diff --git "a/measurement.json" "b/measurement.json" --- "a/measurement.json" +++ "b/measurement.json" @@ -1,77902 +1,50733 @@ { - "measurement": [ - { - "key": "model.layers.0.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.03230390325188637, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.029784446582198143, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.02186223864555359, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.01975933089852333, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.014162356965243816, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.010739832185208797, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.03140970692038536, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.023540202528238297, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.014444740489125252, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.012868860736489296, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.012665276415646076, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.012587270699441433, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.01083245687186718, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.008062059991061687, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.007310627494007349, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.006999222561717033, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.005969742778688669, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.005695451516658068, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.0058198184706270695, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.005528253503143787, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.00546506280079484, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.00560511602088809, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.005096091888844967, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005179483909159899, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.0.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.03175673633813858, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.029609620571136475, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.02038322016596794, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.018657049164175987, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.014062105678021908, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.009873351082205772, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.03141756355762482, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.024532996118068695, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.014313061721622944, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.013049090281128883, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.012956628575921059, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.012832366861402988, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.01152390893548727, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.008138582110404968, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.007084643468260765, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.007484937086701393, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.006042063236236572, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.005641546566039324, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.005928767845034599, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.005509600508958101, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.0057862550020217896, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.00577015895396471, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.005292873829603195, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00524822948500514, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.0.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.1890709102153778, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.17495830357074738, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1698278933763504, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.15344351530075073, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.08814194053411484, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.08184386789798737, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.10108186304569244, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.09211748093366623, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.0894981250166893, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.07742056250572205, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07308852672576904, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05194424092769623, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04424315318465233, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.0421445332467556, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04161277040839195, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02636825293302536, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02144516073167324, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.021204672753810883, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.019248029217123985, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.018886271864175797, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.013634934090077877, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.012807287275791168, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.012339694425463676, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00785725750029087, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.0.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.053076013922691345, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.03815510496497154, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.03296404331922531, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.027283594012260437, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.02372126281261444, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.017015783116221428, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.02817617729306221, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.02603343315422535, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.024947166442871094, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.014882590621709824, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.014337333850562572, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.014466616325080395, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.01276424527168274, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.011754602193832397, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.011499390006065369, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.007423125207424164, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.0066728852689266205, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.0066141001880168915, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.0052263494580984116, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.005061035975813866, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.004322547931224108, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.004935316741466522, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.004045556765049696, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.004069938790053129, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.0.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.13541144132614136, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.12699513137340546, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.12453945726156235, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.11379240453243256, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.06122714281082153, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.058325089514255524, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.06725095212459564, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.06251093745231628, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.061852872371673584, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.055256620049476624, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.0524868443608284, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.03413916379213333, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.029971111565828323, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.029407639056444168, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.02926579676568508, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.01708625815808773, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.01571928709745407, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.015688348561525345, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.01461978629231453, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.014541825279593468, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.009358192794024944, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.010316944681107998, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.009183968417346478, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.00788067001849413, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.0.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.15738537907600403, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.14769551157951355, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.14491017162799835, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.13234739005565643, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.07098409533500671, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.06762091815471649, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.07794409245252609, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.07242552191019058, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.07169885188341141, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.06406544148921967, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.06064144894480705, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.03916986286640167, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.03423396125435829, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.033607348799705505, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.03346450626850128, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.019499391317367554, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.017037635669112206, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.016992496326565742, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.01564527116715908, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.015552791766822338, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.010034192353487015, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.00989194493740797, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.009826322086155415, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.006115985568612814, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.0.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.07462339848279953, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.06711549311876297, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.06340650469064713, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.0574587807059288, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.0325695276260376, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.029676824808120728, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.03982415795326233, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.03661837801337242, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.03372100740671158, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.029274379834532738, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.0280716884881258, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.020092815160751343, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.01765749230980873, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.015863999724388123, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.015424376353621483, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.010135111398994923, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.008830707520246506, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.008696301840245724, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.008257830515503883, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.007993762381374836, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.005695198196917772, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.006403717212378979, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.005172977223992348, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.0050712404772639275, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.1.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.03534870594739914, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.03147692233324051, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.028167836368083954, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.024974951520562172, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.015662850812077522, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.013350101187825203, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.022028453648090363, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.019596051424741745, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.016084345057606697, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.013411153107881546, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.01271353755146265, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.010647699236869812, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.009278953075408936, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.007858090102672577, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.007474545855075121, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.005561539903283119, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.00465261610224843, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.004480893723666668, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.00428995955735445, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.004078152123838663, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.003570699831470847, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.003701635170727968, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.003251073881983757, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0030895634554326534, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.1.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.029926501214504242, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.02669426240026951, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.02312900498509407, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.02055640146136284, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.013153627514839172, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.010999999940395355, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.020159635692834854, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.01746341958642006, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.013508475385606289, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.011330629698932171, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.010803676210343838, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.009398099966347218, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.008143101818859577, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.006708075758069754, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.006308048497885466, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.004970990587025881, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.004108397290110588, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.0039380923844873905, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.0038292964454740286, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.0036250639241188765, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.003321494208648801, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.003399296198040247, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.003039101604372263, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.002904816996306181, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.1.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.16055884957313538, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.14783863723278046, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1437121331691742, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12997612357139587, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.07449045777320862, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06926274299621582, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.08357331901788712, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.07732324302196503, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.07576218247413635, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06522131711244583, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.06148075312376022, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.04249928891658783, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03695463389158249, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03561334311962128, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.035256583243608475, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.021314555779099464, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.018105724826455116, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.017945213243365288, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01620764657855034, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01593770645558834, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.011044352315366268, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.010726889595389366, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.010410876013338566, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00651192432269454, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.1.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.153273805975914, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.12990422546863556, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.12154393643140793, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.1039503738284111, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06994494795799255, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.060216616839170456, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.08316370099782944, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.07621477544307709, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.07242124527692795, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05554450303316116, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05070900171995163, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.042754609137773514, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03724519535899162, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03446095064282417, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03376360610127449, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02180454321205616, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.019220951944589615, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.019001014530658722, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.016655422747135162, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01623133383691311, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01237456128001213, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.013823067769408226, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.011549348011612892, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.011132515035569668, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.1.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.17471274733543396, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.16492117941379547, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.16210776567459106, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.14889872074127197, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.08024512231349945, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.07680158317089081, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.08847419917583466, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.08173129707574844, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.08101088553667068, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.07299139350652695, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.06987179815769196, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.04510660097002983, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.03926655277609825, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.03863569721579552, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.03849169984459877, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.022638149559497833, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.020595412701368332, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02055368945002556, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.01922721229493618, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.01912575028836727, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.012472609989345074, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.013410023413598537, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.012265684083104134, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.010142502374947071, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.1.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2017059177160263, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.19055432081222534, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.18729610741138458, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.17219224572181702, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09255760163068771, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08858363330364227, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10170932114124298, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09422540664672852, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.0934523344039917, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08415020257234573, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.0803622454404831, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05136121064424515, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04471816495060921, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04399426653981209, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.043815162032842636, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.025602787733078003, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.022317245602607727, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.022268952801823616, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.020616039633750916, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.020501814782619476, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013260528445243835, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.012961839325726032, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013014673255383968, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.008059629239141941, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.1.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.14722862839698792, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.13641291856765747, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.1314466893672943, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.11970935761928558, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.06637635827064514, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.062316954135894775, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.07735145837068558, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.07130209356546402, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.06771218031644821, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.060373831540346146, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.05774446204304695, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.03892504423856735, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.03405678644776344, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.03189798444509506, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.03140507638454437, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.019507693126797676, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.01695002056658268, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.016808170825242996, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.01583017222583294, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.015524406917393208, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.01059587299823761, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.011207498610019684, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.00997238326817751, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.008228613063693047, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.2.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.05159660801291466, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.04790686443448067, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.045982055366039276, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.04166054353117943, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.023549938574433327, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.02184896357357502, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.028262706473469734, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.025476427748799324, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.02387407049536705, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.020853713154792786, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.019725650548934937, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.01377807091921568, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.012009107507765293, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.011292139999568462, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.011122181080281734, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.0069027249701321125, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.005896113347262144, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.005824903026223183, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.00536649115383625, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.005268261302262545, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.003723554313182831, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.0037508008535951376, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.0035266492050141096, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.002617023652419448, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.2.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.04054543003439903, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.037604160606861115, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.035977840423583984, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.032591838389635086, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.018444236367940903, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.017070382833480835, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.02258233167231083, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.020109029486775398, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.01869824528694153, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.016330745071172714, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.015454769134521484, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.01092112809419632, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.009482557885348797, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.008873945102095604, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.00873279944062233, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.005500265397131443, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.004717655945569277, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.004652027040719986, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.004314851481467485, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.004228429403156042, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.0030824942514300346, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.0031127629335969687, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.0029167619068175554, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0022786727640777826, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.2.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.19114623963832855, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.17926980555057526, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.17568109929561615, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.16063231229782104, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.08923749625682831, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.08441183716058731, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.09876401722431183, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.0914650559425354, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.09021397680044174, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.07990236580371857, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07586990296840668, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05013111233711243, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04365582391619682, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.042564231902360916, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.042320989072322845, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.025051113218069077, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.021578196436166763, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.021495796740055084, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.019661540165543556, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.019482774659991264, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.012872566469013691, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.012543728575110435, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.012466352432966232, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.007644908502697945, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.2.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.16952520608901978, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1468108892440796, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.13847285509109497, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12025243788957596, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.07776599377393723, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06804652512073517, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.09155970066785812, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.0839441642165184, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.08012600988149643, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06322436779737473, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.057546466588974, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.04726608470082283, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04097380116581917, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03825455531477928, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.0375855378806591, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.0240078866481781, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02117646485567093, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.020961282774806023, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.018532173708081245, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.018120143562555313, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.013683960773050785, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.015025784261524677, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01286364533007145, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.012008982710540295, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.2.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.19296324253082275, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.18210084736347198, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.17897656559944153, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.16427966952323914, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.08994249999523163, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08585315942764282, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.09870002418756485, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09162520617246628, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09081339091062546, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08141402900218964, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07770749926567078, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.050588399171829224, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04420272633433342, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.043440576642751694, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04324336349964142, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.025359714403748512, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.023204082623124123, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.023165026679635048, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02160399965941906, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02149081788957119, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013866991735994816, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015172258950769901, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013635027222335339, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.011541587300598621, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.2.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.22644762694835663, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.2138393521308899, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.21023644506931305, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.19324305653572083, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10531605035066605, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10046801716089249, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11590850353240967, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10722790658473969, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10627469420433044, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09530223160982132, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.0908636748790741, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.0587252713739872, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05106794461607933, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.05016016960144043, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.049943044781684875, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.029278580099344254, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02550732158124447, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025452284142374992, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023467864841222763, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.023348025977611542, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.01521490328013897, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014883077703416348, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014920786954462528, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.00935265701264143, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.2.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.05903501808643341, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.05505094304680824, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.04590924456715584, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.041437339037656784, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.025198379531502724, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.022466378286480904, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.04858943074941635, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.03760317340493202, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.02580188401043415, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.022662848234176636, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.021619411185383797, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.016800355166196823, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.014778319746255875, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.01373329944908619, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.012082899920642376, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.010434512048959732, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.007401198614388704, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.006366416811943054, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.006973918993026018, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.005791026167571545, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.005816615652292967, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.005798433907330036, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.003990013152360916, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.0034828432835638523, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.3.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.07996770739555359, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.07479416579008102, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.07175824791193008, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.0653490275144577, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.03694504126906395, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.03423794358968735, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.04483042284846306, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.04031820222735405, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.03737746924161911, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.03309395909309387, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.031643420457839966, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.022599119693040848, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.01930401101708412, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.017827562987804413, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.017467498779296875, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.011455455794930458, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.009465470910072327, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.009284178726375103, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.00872432067990303, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.008498492650687695, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.006426608189940453, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.006281745154410601, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.005942298099398613, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.004535260610282421, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.3.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.06444535404443741, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.060113415122032166, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.058461762964725494, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.05323544144630432, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.029696233570575714, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.027893560007214546, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.03378530964255333, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.031040962785482407, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.030057117342948914, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.02648768573999405, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.02519277110695839, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.017052199691534042, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.014772262424230576, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.014184234663844109, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.014046370051801205, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.008524065837264061, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.007293579168617725, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.00723435590043664, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.006649124436080456, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.0065618595108389854, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.004470666870474815, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.004438013769686222, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.004279212094843388, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0029119050595909357, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.3.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.21933063864707947, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.2062680572271347, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.20222267508506775, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.1850869059562683, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10258903354406357, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.09732295572757721, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11337923258543015, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10499154031276703, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10370680689811707, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.09209772944450378, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.08739916980266571, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05754419043660164, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05009245127439499, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.04895636811852455, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.048684172332286835, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.028727641329169273, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.024762911722064018, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.024674568325281143, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.022577766329050064, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.022405797615647316, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.014722024090588093, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014310531318187714, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.014300607144832611, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.008646124973893166, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.3.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.17312461137771606, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.15664464235305786, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.15127362310886383, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.13386797904968262, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.08121980726718903, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0743836984038353, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.09183249622583389, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.08451594412326813, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.0825193002820015, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06850700080394745, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.06351248919963837, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.047050658613443375, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.0408363901078701, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03937285766005516, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03904258459806442, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.023645054548978806, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.021023018285632133, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.020903464406728745, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01859360747039318, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01836460456252098, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.013012618757784367, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.013828192837536335, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.012560536153614521, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010394853539764881, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.3.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.20892392098903656, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.196786031126976, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.1931859403848648, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.17705611884593964, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09778991341590881, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09309025853872299, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10774315148591995, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09979282319545746, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09880547970533371, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08811349421739578, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08380074054002762, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.0549539215862751, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04781980440020561, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.046941958367824554, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.0467304028570652, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.027511587366461754, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02446419931948185, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.024421628564596176, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.022554462775588036, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.022430982440710068, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.014722931198775768, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015218201093375683, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014436842873692513, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.010746627114713192, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.3.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.24282535910606384, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.22884128987789154, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.2247030884027481, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.20599092543125153, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.11341246217489243, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10803091526031494, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.1249537467956543, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.11567720770835876, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.11458311975002289, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.1021554172039032, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09715490788221359, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.06331618875265121, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05510124936699867, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.05408409982919693, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05385606735944748, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.03157944604754448, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.027450604364275932, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.027382805943489075, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.025131892412900925, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.024979012086987495, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.01630723476409912, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015935543924570084, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.015969818457961082, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.00985249038785696, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.3.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.23012270033359528, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.21413758397102356, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.20868630707263947, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.18892550468444824, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10567955672740936, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10010260343551636, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.11970848590135574, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.11010008305311203, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.10766498744487762, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09537526220083237, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.09049008786678314, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06063028797507286, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05259585753083229, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05061623454093933, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.050144534558057785, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.030354073271155357, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.026314299553632736, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.026248959824442863, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02428712509572506, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.023989904671907425, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.01627430133521557, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.0164086502045393, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.015653884038329124, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.011325614526867867, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.4.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.102790966629982, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.09626410156488419, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.09279105067253113, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.08470243960618973, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.04790723696351051, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.04462721571326256, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.0576588474214077, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05159999802708626, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.04843294620513916, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04290582984685898, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04113638401031494, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.029203850775957108, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.0246825460344553, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02310531586408615, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.022734547033905983, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.014748561196029186, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.012194212526082993, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.012010610662400723, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.011218578554689884, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.010987729765474796, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008212448097765446, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.007935082539916039, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.0076575689017772675, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00566742941737175, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.4.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.0843391939997673, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.07882413268089294, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.07688678801059723, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.07006409764289856, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.03919734060764313, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0369318388402462, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.04421596974134445, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.04065309092402458, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.03969534859061241, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.034987032413482666, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.03321484476327896, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.022354386746883392, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.01938474178314209, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.018737878650426865, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.018584592267870903, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.01116924174129963, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.009565289132297039, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.00950571522116661, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.008707081899046898, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.008608478121459484, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.005798628553748131, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.005702632945030928, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.005583565682172775, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0036335140466690063, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.4.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.22520288825035095, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.21134836971759796, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.20708592236042023, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.18927094340324402, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10555751621723175, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.10000020265579224, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.1168457567691803, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10812275856733322, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10675031691789627, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.09450351446866989, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.08956869691610336, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05927681550383568, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05157037824392319, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05040544643998146, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05011654645204544, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.029564814642071724, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.0255179051309824, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02543386071920395, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.023236563429236412, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.023043200373649597, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.015178518369793892, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014790074899792671, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.014765020459890366, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00898163951933384, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.4.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.1730976700782776, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.15326833724975586, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.14701363444328308, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12800705432891846, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.08111076802015305, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.07299018651247025, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.09227549284696579, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.08499400317668915, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.08281959593296051, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06637167185544968, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.06126119941473007, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.04747084900736809, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.041436269879341125, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03970557451248169, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03928523138165474, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.024047935381531715, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.021669510751962662, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.021536841988563538, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.018932253122329712, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01867564022541046, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.013564685359597206, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014868834055960178, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01307661086320877, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.011697536334395409, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.4.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.21639786660671234, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.20351114869117737, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.1996537148952484, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.18275615572929382, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10169310867786407, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09661044925451279, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11214753240346909, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10383006930351257, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10274603217840195, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09132052958011627, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08680761605501175, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05718811973929405, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.049756474792957306, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.048805203288793564, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04858317971229553, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.028587747365236282, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.025373900309205055, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025314094498753548, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02331109344959259, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02317136898636818, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015214326791465282, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015679920092225075, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014907696284353733, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.010949176736176014, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.4.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.25389978289604187, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.23893406987190247, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.23444585502147675, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.2145870476961136, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.11913088709115982, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.11327023059129715, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.13137631118297577, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.121636763215065, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.12039158493280411, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.1069963350892067, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.10167234390974045, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.06667856872081757, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.0580294169485569, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.05692297965288162, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.056663017719984055, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.0332958810031414, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.028950318694114685, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.028874145820736885, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.026441948488354683, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.026269778609275818, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.0172380693256855, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.016882549971342087, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.016872059553861618, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.010539455339312553, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.4.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.23657993972301483, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.2193647176027298, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.21338404715061188, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.1928098052740097, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10903364419937134, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10284297168254852, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.12362591177225113, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.114096499979496, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.1113564521074295, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09796147048473358, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.09260815382003784, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06271668523550034, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.054513897746801376, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.052235234528779984, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05170150473713875, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03136032074689865, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.02706066705286503, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.026993483304977417, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02486865594983101, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.024519724771380424, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.016597427427768707, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.016779720783233643, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.01584666222333908, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.011384401470422745, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.5.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.10764948278665543, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.10075895488262177, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.09700749069452286, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.08839825540781021, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.05028269439935684, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.046738553792238235, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06055569648742676, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.054311539977788925, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.050868310034275055, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04494006186723709, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.042992521077394485, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03065258078277111, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.02602774277329445, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02430684305727482, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.023859865963459015, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.01556455995887518, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.01281799003481865, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.012637916021049023, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.011769379489123821, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.011544013395905495, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008654851466417313, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008372760377824306, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.008095315657556057, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.006036918610334396, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.5.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.08646245300769806, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.08076246827840805, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.07867812365293503, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.07168218493461609, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.04032345861196518, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0379098579287529, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.04548811912536621, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.041887860745191574, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.04079668968915939, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.03589119762182236, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.034008901566267014, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.0230115856975317, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.019979111850261688, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.019281046465039253, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.019115734845399857, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.011506857350468636, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.009871723130345345, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.00980654638260603, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.00897077564150095, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.008863293565809727, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.006001310423016548, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.005924876779317856, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.0057685114443302155, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0038160874973982573, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.5.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.23853586614131927, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.22382989525794983, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.21932072937488556, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.20030316710472107, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.11212362349033356, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.1061638668179512, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.1241171658039093, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.11480993777513504, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.11335526406764984, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10011237114667892, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.0948643758893013, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06297926604747772, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.054810069501399994, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.053536687046289444, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05323807895183563, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.031426481902599335, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.027106203138828278, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02701152116060257, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.02461063675582409, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.024423733353614807, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01614571362733841, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.015672702342271805, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.015698641538619995, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009481065906584263, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.5.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.1727810949087143, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.15379182994365692, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1476573348045349, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.1291482001543045, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.08100153505802155, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0729072168469429, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.0918414518237114, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.08466767519712448, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.08273676037788391, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06699194014072418, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.060879193246364594, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.04734807088971138, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04115236550569534, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.039507076144218445, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03910219669342041, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02403370663523674, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.021382519975304604, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02126971073448658, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.018765917047858238, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.018506981432437897, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.013564104214310646, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014456185512244701, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.013066314160823822, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.011192136444151402, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.5.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.20689240097999573, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.19441372156143188, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.19055606424808502, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.174322709441185, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09769954532384872, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09269075840711594, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10793227702379227, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10005397349596024, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09872745722532272, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08764199912548065, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08322129398584366, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05530182272195816, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.048302583396434784, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.047286685556173325, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04703835770487785, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.027715908363461494, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02522403374314308, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02516610361635685, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023314541205763817, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.023155920207500458, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015125722624361515, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.016477664932608604, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014796900562942028, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.012473470531404018, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.5.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.25308647751808167, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.23785808682441711, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.23336154222488403, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.21352022886276245, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.11904557794332504, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.11299676448106766, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.13179470598697662, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.12183050811290741, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.1203702986240387, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.1067824736237526, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.10129484534263611, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.06697940826416016, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05814240872859955, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.05692308396100998, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.056635692715644836, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.03342907503247261, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.029099583625793457, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.029002252966165543, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.026576552540063858, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.026390839368104935, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.01744798757135868, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.017215419560670853, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.017036037519574165, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.011060243472456932, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.5.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.046718087047338486, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.042680781334638596, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.040529925376176834, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.03635390102863312, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.021945465356111526, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.019901499152183533, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.025560520589351654, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.02449067495763302, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.022531796246767044, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.01940131187438965, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.01827116869390011, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.014430774375796318, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.011907788924872875, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.010581516660749912, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.010272757150232792, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.0069638993591070175, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.005534140393137932, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.0054743289947509766, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.005025467835366726, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.0048856050707399845, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.0034595595207065344, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.003634590422734618, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.0032121706753969193, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.002568020485341549, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.6.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.11902957409620285, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11133553087711334, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1079295352101326, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09839572757482529, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.05573156476020813, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05214904993772507, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06565845012664795, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05906219780445099, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.0564018152654171, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.049783505499362946, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04774110019207001, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03358117863535881, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.028334232047200203, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.026876728981733322, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.026536831632256508, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.016934484243392944, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.014141716063022614, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.013993840664625168, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.012975621037185192, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.012761090882122517, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.009303630329668522, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009104039520025253, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.008744360879063606, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.006495154928416014, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.6.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.09949972480535507, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.09291025996208191, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.09068355709314346, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.08261038362979889, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.04643649235367775, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0437525250017643, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.05209078639745712, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.04792306572198868, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.047021910548210144, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.041314516216516495, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.03920622915029526, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.026381203904747963, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.022857796400785446, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.022196870297193527, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.022038869559764862, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.013172514736652374, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.011355948634445667, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.011303628794848919, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.0103123988956213, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.010210176929831505, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.006844031158834696, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.006782940588891506, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.006613933946937323, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.004371808376163244, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.6.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.23898932337760925, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.22417275607585907, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.2196040004491806, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.20059706270694733, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.11238209158182144, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.10636750608682632, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.12438414990901947, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.11500655859708786, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.11365558207035065, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10031379759311676, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.094990573823452, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06304016709327698, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05486680939793587, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.0536516010761261, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05338259041309357, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.031455300748348236, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.027195753529667854, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02710846997797489, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.024686377495527267, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02449621632695198, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.016154326498508453, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.01574154570698738, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.015738073736429214, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009567425586283207, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.6.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.16943922638893127, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.144352987408638, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1342983841896057, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.11556942015886307, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.07748635113239288, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06683982163667679, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.09393837302923203, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.08677650988101959, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.08162500709295273, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06269842386245728, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.058101631700992584, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.049092765897512436, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.042357176542282104, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03803851827979088, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03697344288229942, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.024759789928793907, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02087915875017643, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.020550837740302086, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01816711388528347, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01750108040869236, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.013695296831429005, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014848684892058372, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.012263313867151737, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.011480429209768772, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.6.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.19779500365257263, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.18587766587734222, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.18217071890830994, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.16667413711547852, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09329400211572647, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.088478684425354, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10315010696649551, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09554152190685272, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.0943310409784317, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08358969539403915, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07929789274930954, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05263431370258331, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04579303041100502, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04483896121382713, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04461977630853653, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.02629619836807251, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.023338980972766876, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.023279234766960144, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02140720747411251, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.021266452968120575, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013979896903038025, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014473650604486465, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.0136740542948246, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.010150056332349777, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.6.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.252116322517395, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.2370356172323227, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.23254947364330292, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.21291573345661163, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.11895839869976044, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.1129140704870224, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.1314505785703659, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.12159153819084167, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.12025953084230423, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.10663434863090515, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.10118444263935089, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.06691265106201172, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05817706137895584, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.05701812356710434, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05674217268824577, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.03342930227518082, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.029331443831324577, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02924802154302597, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.026823388412594795, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02665148861706257, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.017560560256242752, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.017664512619376183, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.017171498388051987, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.011758037842810154, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.6.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.11825074255466461, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.10829847306013107, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.1031406819820404, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.09275054931640625, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.05468102917075157, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.05014907568693161, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.0662902295589447, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.0605231337249279, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.056117910891771317, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.04866526275873184, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.04561511427164078, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.03320688754320145, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.029193706810474396, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.026312286034226418, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.02583557553589344, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.016442634165287018, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.014126872643828392, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.014046375639736652, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.012971324846148491, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.012720786035060883, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.009232714772224426, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.009539385326206684, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.008736243471503258, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.007210031151771545, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.7.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.10980059951543808, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.10261372476816177, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10026261955499649, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09149814397096634, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.051397860050201416, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.04848257824778557, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.058953914791345596, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05293509364128113, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.05203288793563843, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.045825712382793427, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.044375624507665634, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03000146895647049, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.025320887565612793, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.024625923484563828, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.024455511942505836, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.014988242648541927, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.012568842619657516, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.012509211897850037, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01143430732190609, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.011323952116072178, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.007768369745463133, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.0074547152034938335, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.007540724240243435, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0047446745447814465, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.7.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.08938213437795639, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.08343709260225296, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.08144435286521912, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.07417131215333939, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.04169930890202522, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.03925752267241478, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.04672536253929138, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.04305475205183029, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.04219828173518181, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.037080153822898865, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.03515513613820076, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.02369081601500511, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.0205459576100111, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.01993577368557453, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.019788602367043495, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.011818175204098225, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.010190783999860287, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.010139571502804756, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.009245382621884346, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.009152770042419434, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.006129931658506393, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.006077486090362072, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.005924491677433252, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0038990583270788193, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.7.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.22754721343517303, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.21352410316467285, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.20915710926055908, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.19090013206005096, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10704988986253738, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.1012234017252922, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.1184055432677269, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10955510288476944, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10821380466222763, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.09544561803340912, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.09031977504491806, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.060099177062511444, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05227132886648178, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05108335241675377, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05082849785685539, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.029967959970235825, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.025881899520754814, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.025798533111810684, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.023487873375415802, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02330402471125126, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.015388625673949718, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014989781193435192, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.014972870238125324, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009126048535108566, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.7.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.19697892665863037, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.17646698653697968, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1697211116552353, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.14759518206119537, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.09259555488824844, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.08389399200677872, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.10543984174728394, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.09697792679071426, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.09456494450569153, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.0768948644399643, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07087763398885727, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05451114475727081, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04718714952468872, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.045245878398418427, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.044797636568546295, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.027478348463773727, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.024360958486795425, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.024223554879426956, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.021358199417591095, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.021055832505226135, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.015231011435389519, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016316916793584824, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.014637854881584644, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.012475160881876945, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.7.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.18349404633045197, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.1724083572626114, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.16908800601959229, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.15476863086223602, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.08652464300394058, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08204633742570877, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.09532579034566879, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.08839081972837448, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.08746970444917679, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.07745729386806488, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07343528419733047, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.048589203506708145, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.042325641959905624, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04149353876709938, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04130106791853905, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.024278581142425537, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.021448850631713867, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.021392080932855606, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.01961928978562355, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.019493186846375465, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.012773008085787296, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.013055024668574333, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01249791495501995, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.008886901661753654, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.7.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.24589522182941437, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.2311146855354309, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.2268570214509964, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.2077142298221588, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.1158728301525116, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10992388427257538, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.12758216261863708, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.11824727803468704, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.11709035933017731, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.10378725081682205, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09839466959238052, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.06480187922716141, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05644199252128601, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.055357757955789566, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.0550931952893734, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.032338134944438934, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02810518816113472, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.028037091717123985, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.025606980547308922, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.025436287745833397, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.01666918955743313, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.01632506772875786, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01630248688161373, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.010088598355650902, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.7.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.23471567034721375, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.21402274072170258, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.2057631015777588, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.18420615792274475, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10829458385705948, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10006935894489288, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.12739619612693787, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.11630374193191528, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11130864173173904, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09552124887704849, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.09004934132099152, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06450801342725754, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05585946515202522, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.052243996411561966, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05134078860282898, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03242268040776253, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.02751445211470127, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.027330758050084114, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.025027453899383545, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.024469437077641487, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.017626669257879257, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.017890987917780876, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.016468890011310577, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.012713396921753883, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.8.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.11127451062202454, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.10420023649930954, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10173998773097992, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.0927407518029213, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.052112530916929245, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.049166515469551086, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.05901443213224411, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05360914021730423, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.05271561071276665, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.046424370259046555, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.044611383229494095, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.030027132481336594, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.025628501549363136, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02495470829308033, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.024791844189167023, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.015033775940537453, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.012737213633954525, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.012688725255429745, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01158075500279665, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.011480345390737057, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.007832168601453304, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.00754639832302928, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.007609360385686159, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.004801752977073193, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.8.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.09519032388925552, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.08894002437591553, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.0869046002626419, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.07922432571649551, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.044487979263067245, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.04189468175172806, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.04963456094264984, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.04582412168383598, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.04500702768564224, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.03956631198525429, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.0374998077750206, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.025136549025774002, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.021858420222997665, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02124934084713459, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.0211094468832016, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.012550905346870422, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.010837976820766926, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.010783610865473747, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.009830283001065254, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.00973979476839304, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.006488240323960781, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.006409359164535999, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.006284231785684824, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.004052178468555212, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.8.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.22532419860363007, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.21128776669502258, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.20705345273017883, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.18895423412322998, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10589644312858582, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.10025534778833389, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11717914044857025, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10842379927635193, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10708639770746231, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.09451660513877869, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.0894625335931778, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05943150818347931, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05168234556913376, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.050561223179101944, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.050284747034311295, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.029650727286934853, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.025612886995077133, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02553373947739601, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.023242315277457237, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.023065786808729172, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.015220511704683304, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014835992828011513, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.014818433672189713, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009023422375321388, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.8.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.20711010694503784, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1871723085641861, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.18063917756080627, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.15902385115623474, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.09766966849565506, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.08896040171384811, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11031010001897812, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10185875743627548, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.09941524267196655, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.08210758864879608, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07525745779275894, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05673789232969284, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04923639073967934, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.04729748144745827, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.046846430748701096, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.028473015874624252, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02500917948782444, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.024860821664333344, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.021958444267511368, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02165548875927925, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01537347212433815, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016181927174329758, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01477944664657116, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.011805381625890732, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.8.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.18724161386489868, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.1759393960237503, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.17257411777973175, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.15788619220256805, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.08840065449476242, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08384176343679428, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.09737501293420792, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09027653187513351, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.08934623748064041, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.07911240309476852, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07503043115139008, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.04965728893876076, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04326877370476723, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04241293668746948, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.042222097516059875, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.024817511439323425, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02198009006679058, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02192896604537964, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02012573555111885, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.01999683491885662, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013085033744573593, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.013475515879690647, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.012806599028408527, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009287471882998943, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.8.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2413172572851181, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.22683729231357574, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.22258034348487854, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.20374387502670288, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.11369740217924118, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10793096572160721, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.1252441704273224, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.11601077020168304, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.11495629698038101, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.10185002535581589, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09654070436954498, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.06355135887861252, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05535734444856644, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.05430551990866661, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05405336618423462, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.031712599098682404, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.027539100497961044, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02747100591659546, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02506878413259983, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.024907978251576424, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.016309000551700592, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.01595393754541874, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.015954315662384033, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.00977935828268528, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.8.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.2452506422996521, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.22227351367473602, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.21359038352966309, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.19082894921302795, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.11300785094499588, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10418257862329483, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.13143494725227356, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.12102453410625458, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11644663661718369, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09902934730052948, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.09309613704681396, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06692174077033997, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05797097086906433, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05435730144381523, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05346840247511864, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03355568274855614, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.02831246890127659, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.028185700997710228, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.025538384914398193, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.0249851755797863, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.017923589795827866, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.01796211116015911, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.0167447030544281, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.012308033183217049, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.9.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.10255251079797745, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.09583067148923874, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.09358014166355133, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.0853947326540947, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.04789820313453674, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.045137446373701096, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.0547490119934082, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.04938406124711037, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.04848824813961983, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04265343025326729, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04098989814519882, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.0277263056486845, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.023632628843188286, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.022941917181015015, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.02278757095336914, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.013868329115211964, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.011721193790435791, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.011664972640573978, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.010649792850017548, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.010546267032623291, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.007237666752189398, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.006957689765840769, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.007006957661360502, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00442010210826993, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.9.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.08511049300432205, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.07951992005109787, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.07755565643310547, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.07065784931182861, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.03976742923259735, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.03740956634283066, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.04452481493353844, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.041113343089818954, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.040220845490694046, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.035365279763936996, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.03350544720888138, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.02255777269601822, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.01961967721581459, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.018998751416802406, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.018854942172765732, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.011267002671957016, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.009696636348962784, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.009644022211432457, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.008794927038252354, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.00869704782962799, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.005836266092956066, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.005762606859207153, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.005629430525004864, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.003654486034065485, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.9.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.2256675809621811, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.21147257089614868, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.2072169929742813, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.18912722170352936, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10605580359697342, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.10029357671737671, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11758294701576233, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10855937749147415, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10728064924478531, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.0945449247956276, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.08951325714588165, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.059706419706344604, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05182470381259918, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05064740777015686, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.050387851893901825, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.029774507507681847, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02565918304026127, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.025567246600985527, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.023262344300746918, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.023082736879587173, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.015280312858521938, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014842955395579338, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.014853636734187603, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009002137929201126, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.9.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.19787156581878662, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.18060900270938873, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.17392319440841675, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.15529467165470123, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.09259910136461258, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.08473212271928787, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.10646578669548035, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.09806748479604721, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.09445393830537796, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.07976184040307999, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07423610240221024, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05466979742050171, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04721061885356903, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.04469382017850876, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04407104104757309, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.027381636202335358, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.023526998236775398, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02329311892390251, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.021001942455768585, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02061641402542591, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.014698012731969357, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.0151962386444211, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.013871350325644016, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.01086970791220665, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.9.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.17915546894073486, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.16845478117465973, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.16531679034233093, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.15136654675006866, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.0844869613647461, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08019711822271347, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.09306925535202026, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.08622147887945175, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.08537937700748444, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.07577173411846161, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07188756763935089, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.047440070658922195, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04136456176638603, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04056728258728981, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.040381379425525665, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.023714013397693634, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.021068818867206573, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.021019572392106056, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.019331051036715508, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.019215261563658714, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.012559070251882076, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.012983555905520916, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01229412667453289, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009026437997817993, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.9.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2258313000202179, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.2125554233789444, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.20860521495342255, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.19107140600681305, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10623262077569962, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10092112421989441, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11691384762525558, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.1083826795220375, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10736952722072601, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09530474245548248, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09041117131710052, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.0593692846596241, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.051714155822992325, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.050730668008327484, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05050645023584366, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.029625535011291504, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02575121633708477, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025681868195533752, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023482689633965492, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.023325210437178612, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015259861014783382, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014941263943910599, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014934358187019825, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009203712455928326, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.9.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.23376570641994476, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.2111617624759674, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.20266038179397583, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.18061649799346924, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10784362256526947, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.0990186482667923, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.12570765614509583, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.11558122932910919, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11119142919778824, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.0940009132027626, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.08812443912029266, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06385847926139832, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05539238452911377, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05188455805182457, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05102358013391495, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03200243040919304, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.027041567489504814, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.026910850778222084, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02429785206913948, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.023745376616716385, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.017027612775564194, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.017182936891913414, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.01588599383831024, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.011773924343287945, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.10.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.11447321623563766, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.10654667764902115, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10424862802028656, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09464993327856064, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.053199153393507004, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05009791627526283, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.05992332100868225, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05463682860136032, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.05383400246500969, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04719281941652298, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04498329013586044, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.030447708442807198, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.026121558621525764, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02548230066895485, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.025329524651169777, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.015247485600411892, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.013115053065121174, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.013064676895737648, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01191038265824318, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.011811500415205956, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.007993104867637157, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.007921653799712658, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.00778066786006093, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005264063831418753, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.10.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.1045013964176178, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.09734932333230972, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.09507685899734497, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.08629470318555832, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.04851212352514267, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.04561317712068558, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.05394681170582771, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.049876801669597626, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.049111753702163696, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.042950861155986786, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04053771123290062, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.027310486882925034, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.023788943886756897, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.0231768935918808, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.02303164452314377, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.013632379472255707, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.011858777143061161, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.011816524900496006, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.010736781172454357, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.010641636326909065, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.007084658369421959, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.007090516854077578, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.006874420680105686, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.004590931348502636, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.10.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.21807080507278442, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.204308420419693, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.20030155777931213, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.1826866865158081, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.1023874282836914, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.09689155966043472, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11311718821525574, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10469871759414673, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10353468358516693, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.09131264686584473, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.0863809660077095, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05741292983293533, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.049961209297180176, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.0489317961037159, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04868265241384506, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02866416610777378, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.024830279871821404, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.024754798039793968, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.022521832957863808, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.022364120930433273, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.014752731658518314, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014435838907957077, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.014394083991646767, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.008909760974347591, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.10.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.18520213663578033, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.16719025373458862, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.16080552339553833, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.144624263048172, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.087563656270504, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.07946193218231201, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.0999998152256012, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.09189750999212265, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.08934479206800461, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.07462461292743683, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07009995728731155, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05166500061750412, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04455600678920746, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.0425427071750164, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04207582771778107, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.025947941467165947, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02259005792438984, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.022426849231123924, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.020043455064296722, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.019735246896743774, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.014051233418285847, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014764893800020218, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.013428163714706898, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010863867588341236, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.10.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.18445605039596558, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.1734076738357544, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.17015071213245392, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.15576130151748657, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.08712012320756912, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08265821635723114, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.0959714874625206, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.08891453593969345, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.08803035318851471, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.07801459729671478, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07401019334793091, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.04897664114832878, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04268266260623932, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.041861530393362045, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04166736826300621, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.024487033486366272, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02178344689309597, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.021735206246376038, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.019970642402768135, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.01986006647348404, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.012984409928321838, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.013491390272974968, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.012714183889329433, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009448288008570671, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.10.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.23320625722408295, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.21939817070960999, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.21535548567771912, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.19726768136024475, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10999338328838348, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10441097617149353, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.12104501575231552, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.11220429837703705, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.11119258403778076, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09856319427490234, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09346537292003632, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.06150445714592934, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05356947332620621, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.0525665283203125, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05232658237218857, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.030678218230605125, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.026660606265068054, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02659866213798523, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.024282455444335938, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.0241287499666214, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015772050246596336, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015456386841833591, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.015431784093379974, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009511519223451614, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.10.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.24371828138828278, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.2183934599161148, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.20824569463729858, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.18525753915309906, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.1119791641831398, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10179458558559418, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.13200433552265167, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.12196436524391174, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11603040993213654, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09718292206525803, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.09111372381448746, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06726355105638504, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.058456651866436005, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.053920526057481766, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05282729119062424, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.033641643822193146, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.02814345620572567, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.027949972078204155, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.025221850723028183, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.02451224811375141, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.017775118350982666, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.01809748262166977, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.016306329518556595, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.01237876620143652, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.11.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.1208772137761116, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11285299062728882, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.11033549159765244, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10049282759428024, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.05643557012081146, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05315450206398964, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06375551223754883, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05814741551876068, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.05714234709739685, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05017025023698807, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04790886491537094, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.032383885234594345, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.027831794694066048, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02710324339568615, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.026917103677988052, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.016229825094342232, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.014029612764716148, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01397501491010189, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.012778408825397491, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01267279777675867, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008578612469136715, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.00862344540655613, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.008331483229994774, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005888960789889097, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.11.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.104972705245018, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.09796667844057083, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.09571317583322525, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.08710071444511414, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.04903828725218773, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.04615441709756851, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.05465099215507507, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05054645612835884, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.049637772142887115, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04355058819055557, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04114336520433426, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.027688950300216675, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.02412325330078602, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02344541996717453, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.02328304760158062, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.013828016817569733, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.011993316002190113, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.011935646645724773, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.010869632475078106, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01076793484389782, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.007169216405600309, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.007147985976189375, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.006937566678971052, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.004601681139320135, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.11.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.22378699481487274, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.20943894982337952, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.20520244538784027, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.18704937398433685, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10507450997829437, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0993257686495781, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11631979793310165, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10750360786914825, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.1062939390540123, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.09355980902910233, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.08844102919101715, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05901119485497475, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.051285870373249054, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05018550157546997, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04991619661450386, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02944137714803219, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02544882893562317, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.025370197370648384, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.02305302582681179, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.022876126691699028, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01511243637651205, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014757534489035606, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01469728909432888, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009040121920406818, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.11.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.21076616644859314, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.19141337275505066, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.18237602710723877, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.16256392002105713, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.09907195717096329, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.08938445895910263, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.1165655106306076, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10790187120437622, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10109951347112656, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.08459615707397461, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07931719720363617, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06028179079294205, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05185094475746155, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.047847867012023926, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04685700312256813, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.030173931270837784, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02506316639482975, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.024620572105050087, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.02219153195619583, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.021553169935941696, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.015933139249682426, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016219936311244965, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01456424966454506, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.01124954130500555, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.11.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.19042451679706573, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.1788182109594345, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.17537586390972137, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.160474494099617, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.08993920683860779, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08529110252857208, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.0991208404302597, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09187068045139313, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09094834327697754, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08046485483646393, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07632087916135788, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05057762190699577, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04408321902155876, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04321695491671562, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04300948604941368, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.025286590680480003, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02243734523653984, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.022384529933333397, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.020541001111268997, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.020412243902683258, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.0133889801800251, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.013832527212798595, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013099666684865952, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009609110653400421, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.11.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.230258971452713, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.21627479791641235, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.21220466494560242, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.19413580000400543, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10859785974025726, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10296186804771423, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11958206444978714, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.11081729829311371, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10981331765651703, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09711026400327682, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09201638400554657, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.060738857835531235, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05290063098073006, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.051874566823244095, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05163399502635002, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.030297227203845978, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.026299988850951195, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02623092383146286, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023914575576782227, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.023755494505167007, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.01557457260787487, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015224959701299667, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.015223158523440361, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009323072619736195, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.11.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.24779345095157623, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.22140738368034363, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.21122603118419647, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.18713203072547913, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.11397875100374222, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.1034514531493187, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.134059876203537, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.12365405261516571, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11805945634841919, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09824125468730927, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.09190379828214645, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06831863522529602, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05923108756542206, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05484781041741371, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05376899242401123, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03419063240289688, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.02855277806520462, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.028387024998664856, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.025411702692508698, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.024722568690776825, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.018160689622163773, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.01820395141839981, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.016757359728217125, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.012329966761171818, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.12.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.12400861084461212, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11547456681728363, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.11283902078866959, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10265147686004639, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.05797242745757103, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.054593272507190704, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06547592580318451, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05963774397969246, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.05867113918066025, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05144285783171654, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.049171049147844315, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.033270467072725296, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.028603632003068924, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02784103713929653, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.027652891352772713, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.016665007919073105, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.014407835900783539, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01434832438826561, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.013116259127855301, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.013001681305468082, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008769025094807148, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.00884412880986929, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.008509607054293156, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0060434602200984955, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.12.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.10232221335172653, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.09535089880228043, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.09303556382656097, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.0846225693821907, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.04780754819512367, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.04497074335813522, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.053411323577165604, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.04937935993075371, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.04841474071145058, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04240546002984047, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.040075838565826416, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.02707372047007084, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.023602530360221863, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02288617193698883, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.02272447943687439, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.013538450002670288, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.011744756251573563, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.011684497818350792, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01064563449472189, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.010530530475080013, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.0070417108945548534, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.0070700496435165405, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.006805929355323315, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00461866520345211, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.12.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.2282271534204483, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.21341460943222046, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.2089989334344864, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.1903342753648758, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10715588182210922, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.1012081429362297, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11882764846086502, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10975789278745651, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10843934118747711, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.095297671854496, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.09008511900901794, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06032584607601166, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05236325412988663, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.0511995367705822, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.050929512828588486, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.030097855255007744, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02597164921462536, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.025891749188303947, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.023499496281147003, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.023318355903029442, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.0154655110090971, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.015090477652847767, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.015035077929496765, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00924698170274496, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.12.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.22942133247852325, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.20638862252235413, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.19765879213809967, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.1778465360403061, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10738248378038406, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.09677870571613312, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.12437593936920166, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.11448515951633453, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10997427999973297, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.0912608876824379, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.08600163459777832, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06380156427621841, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05508935824036598, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.051802314817905426, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05102052167057991, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03191511705517769, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.027082782238721848, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02679493837058544, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.02386816032230854, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.023366259410977364, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.016931649297475815, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.017290903255343437, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.015877190977334976, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.012040963396430016, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.12.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.19373808801174164, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.18178531527519226, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.17825660109519958, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.16307376325130463, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09151189774274826, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.0867079645395279, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10091669112443924, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.0935027226805687, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09254271537065506, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08181454986333847, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07752063870429993, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.051480285823345184, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04489652439951897, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04399123042821884, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.043778181076049805, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.025738738477230072, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.022874008864164352, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02281402051448822, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.020925376564264297, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02079707384109497, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013629967346787453, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014130166731774807, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01333596184849739, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009865375235676765, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.12.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.22673049569129944, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.2129223346710205, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.20880825817584991, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.19100216031074524, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.1068899929523468, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10130737721920013, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11781074851751328, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10915762931108475, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10808553546667099, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.0955314189195633, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09049507975578308, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05981085076928139, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05210070684552193, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.05106990411877632, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.050838910043239594, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.029843628406524658, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02592039294540882, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025852108374238014, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023554548621177673, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.023392630741000175, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015352855436503887, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015027865767478943, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014998547732830048, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009234602563083172, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.12.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.25185802578926086, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.2245400846004486, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.21421851217746735, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.18920810520648956, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.11602573841810226, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10524456202983856, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.1360069066286087, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.12533694505691528, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.12012636661529541, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.0994279608130455, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.0927928164601326, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06926175206899643, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.060070548206567764, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.055860377848148346, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05483739450573921, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03472914174199104, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.029114514589309692, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.028965072706341743, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.025819886475801468, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.025145096704363823, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.018478432670235634, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.018568724393844604, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.01710272580385208, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.012658447027206421, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.13.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.13233718276023865, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1234910786151886, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.12066864967346191, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10999077558517456, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06204961985349655, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05847260728478432, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07012740522623062, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06375081837177277, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06278390437364578, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05511821061372757, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.052833281457424164, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.035643741488456726, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03052334487438202, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.029735105112195015, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.029560986906290054, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.017837751656770706, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.015287641435861588, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.015230199322104454, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.013890433125197887, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.013776184991002083, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.009335034526884556, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009209074079990387, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.009081240743398666, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.006090555805712938, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.13.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.10990343987941742, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1025506779551506, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10013782232999802, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09107580035924911, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.051392801105976105, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.04833626747131348, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.057349447160959244, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05302305892109871, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.052009258419275284, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04561882093548775, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04307807981967926, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.029083188623189926, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.025338532403111458, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.024585004895925522, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.024410659447312355, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.014538979157805443, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.01258128508925438, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01252263318747282, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01140893716365099, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.011295043863356113, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.00753732630982995, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.0075326766818761826, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.00728584872558713, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.004874362610280514, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.13.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.2365582287311554, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.22119328379631042, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.2166043519973755, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.19719721376895905, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.111081562936306, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.10492511838674545, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.12291976809501648, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.11364411562681198, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.1123674139380455, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.09873311221599579, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.09325940161943436, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.062379490584135056, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.054216377437114716, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05305693671107292, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05277270823717117, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.031134357675909996, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.026915408670902252, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02682678960263729, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.024351999163627625, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02416204661130905, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.015979908406734467, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.015623611398041248, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.015548678115010262, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009574019350111485, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.13.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.22140401601791382, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.20007066428661346, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.19340604543685913, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.17097555100917816, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10386523604393005, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.09489385783672333, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11781296879053116, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10838445276021957, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10607356578111649, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.08809172362089157, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.080961674451828, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.060519661754369736, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05222141742706299, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.050115399062633514, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04963036999106407, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03028654307126999, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02627554163336754, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.026132896542549133, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.023162512108683586, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02283686399459839, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.016174180433154106, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016691407188773155, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.015499478206038475, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.011858669109642506, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.13.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.19743706285953522, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.18533150851726532, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.18171444535255432, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.1662396490573883, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09331581741571426, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08843939751386642, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10297204554080963, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.0953403040766716, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09433828294277191, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08342102915048599, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07907002419233322, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05254857987165451, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04580565169453621, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04488182067871094, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04466473311185837, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.026291580870747566, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02339939959347248, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.023343496024608612, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.021436845883727074, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.021295836195349693, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013980166986584663, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014557220041751862, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013681099750101566, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.01026714127510786, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.13.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2271219789981842, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.21324674785137177, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.20923563838005066, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.1913132220506668, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10708063840866089, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10148207098245621, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11796092987060547, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10934801399707794, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10824275016784668, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09571495652198792, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09062163531780243, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.059934720396995544, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05218248814344406, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.051150668412446976, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.050904933363199234, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.02990577556192875, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02595643140375614, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02588639035820961, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02359561063349247, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02344122715294361, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015392613597214222, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.01506340503692627, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.015039853751659393, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009265613742172718, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.13.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.2454363852739334, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.21840932965278625, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.20753000676631927, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.18298767507076263, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.11296367645263672, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10201369225978851, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.13455457985401154, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.12337626516819, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11718426644802094, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.0967583954334259, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.09047191590070724, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06850696355104446, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.059274010360240936, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05454691872000694, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05339296907186508, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03443659469485283, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.028715791180729866, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.02850733883678913, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.025548450648784637, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.024793092161417007, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.01857929676771164, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.01878477819263935, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.01705664023756981, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.01318351924419403, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.14.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.13948243856430054, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1304534524679184, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1276513934135437, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.11641185730695724, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06527963280677795, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06164247915148735, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07447042316198349, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06702008098363876, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06607262045145035, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.0581771656870842, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05607986822724342, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03779983147978783, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03212679177522659, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.031368810683488846, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.031187772750854492, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.018893733620643616, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.016136296093463898, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.016082480549812317, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.014702947810292244, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.014580974355340004, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.009888476692140102, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009749543853104115, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.00963597372174263, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0064965710043907166, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.14.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.11589020490646362, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.10826071351766586, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10579723119735718, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09619889408349991, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.05416993424296379, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.051033515483140945, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06036749482154846, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05581546947360039, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.05481299012899399, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04813646152615547, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04551786556839943, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.030607277527451515, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.02664276957511902, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02589181810617447, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.025711795315146446, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.015304559841752052, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.013254807330667973, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.013194230385124683, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.012020222842693329, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01190940197557211, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.007943675853312016, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.007915031164884567, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.0076899356208741665, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005111189559102058, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.14.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.25076285004615784, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.23490594327449799, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.23027724027633667, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.20988047122955322, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.11798395216464996, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.11159397661685944, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.1304692178964615, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.12064451724290848, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.11933980137109756, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10514064133167267, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.09938116371631622, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06618379056453705, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05757150426506996, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05636737123131752, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05606976896524429, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.033031828701496124, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.028560802340507507, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.028478462249040604, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.025890257209539413, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02570425532758236, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.016959184780716896, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016542453318834305, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.016525395214557648, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010096844285726547, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.14.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.21557234227657318, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1986982673406601, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.19350466132164001, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.1719345897436142, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10182162374258041, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0947527140378952, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11372148245573044, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.1045476496219635, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.1032063439488411, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.0876428559422493, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.0811222493648529, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05842694640159607, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.050276078283786774, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.04902084916830063, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.048730168491601944, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.029219433665275574, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.025473972782492638, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02539731189608574, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.022642012685537338, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.022443456575274467, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.015594204887747765, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.015789002180099487, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01518335286527872, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010986704379320145, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.14.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.19886107742786407, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.18685005605220795, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.1832175999879837, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.16764602065086365, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09391093254089355, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08902174979448318, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10348436236381531, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09591349959373474, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09493233263492584, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.0840211808681488, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.07971122115850449, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05277121439576149, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.046018436551094055, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.045100126415491104, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04488703981041908, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.02637457102537155, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.023382989689707756, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.023327454924583435, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.021396122872829437, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.021265631541609764, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013915393501520157, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014351531863212585, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013618817552924156, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009902684949338436, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.14.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.22581729292869568, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.21223236620426178, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.20818521082401276, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.19043169915676117, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10649250447750092, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10094606876373291, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11722171306610107, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10873068869113922, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10765985399484634, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09525717049837112, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09024588018655777, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05949854850769043, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05188288167119026, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.05087029188871384, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05062456056475639, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.029701851308345795, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.025817876681685448, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02574814110994339, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023475589230656624, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02332194335758686, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015269468538463116, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.01496732048690319, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01492576114833355, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.0092028072103858, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.14.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.25115999579429626, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.22453516721725464, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.2142781913280487, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.18865004181861877, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.11571941524744034, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10521332174539566, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.13573476672172546, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.1251193732023239, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11977040767669678, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09928250312805176, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.09249161928892136, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.0692046582698822, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.06003028526902199, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05577416718006134, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05474509298801422, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.0346512533724308, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.029208485037088394, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.029050905257463455, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.025962889194488525, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.025299716740846634, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.01845533214509487, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.018815170973539352, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.017085440456867218, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.013057468459010124, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.15.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.14162388443946838, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.13260866701602936, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.12987719476222992, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.11857090145349503, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06635192036628723, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06270750612020493, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07525431364774704, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06813398748636246, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06714271754026413, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.059231169521808624, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.057020362466573715, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03828722611069679, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.032618019729852676, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03180745616555214, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03162670135498047, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.019144544377923012, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.016301851719617844, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.016252607107162476, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.0148544292896986, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.014735191129148006, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.009998776018619537, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.00974875409156084, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.009735288098454475, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.006339144427329302, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.15.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.11523445695638657, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.10768011957406998, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10515383630990982, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09578180313110352, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.0538918673992157, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05074022710323334, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.060115814208984375, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05556116998195648, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.054543837904930115, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.047926824539899826, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.045392218977212906, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03049379028379917, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.026540670543909073, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.025773346424102783, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.025595609098672867, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.01522411871701479, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.013172722421586514, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.013107642531394958, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01195753738284111, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.011839245446026325, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.00791048165410757, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.007853993214666843, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.007644684985280037, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005036332178860903, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.15.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.24286943674087524, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.2274719476699829, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.22298002243041992, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.20335286855697632, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.11418163031339645, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.10807212442159653, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.12635831534862518, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.11673417687416077, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.11550803482532501, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10182718187570572, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.09636816382408142, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.0641254261136055, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05571011081337929, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05454254522919655, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05426541343331337, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03198738396167755, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.027660813182592392, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.027584295719861984, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.025075292214751244, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02490660734474659, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01642453484237194, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016046328470110893, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.016004974022507668, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0098291439935565, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.15.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.21778535842895508, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.19537445902824402, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.18835106492042542, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.16714757680892944, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.10300545394420624, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.09324809908866882, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.11566445231437683, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.10642030835151672, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.10485317558050156, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.0854174792766571, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07965550571680069, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05949283391237259, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.051212552934885025, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.04967454820871353, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04930558800697327, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.029739201068878174, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.025867953896522522, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.025764500722289085, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.022352976724505424, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.022101404145359993, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.015811778604984283, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016138244420289993, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.015314675867557526, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.011269350536167622, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.15.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.20399849116802216, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.19163863360881805, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.18791905045509338, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.17193172872066498, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.0962400883436203, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09133531898260117, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10604005306959152, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09831846505403519, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09732019901275635, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08616547286510468, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08170483261346817, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05402272567152977, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.047104693949222565, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.046179238706827164, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04595807567238808, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.026991138234734535, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.023859651759266853, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.023795055225491524, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.021824343129992485, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02168436162173748, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.014188658446073532, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014527343213558197, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013882629573345184, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009874748066067696, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.15.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2282380312681198, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.21456323564052582, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.21054188907146454, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.19270846247673035, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10755057632923126, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10202470421791077, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11839251965284348, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10977081954479218, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10872292518615723, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09625121206045151, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09128180891275406, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.06009683758020401, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05239735171198845, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.051375046372413635, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05113871768116951, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.02999064326286316, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.026057299226522446, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025987399742007256, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02371247112751007, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02355486899614334, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015411410480737686, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015093065798282623, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.015074181370437145, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009258836507797241, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.15.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.25170785188674927, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.22483736276626587, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.2148113250732422, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.189347505569458, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.11604941636323929, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.1055607944726944, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.13551701605319977, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.12506744265556335, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.12003251165151596, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09950113296508789, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.09261302649974823, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.0691242590546608, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05995108187198639, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.055851198732852936, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05485237389802933, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.034586772322654724, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.029106101021170616, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.028978103771805763, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02582686021924019, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.025180336087942123, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.01836838200688362, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.018543731421232224, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.01705409772694111, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.012658815830945969, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.16.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.13983148336410522, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.13111478090286255, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1282833367586136, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.11741691827774048, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06583845615386963, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.062293000519275665, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07480103522539139, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.0675690621137619, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06661677360534668, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.058904800564050674, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05689208582043648, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03815922886133194, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03232618793845177, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03155548870563507, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03135807439684868, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.01908143050968647, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.016094697639346123, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.016031716018915176, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.014661600813269615, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01454479806125164, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.0098743736743927, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009482668712735176, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.00961719173938036, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005992680788040161, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.16.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.11835885047912598, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11063980311155319, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10811083018779755, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09869898855686188, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.055435530841350555, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05226339027285576, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06181244179606438, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.05710652098059654, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.056097596883773804, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04937075078487396, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04675887152552605, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03134574368596077, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.027279311791062355, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.026509836316108704, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.026328086853027344, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.01566118560731411, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.013524291105568409, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.013458453118801117, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.012287269346415997, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01216613594442606, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008119095116853714, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008026547729969025, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.007854701951146126, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0051072025671601295, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.16.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.25436931848526, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.238735169172287, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.23409536480903625, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.21400780975818634, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.11990669369697571, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.11356112360954285, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.13237115740776062, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.12250228226184845, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.12121662497520447, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.1070934608578682, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.10140898823738098, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06720180809497833, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05849483609199524, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05729665234684944, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05701540410518646, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.033540770411491394, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02904074639081955, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.0289643332362175, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.02637573704123497, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.026184584945440292, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.017204880714416504, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016826270148158073, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01678205095231533, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.01029493659734726, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.16.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.19911225140094757, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.18114982545375824, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.17570656538009644, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.15411077439785004, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.09382317215204239, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.08637237548828125, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.10474781692028046, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.0965547040104866, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.09534266591072083, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.0786491185426712, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07210967689752579, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.053608689457178116, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.0463157519698143, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.045027151703834534, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04472312703728676, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.026791490614414215, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02324763499200344, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.023169709369540215, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.02018074132502079, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.019975116476416588, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.0141209801658988, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.01419780682772398, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.013700996525585651, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009564867243170738, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.16.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2019214928150177, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.18963997066020966, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.1860998272895813, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.17044249176979065, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09519145637750626, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09027576446533203, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10485463589429855, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09718982130289078, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09623599052429199, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08524616062641144, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08091027289628983, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05340452119708061, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04653303325176239, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.045635972172021866, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.045420385897159576, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.026671575382351875, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.023537252098321915, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.023482024669647217, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.021533973515033722, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.021397996693849564, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013986704871058464, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014271795749664307, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013689248822629452, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009635270573198795, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.16.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2229466736316681, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.2096204161643982, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.2056741565465927, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.1884106546640396, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10497710853815079, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09960268437862396, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.1155407652258873, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.1071387305855751, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10613952577114105, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09405864030122757, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08922097086906433, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05863958224654198, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05112796276807785, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.050148580223321915, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.049913957715034485, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.029251206666231155, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.0254381000995636, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025374434888362885, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023161478340625763, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02302032709121704, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015049145556986332, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014745188876986504, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014721715822815895, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009069227613508701, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.16.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.2488427609205246, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.22244831919670105, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.21243660151958466, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.18734097480773926, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.11462733149528503, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10420586168766022, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.1338558793067932, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.12364315986633301, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11852989345788956, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09835024178028107, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.0915956124663353, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06825846433639526, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05926928669214249, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05512883514165878, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.0541277676820755, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.034162648022174835, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.028651617467403412, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.028505288064479828, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02540627494454384, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.024760188534855843, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.01805064082145691, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.018171118572354317, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.016696130856871605, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.012277258560061455, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.17.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.13782572746276855, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.12921547889709473, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.12661802768707275, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.11580445617437363, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06463710218667984, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0612117238342762, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07360786944627762, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06634239107370377, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06537137925624847, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05783717334270477, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05591060593724251, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03750917688012123, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03174857422709465, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03100191056728363, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03083227202296257, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.018747057765722275, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.015834782272577286, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.015777215361595154, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.014439411461353302, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.014329755678772926, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.009754294529557228, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.00934506580233574, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.009511822834610939, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005949027836322784, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.17.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.11817828565835953, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11059071123600006, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10817524045705795, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09879027307033539, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.05535144358873367, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0522269569337368, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06163700297474861, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.056989558041095734, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.0560157410800457, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.049342356622219086, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04678698629140854, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.031257160007953644, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.027213502675294876, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.026463903486728668, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.02628328651189804, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.01560742687433958, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.013514566235244274, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01344917993992567, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01228241715580225, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.012167477048933506, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008095646277070045, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008014688268303871, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.007837500423192978, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005102427676320076, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.17.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.2535528838634491, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.23799730837345123, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.23343269526958466, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.21336376667022705, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.11932410299777985, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.11311809718608856, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.13187755644321442, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.12192708998918533, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.12071158736944199, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.1066693589091301, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.101110078394413, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06690683215856552, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05817825719714165, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05697283893823624, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05670510604977608, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03336473926901817, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.028864603489637375, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.028776565566658974, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.026210537180304527, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.026024004444479942, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.0171092227101326, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.01666352152824402, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.016689451411366463, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010097148828208447, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.17.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.20612835884094238, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.18712984025478363, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.18165063858032227, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.15851525962352753, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.09712515771389008, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.08941197395324707, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.10822706669569016, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.09972960501909256, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.09863657504320145, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.08127070963382721, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07379334419965744, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.055285073816776276, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.047800298780202866, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.04663407802581787, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04634756222367287, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02763989381492138, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.024153830483555794, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02409808337688446, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.0209883414208889, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02080555446445942, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.014607853256165981, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014840529300272465, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01423495914787054, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.01016731932759285, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.17.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.19990667700767517, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.18792076408863068, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.18434913456439972, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.16893841326236725, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.0941852554678917, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08937960118055344, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.1037273034453392, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09614169597625732, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09522528946399689, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.0843980461359024, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08015254139900208, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.0527583546936512, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04603071138262749, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04514700919389725, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04494118690490723, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.02634967677295208, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02326914295554161, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02321903221309185, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.0212929118424654, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02115674689412117, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013804525136947632, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014086998999118805, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013515383936464787, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009476860985159874, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.17.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.22096483409404755, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.20788024365901947, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.2040255218744278, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.1868956983089447, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10405412316322327, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09874846041202545, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11450062692165375, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10615618526935577, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10516874492168427, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09326356649398804, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08849376440048218, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.058134790509939194, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.050662361085414886, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04969865828752518, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04947148263454437, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.028995338827371597, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.025213219225406647, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02515104040503502, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.022967956960201263, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02281971089541912, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.014900699257850647, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014623547904193401, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01458031591027975, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.008997485041618347, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.17.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.2460920661687851, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.21981362998485565, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.2099081575870514, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.1850639283657074, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.11315187066793442, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.10294903069734573, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.13234856724739075, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.12205345183610916, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11713004112243652, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09713082760572433, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.0904860571026802, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06734947860240936, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05851050466299057, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05443252995610237, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.053451057523489, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03374218940734863, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.028340695425868034, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.02821689285337925, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02515852451324463, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.02452383004128933, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.01787717081606388, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.01803339645266533, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.016550466418266296, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.012275304645299911, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.18.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.13862590491771698, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.13009023666381836, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.12747229635715485, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.11675377190113068, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06506822258234024, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06161244213581085, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07446299493312836, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06671743094921112, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06581684947013855, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.058318547904491425, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05655710771679878, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03774311766028404, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03194418549537659, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.031204307451844215, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03102557361125946, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.018898246809840202, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.015895776450634003, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01583261229097843, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.014499044977128506, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.014394313097000122, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.009831473231315613, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009338042698800564, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.009591351263225079, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005874369293451309, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.18.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.11861499398946762, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11113302409648895, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10871917754411697, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09927916526794434, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.055573415011167526, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05249891057610512, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.061869461089372635, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.057234253734350204, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.05624529719352722, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.049626246094703674, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.04710110276937485, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03139369934797287, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.02731860615313053, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02658875100314617, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.026418408378958702, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.015686189755797386, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.01356285810470581, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.013504019007086754, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.012333456426858902, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.012227541767060757, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008124567568302155, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008035238832235336, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.00787094421684742, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005100534297525883, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.18.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.25396087765693665, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.23852376639842987, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.23405149579048157, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.21428321301937103, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.11962194740772247, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.11333666741847992, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.13202817738056183, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.12222205847501755, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.12087180465459824, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10701313614845276, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.10148800909519196, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.0670086219906807, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05830228328704834, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05712202936410904, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05682298541069031, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.0333966501057148, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.028918243944644928, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.028841109946370125, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.02628537453711033, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.026111235842108727, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.017127370461821556, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016696635633707047, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.016717512160539627, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010117732919752598, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.18.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.20256666839122772, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.18270938098430634, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.17678488790988922, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.15529319643974304, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.09500987082719803, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.08680659532546997, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.10629382729530334, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.09815967082977295, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.09678622335195541, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.07937260717153549, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.07233971357345581, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.05423572659492493, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04700008034706116, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.04557044431567192, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04521188884973526, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02707662060856819, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.023481054231524467, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02340790443122387, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.020340045914053917, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02010839246213436, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01419802661985159, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.014301817864179611, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.013722921721637249, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009541355073451996, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.18.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.20001046359539032, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.18817579746246338, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.18463535606861115, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.16924983263015747, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09421144425868988, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.08947642892599106, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10375043749809265, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09616746753454208, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09522850811481476, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08453971147537231, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08022232353687286, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.052755773067474365, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.045996345579624176, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04510302469134331, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.0448952242732048, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.02634281851351261, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.023168757557868958, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02311098761856556, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02118847146630287, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.021056752651929855, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013738618232309818, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.01389390043914318, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013450848869979382, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009172500111162663, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.18.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.219919815659523, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.20684237778186798, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.20306722819805145, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.18614594638347626, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10347434878349304, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09822005778551102, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11389202624559402, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.1055661290884018, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10458843410015106, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09286060184240341, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08812309056520462, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05779748409986496, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.050376906991004944, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.049414440989494324, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.049193527549505234, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.028828153386712074, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02507607452571392, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025009408593177795, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.022861657664179802, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.022719213739037514, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.014844316989183426, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014539379626512527, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014520732685923576, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.00894497986882925, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.18.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.23834732174873352, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.21200799942016602, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.2018144577741623, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.17772026360034943, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10929019004106522, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.09893041104078293, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.1285630166530609, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.11872565001249313, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11335296928882599, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09349217265844345, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.08703593909740448, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06538640707731247, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.056823763996362686, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05254676565527916, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05149953067302704, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03271108120679855, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.0272719357162714, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.027117356657981873, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.024111146107316017, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.023432141169905663, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.017263319343328476, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.017292380332946777, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.015875358134508133, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.011562490835785866, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.19.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.14314576983451843, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.13436461985111237, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1315939724445343, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12072128057479858, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06715173274278641, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06357771158218384, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07663358002901077, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.0688149556517601, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06789959967136383, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06020315736532211, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.058446988463401794, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03896147012710571, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03291771560907364, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03215347230434418, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03198747709393501, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.019465861842036247, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.01639670692384243, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01633846014738083, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.014969469048082829, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.014855421148240566, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.010099519044160843, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009649666026234627, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.009853164665400982, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.006102549843490124, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.19.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.12670688331127167, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11868872493505478, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.11612701416015625, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10612496733665466, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.05933765321969986, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.056056998670101166, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.0661885216832161, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06106593832373619, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.060060422867536545, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.053012825548648834, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.050349947065114975, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.033588118851184845, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.02914273738861084, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.028367463499307632, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.028189735487103462, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.016755100339651108, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.014451098628342152, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.014390111900866032, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.013150634244084358, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.013028278015553951, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.00867460947483778, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008522890508174896, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.008398273028433323, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00537140155211091, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.19.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.2531066834926605, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.23772075772285461, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.23322923481464386, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.2135002613067627, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.1192575991153717, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.1130313128232956, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.13162249326705933, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.1218205988407135, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.12054669111967087, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10668599605560303, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.10129621624946594, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06684970110654831, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.0581616573035717, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05697663873434067, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05672250688076019, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.0333632193505764, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.029044196009635925, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02896382287144661, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.026455195620656013, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.026278510689735413, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01723043993115425, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.017037753015756607, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01683049276471138, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010767935775220394, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.19.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.18300634622573853, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.16554880142211914, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.16047590970993042, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.13974407315254211, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.0864657312631607, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.07943621277809143, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.09694398939609528, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.08890999853610992, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.0878530740737915, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.07191010564565659, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.06586433202028275, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.04977721720933914, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.04280659928917885, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.04170268028974533, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.04145898297429085, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02494252845644951, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.021811462938785553, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.021747812628746033, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.018919669091701508, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.018749818205833435, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.013376409187912941, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.013689147308468819, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.013036671094596386, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009716492146253586, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.19.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.20165975391864777, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.18965357542037964, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.1860904097557068, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.17049044370651245, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09496597945690155, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09016677737236023, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10459128767251968, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09699499607086182, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09601578861474991, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08516623824834824, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08086109161376953, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05317782983183861, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04632727429270744, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04542980343103409, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.045218855142593384, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.026557590812444687, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02324685826897621, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.023183023557066917, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.021230272948741913, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.021094132214784622, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.013808912597596645, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.013784981332719326, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.013510788790881634, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.008922240696847439, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.19.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.21867768466472626, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.20578758418560028, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.20207765698432922, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.1853613257408142, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.102934330701828, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09776132553815842, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11331702768802643, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10506968200206757, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10407115519046783, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09236232191324234, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08769051730632782, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.057555485516786575, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.050103913992643356, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.049145281314849854, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.048926129937171936, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.028715383261442184, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02493942342698574, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.024877125397324562, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.022742867469787598, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.022602232173085213, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.014834682457149029, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.0144620630890131, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014514219015836716, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.00889480859041214, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.19.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.22727333009243011, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.20227503776550293, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.19208481907844543, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.1688685268163681, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10415606200695038, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.09404739737510681, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.12381276488304138, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.11401344835758209, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.10803709179162979, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.08911868184804916, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.08308953791856766, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06279825419187546, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.054488323628902435, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.050037186592817307, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.04894155263900757, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03137343376874924, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.025909338146448135, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.025702303275465965, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02289222739636898, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.022181851789355278, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.01650266721844673, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.016414372250437737, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.015049755573272705, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.01083010621368885, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.20.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.14341716468334198, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.13455171883106232, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.13205890357494354, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.1209348738193512, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06740327179431915, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06386515498161316, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07683317363262177, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06906737387180328, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06817538291215897, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.060459204018116, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.058690957725048065, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03915136307477951, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.033028494566679, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.032280780375003815, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03209418058395386, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.019551416859030724, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.016411202028393745, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.016359012573957443, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.014978528022766113, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.014866000041365623, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.010136437602341175, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009597623720765114, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.009884648025035858, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005977766588330269, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.20.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.12770022451877594, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11973707377910614, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.11713780462741852, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10712462663650513, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.059903256595134735, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05661039426922798, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06665055453777313, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06157733127474785, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.060585349798202515, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05350131541490555, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05079333111643791, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03383636102080345, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.029376978054642677, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.028619306161999702, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.0284409262239933, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.016880568116903305, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.014595676213502884, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.014531238004565239, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.013282956555485725, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01316536869853735, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.0087510971352458, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.0086178258061409, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.00848130788654089, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005448906682431698, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.20.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.25543874502182007, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.24001456797122955, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.23540887236595154, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.21576200425624847, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.12028886377811432, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.11407426744699478, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.13275396823883057, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.12290143221616745, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.12156940251588821, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10771604627370834, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.10216333717107773, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06735388934612274, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05863694101572037, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.057433780282735825, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05715540796518326, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03359723091125488, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.029121601954102516, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02903682366013527, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.026506802067160606, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.026316868141293526, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.017249910160899162, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016866037622094154, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01684986986219883, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010303394868969917, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.20.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.15441103279590607, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.13498930633068085, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1296164095401764, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.11513679474592209, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.07224857807159424, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06411423534154892, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.0815562903881073, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.07458613812923431, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.07389822602272034, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05909300222992897, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05499792844057083, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.041887227445840836, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.036231547594070435, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03513294458389282, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.0348488874733448, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02108994871377945, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.018849093466997147, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01881108619272709, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.016353074461221695, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.016188442707061768, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.011698932386934757, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.012498660013079643, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.011364525184035301, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009521082043647766, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.20.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.20611755549907684, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.1940130591392517, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.19039922952651978, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.17461562156677246, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09706863015890121, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09216851741075516, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.1069478690624237, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.09910241514444351, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.09811224043369293, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08708460628986359, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08275798708200455, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.054375987499952316, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.047331515699625015, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04641444608569145, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04621146619319916, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.027155116200447083, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02371882274746895, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02365879714488983, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.021666748449206352, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.021531833335757256, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.01415195781737566, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014013171195983887, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01385258138179779, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.008993489667773247, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.20.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2200937420129776, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.20713673532009125, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.203372985124588, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.18658128380775452, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10359205305576324, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09838467836380005, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11407521367073059, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10570665448904037, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10470294207334518, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09297040849924088, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08831106871366501, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05791373923420906, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05043491721153259, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.049462735652923584, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04924148693680763, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.028885554522275925, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.025094328448176384, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025031091645359993, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02288191020488739, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.022734761238098145, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.014884300529956818, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014546426944434643, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014563530683517456, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.008920127525925636, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.20.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.2275831252336502, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.20303435623645782, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.19316090643405914, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.1701783537864685, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10443529486656189, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.09457297623157501, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.12288148701190948, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.11371370404958725, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.10816394537687302, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.0895586833357811, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.08346112072467804, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06250868737697601, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.0544419102370739, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05022639408707619, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.04920084774494171, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.031277600675821304, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.02609388902783394, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.02590673416852951, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.023121101781725883, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.02244168147444725, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.016470348462462425, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.01660013571381569, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.015110127627849579, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.01112616527825594, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.21.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.1444173902273178, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1357915848493576, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1330721229314804, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12194399535655975, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06788647919893265, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06432712823152542, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07737476378679276, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06957543641328812, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06870517879724503, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06087140738964081, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05910865217447281, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.039273664355278015, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.033259067684412, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03250189870595932, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.032318077981472015, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.019589412957429886, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.016511905938386917, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01645800843834877, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.015063682571053505, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.014942368492484093, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.010132724419236183, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009649264626204967, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.009878459386527538, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00598738482221961, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.21.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.12875548005104065, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1207335814833641, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.11816428601741791, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10804414004087448, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06044397130608559, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05715043470263481, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06731980293989182, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.062141891568899155, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06116623803973198, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05400337278842926, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05130702257156372, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03415168821811676, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.0296653863042593, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.028895525261759758, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.028711434453725815, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.01704990677535534, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.014754085801541805, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01468760333955288, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.013421918265521526, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.013306588865816593, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008841965347528458, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008728711865842342, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.008576528169214725, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005541397258639336, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.21.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.2529781758785248, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.23772548139095306, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.23326179385185242, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.21369601786136627, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.11911892145872116, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.11287347972393036, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.13136166334152222, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.12166471034288406, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.12040889263153076, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10664442181587219, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.10112583637237549, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06659670919179916, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05803508684039116, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05686812847852707, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05659419670701027, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03322929888963699, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.028777901083230972, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.028704756870865822, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.026173090562224388, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.026005985215306282, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.017031632363796234, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016614215448498726, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01663377322256565, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.01007852517068386, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.21.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.16780367493629456, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1477764993906021, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.14149068295955658, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12626202404499054, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.07878562808036804, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.07001198083162308, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.08895529806613922, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.08210951834917068, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.08048757165670395, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06472572684288025, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.060511719435453415, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.04565016180276871, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.039489880204200745, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03798528388142586, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.0376286655664444, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.022811468690633774, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.019885197281837463, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.019790777936577797, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.017103739082813263, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.016871249303221703, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.012169182300567627, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.012597886845469475, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.011685232631862164, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.008928277529776096, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.21.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.21044792234897614, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.1979154795408249, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.19423127174377441, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.17809544503688812, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.09909678995609283, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09405159205198288, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10918045789003372, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10118132829666138, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.1001758947968483, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08885364234447479, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08437573164701462, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05546419695019722, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.0483231395483017, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.047373559325933456, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04714968800544739, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.027702443301677704, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.024166179820895195, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02409859001636505, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.022056594491004944, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.021917007863521576, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.01438163872808218, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014223437756299973, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01406572200357914, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.00903978943824768, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.21.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2249213606119156, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.21161288022994995, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.20768916606903076, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.1905214935541153, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10595736652612686, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10058864951133728, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11666125059127808, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10812466591596603, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10710534453392029, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09509243071079254, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09029312431812286, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.059243831783533096, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05160451680421829, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.050616517663002014, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05037381872534752, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.029565850272774696, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.025685152038931847, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02561981789767742, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023417260497808456, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.023267941549420357, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015254956670105457, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014903491362929344, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01492367871105671, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.00918327271938324, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.21.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.23531071841716766, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.21009689569473267, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.2003958821296692, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.17644710838794708, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10800448060035706, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.09808023273944855, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.12636567652225494, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.1169036477804184, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.1117774099111557, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.0925813764333725, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.08619856834411621, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06421582400798798, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.055805690586566925, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05178547278046608, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05080400034785271, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.032079584896564484, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.026597099378705025, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.0264570452272892, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02346619963645935, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.022813841700553894, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.01678072288632393, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.01647070050239563, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.015471416525542736, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.010528280399739742, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.22.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.14518995583057404, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.13639014959335327, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.13346198201179504, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12252120673656464, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.0683637410402298, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.0647772029042244, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07799161225557327, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.07007209211587906, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06915601342916489, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.061329640448093414, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05938390642404556, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03961099684238434, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.033498361706733704, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03272886574268341, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.0325537770986557, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.019777212291955948, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.01664099097251892, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.016583215445280075, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01518093328922987, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.015067734755575657, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01022936962544918, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.00973854772746563, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.009978842921555042, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.006070688832551241, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.22.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.12800027430057526, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11994028836488724, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.11728285998106003, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10733652114868164, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06005708873271942, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05675068870186806, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06703127920627594, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.06179770082235336, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.0607762411236763, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05364247411489487, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05099618434906006, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03399891406297684, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.0294969342648983, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02873177081346512, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.028554268181324005, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.01699356734752655, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.014687350019812584, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.014619520865380764, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.013356773182749748, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01324409805238247, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008835416287183762, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008724668063223362, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.00856034830212593, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005586417857557535, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.22.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.2588341236114502, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.24322223663330078, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.2385697066783905, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.21848896145820618, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.12201983481645584, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.11570729315280914, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.13456310331821442, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.12469889968633652, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.12334248423576355, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10918718576431274, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.10363827645778656, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06833699345588684, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.059511423110961914, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.058316223323345184, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.058023665100336075, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03411329537630081, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.029571404680609703, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.029486829414963722, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.026891814544796944, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.02671586349606514, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.017506282776594162, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.01712910085916519, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.017093084752559662, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010496992617845535, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.22.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.16582970321178436, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.14574426412582397, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.13966307044029236, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12253310531377792, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.0772428959608078, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06905660033226013, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.08813897520303726, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.08117052167654037, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.07969523966312408, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06337802112102509, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05877091735601425, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.045294150710105896, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03917008265852928, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03743767365813255, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03701566532254219, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.022711554542183876, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.019814029335975647, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.019724318757653236, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01713067851960659, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.016866818070411682, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.012271536514163017, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.012900959700345993, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.011722038500010967, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.009456413798034191, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.22.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.21472732722759247, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.20182645320892334, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.19796817004680634, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.1815938800573349, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10115010291337967, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09597964584827423, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.1115260049700737, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10332759469747543, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10228568315505981, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09066777676343918, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08606542646884918, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.056658465415239334, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.0493420772254467, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.048367977142333984, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04813045263290405, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.028307653963565826, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.024672171100974083, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.024603858590126038, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02251027151942253, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.02236047573387623, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.014719732105731964, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014515524730086327, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014400606043636799, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009222028777003288, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.22.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.22482110559940338, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.2114790678024292, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.20758801698684692, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.19033707678318024, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10597153007984161, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10058960318565369, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11671875417232513, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.1081463098526001, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10711836069822311, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09506680816411972, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09022407978773117, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.0592709481716156, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05162505805492401, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.050623804330825806, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05039140209555626, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.029593292623758316, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.025703126564621925, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025632116943597794, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023422183468937874, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.023264218121767044, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015288863331079483, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014930331148207188, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014959057793021202, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009214591234922409, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.22.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.23549656569957733, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.21044664084911346, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.20071692764759064, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.17683526873588562, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10832957178354263, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.09832609444856644, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.12731805443763733, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.11730222404003143, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11206676065921783, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.09297657757997513, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.0866839736700058, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.064805768430233, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05630158632993698, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05214463919401169, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.05113014578819275, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.03253503516316414, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.02718980982899666, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.027025241404771805, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02414456196129322, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.02348804846405983, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.017277033999562263, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.017400339245796204, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.015919698402285576, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.01186734065413475, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.23.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.1454857438802719, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.13655763864517212, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1338057816028595, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12261617928743362, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06857234239578247, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06495147198438644, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07768061012029648, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.0703090950846672, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06937988102436066, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06140822172164917, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05928808078169823, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.039446305483579636, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03356364741921425, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.032784461975097656, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03258707746863365, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.019685834646224976, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.016650686040520668, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.016594383865594864, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.015164931304752827, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.015043603256344795, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01014303881675005, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009712168015539646, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.00988694652915001, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00600779103115201, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.23.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.12887205183506012, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.12076430022716522, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.11806975305080414, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10806158930063248, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.060666508972644806, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.057284049689769745, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.0676237940788269, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.062356021255254745, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06137053668498993, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.054123878479003906, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05139199644327164, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.034374359995126724, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.029811877757310867, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.029034556820988655, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.028848370537161827, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.017177825793623924, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.014870326034724712, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.014810127206146717, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.013536092825233936, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.013419533148407936, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008974825032055378, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008889357559382915, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.00869837962090969, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.005766647402197123, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.23.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.2548714280128479, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.23965777456760406, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.23498490452766418, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.21522462368011475, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.1203666403889656, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.11414679884910583, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.13278256356716156, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.12304496765136719, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.12171542644500732, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10772284120321274, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.10217474400997162, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06739206612110138, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05872458219528198, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.0575624518096447, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05728621408343315, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03364262357354164, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02916564792394638, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.029083415865898132, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.02650785632431507, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.026342209428548813, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.017267638817429543, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016891706734895706, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.016864217817783356, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010312376543879509, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.23.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.14615510404109955, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.12520156800746918, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.11894338577985764, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10224401205778122, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06820745766162872, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.05892413109540939, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07799497991800308, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.07152938097715378, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.07017140090465546, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.05375879630446434, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.049291569739580154, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.040189579129219055, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03466345742344856, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.033108681440353394, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.03273671865463257, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02017257921397686, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.017707977443933487, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.017619889229536057, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.014953810721635818, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.014705696143209934, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.011035536415874958, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.01173563115298748, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.01055755652487278, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00880598183721304, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.23.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2159627079963684, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.20296524465084076, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.19923625886440277, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.18252745270729065, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.1019270047545433, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.0966518297791481, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11247771233320236, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10412774235010147, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10306619107723236, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.09130600839853287, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08675737679004669, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.057287558913230896, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04977806657552719, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.04879622161388397, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.04856017976999283, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.028624868020415306, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.02499019168317318, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02492031268775463, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.022807277739048004, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.022660112008452415, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015009729191660881, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.014842327684164047, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014690618962049484, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009633062407374382, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.23.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.22634652256965637, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.21294716000556946, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.20894156396389008, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.19146890938282013, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10689284652471542, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.10138554126024246, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.1178300678730011, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.1091136708855629, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10804735869169235, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.0957818254828453, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.09096531569957733, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.06006273627281189, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.05214859917759895, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.051132626831531525, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.05088400840759277, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.029993819072842598, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.026112595573067665, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.02604636736214161, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023827167227864265, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.023665376007556915, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015676943585276604, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015389041975140572, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.015351401641964912, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.009838318452239037, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.23.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.2326476275920868, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.207753524184227, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.19775760173797607, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.17425115406513214, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.10689593851566315, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.09685611724853516, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.12784193456172943, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.11642196029424667, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.11066057533025742, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.0919051244854927, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.08575235307216644, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.06473514437675476, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.05598801001906395, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.05152514576911926, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.050425685942173004, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.032547786831855774, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.027013199403882027, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.026786817237734795, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.02405993826687336, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.02334412932395935, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.017500966787338257, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.01752910390496254, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.016010187566280365, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.012120410799980164, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.24.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.14661987125873566, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.13763752579689026, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1348012387752533, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12367857992649078, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06913929432630539, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06547852605581284, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07821737229824066, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.07088427990674973, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06994965672492981, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06191891431808472, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.059777773916721344, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03973753750324249, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.033861834555864334, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03307928517460823, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.032897453755140305, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.019853342324495316, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.016827888786792755, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01676870696246624, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.0153347821906209, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.01521651167422533, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01028221845626831, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009846328757703304, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.010025691241025925, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00614262605085969, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.24.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.1271236687898636, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.11918609589338303, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.11665181815624237, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10667489469051361, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.05988960340619087, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.056592199951410294, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.06672855466604233, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.061524998396635056, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.0606011264026165, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.053464002907276154, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.050800494849681854, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03391532227396965, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.0294073224067688, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.028681376948952675, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.028496475890278816, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.016945285722613335, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.014676607213914394, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.014619697816669941, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.013361388817429543, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.013249016366899014, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.00882686022669077, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008754357695579529, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.008566354401409626, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00567243155092001, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.24.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.25743186473846436, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.24201348423957825, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.23725560307502747, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.21731549501419067, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.12153765559196472, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.11523079127073288, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.13399426639080048, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.12410873174667358, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.12285064905881882, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10871393233537674, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.10309570282697678, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06807035207748413, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.059270117431879044, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05809199437499046, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.057813551276922226, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03397807851433754, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02945038117468357, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.02936631627380848, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.026775037869811058, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.026597242802381516, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01743677444756031, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.01704547554254532, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.017029451206326485, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010418127290904522, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.24.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.15161611139774323, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.12884779274463654, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.12088136374950409, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.10518001765012741, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06971026957035065, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.060077689588069916, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.08283381164073944, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.07606106996536255, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.07320544868707657, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.0563041977584362, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.05089040845632553, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.04259109869599342, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.036795783787965775, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03395271673798561, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.033255863934755325, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.02137165144085884, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.018236882984638214, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.018060287460684776, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.015697218477725983, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.015241528861224651, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.01165956724435091, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.012391820549964905, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.0107711898162961, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.00926202442497015, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.24.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.2092934548854828, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.19675375521183014, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.19299468398094177, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.17686204612255096, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.0990162342786789, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09385767579078674, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.10942989587783813, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10112587362527847, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.1001068651676178, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.08864196389913559, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08423217386007309, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05600164458155632, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04853291064500809, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.0475776270031929, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.047346726059913635, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.028036339208483696, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.0246818158775568, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.024610092863440514, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.02258419804275036, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.022438863292336464, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.01503574289381504, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.01514427363872528, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.014722399413585663, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.010461967438459396, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.24.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.21274413168430328, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.20000873506069183, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.19623693823814392, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.17989885807037354, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.10074187070131302, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.09557008743286133, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.11146622896194458, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.10286534577608109, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.10185786336660385, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.0902334675192833, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.08592050522565842, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.05713456869125366, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.04943370074033737, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.048468876630067825, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.048242129385471344, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.02862831950187683, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.025267167016863823, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.025204751640558243, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.023165004327893257, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.0230289027094841, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.015500693581998348, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.015697410330176353, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.015185362659394741, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.011058187112212181, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.24.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.20348414778709412, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.18198680877685547, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.17292791604995728, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.1520152986049652, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.09375543147325516, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.08473530411720276, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.11406999826431274, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.10267754644155502, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.09693729877471924, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.08065222948789597, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.07537679374217987, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.0575062595307827, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.04964646324515343, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.045410770922899246, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.044390782713890076, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.02889842912554741, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.02417767234146595, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.02395208179950714, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.021647505462169647, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.02100905030965805, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.015769798308610916, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.016188858076930046, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.014452554285526276, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.011762655340135098, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.25.self_attn.q_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.1464356631040573, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.1375141590833664, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.1347154974937439, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.12320912629365921, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.06903116405010223, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.06535939872264862, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.07695188373327255, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.07067370414733887, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.06978702545166016, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.06169532984495163, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.058833323419094086, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.03913841396570206, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.03375132009387016, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.03300619125366211, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.032812293618917465, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.019539855420589447, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.016788024455308914, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.01673712581396103, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01528443768620491, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.015168924815952778, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.010121665894985199, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.009841706603765488, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.009866515174508095, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.006172092165797949, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.25.self_attn.k_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.11629855632781982, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.10901308059692383, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.10673041641712189, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.09756254404783249, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.0548287108540535, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.051820725202560425, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.061163026839494705, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.056274380534887314, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.05544334650039673, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.04894198477268219, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.046534452587366104, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.031223032623529434, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.026963017880916595, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.026297755539417267, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.02614857628941536, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.015640536323189735, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.013562341220676899, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.013508697040379047, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.012363381683826447, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.012266717851161957, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.008275890722870827, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.00822184607386589, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.008037367835640907, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0055092088878154755, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.25.self_attn.v_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.24555432796478271, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.23102794587612152, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.22680096328258514, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.20747560262680054, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.11581214517354965, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.10987112671136856, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.1287524551153183, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.11817196756601334, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.11706817150115967, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.10360956192016602, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.09887416660785675, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.06538081914186478, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.05651138350367546, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.05542762205004692, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.05515563488006592, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.03267846256494522, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.02815995365381241, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.028077557682991028, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.025618573650717735, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.025457043200731277, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.016963809728622437, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.016409078612923622, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.016597015783190727, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.010198758915066719, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.25.self_attn.o_proj", - "numel": 10240000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.16546875, - "total_bits": 22174400.0, - "err": 0.09425117820501328, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.36546875, - "total_bits": 24222399.999999996, - "err": 0.07968917489051819, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.60546875, - "total_bits": 26680000.0, - "err": 0.0758289024233818, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.68546875, - "total_bits": 27499200.0, - "err": 0.06574373692274094, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.20546875, - "total_bits": 32824000.0, - "err": 0.04351038858294487, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.72546875, - "total_bits": 38148800.0, - "err": 0.038130983710289, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0313671875, - "total_bits": 31041200.0, - "err": 0.05009010434150696, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 32004800.0, - "err": 0.045428093522787094, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.16546875, - "total_bits": 32414400.0, - "err": 0.04453252628445625, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 36100800.0, - "err": 0.03426105156540871, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.662734375, - "total_bits": 37506400.0, - "err": 0.03231264650821686, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.0313671875, - "total_bits": 41281200.0, - "err": 0.025693228468298912, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 42244799.99999999, - "err": 0.02259751968085766, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.20546875, - "total_bits": 43064000.0, - "err": 0.02167799323797226, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.28546875, - "total_bits": 43883200.0, - "err": 0.021482549607753754, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.0313671875, - "total_bits": 51521200.0, - "err": 0.013195756822824478, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.20546875, - "total_bits": 53304000.0, - "err": 0.012196114286780357, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.28546875, - "total_bits": 54123200.0, - "err": 0.012135125696659088, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 56580800.0, - "err": 0.01053757593035698, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.68546875, - "total_bits": 58219200.0, - "err": 0.010408276692032814, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.0313671875, - "total_bits": 61761200.0, - "err": 0.007627642247825861, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 62724799.99999999, - "err": 0.008762137964367867, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.1913671875, - "total_bits": 63399600.0, - "err": 0.007364828139543533, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 83204800.0, - "err": 0.0072377147153019905, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.25.mlp.gate_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.1721547693014145, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.1617841273546219, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.1586536020040512, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.14526189863681793, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.08132694661617279, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.07705539464950562, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.08979707956314087, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.08309578895568848, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.08223839849233627, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.07273764908313751, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.0690547302365303, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.04570537433028221, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.039763711392879486, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.03895808011293411, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.038773298263549805, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.022831523790955544, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.019960181787610054, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.019906707108020782, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.018205417320132256, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.018084215000271797, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.011876154690980911, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.011897093616425991, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.01160617358982563, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.007762377616018057, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.25.mlp.up_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1651736111111113, - "total_bits": 59862720.00000001, - "err": 0.16338960826396942, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.365173611111111, - "total_bits": 65392320.0, - "err": 0.15349538624286652, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6051736111111112, - "total_bits": 72027840.0, - "err": 0.15046407282352448, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.6851736111111113, - "total_bits": 74239680.0, - "err": 0.13777965307235718, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2051736111111113, - "total_bits": 88616640.0, - "err": 0.07713047415018082, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.725173611111111, - "total_bits": 102993600.0, - "err": 0.07311083376407623, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.0312934027777776, - "total_bits": 83809200.0, - "err": 0.08545268326997757, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.1251736111111112, - "total_bits": 86404800.0, - "err": 0.07883738726377487, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1651736111111113, - "total_bits": 87510720.0, - "err": 0.07801057398319244, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.525173611111111, - "total_bits": 97464000.0, - "err": 0.06898193806409836, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6625868055555557, - "total_bits": 101263200.0, - "err": 0.0655483603477478, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031293402777778, - "total_bits": 111457200.0, - "err": 0.04354318603873253, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.125173611111111, - "total_bits": 114052800.0, - "err": 0.03770548105239868, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.205173611111111, - "total_bits": 116264640.0, - "err": 0.036927711218595505, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.285173611111111, - "total_bits": 118476480.00000001, - "err": 0.036738283932209015, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031293402777778, - "total_bits": 139105200.0, - "err": 0.021770557388663292, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.205173611111111, - "total_bits": 143912640.0, - "err": 0.018861861899495125, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.285173611111111, - "total_bits": 146124480.0, - "err": 0.01881159096956253, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.525173611111111, - "total_bits": 152760000.0, - "err": 0.017182955518364906, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.685173611111111, - "total_bits": 157183680.0, - "err": 0.017064694315195084, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031293402777778, - "total_bits": 166753200.0, - "err": 0.011423103511333466, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.125173611111111, - "total_bits": 169348800.0, - "err": 0.011146057397127151, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.191293402777778, - "total_bits": 171176880.0, - "err": 0.011152730323374271, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12517361111111, - "total_bits": 224644799.99999997, - "err": 0.007139462046325207, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - }, - { - "key": "model.layers.25.mlp.down_proj", - "numel": 27648000, - "options": [ - { - "desc": "0.05:3b/0.95:2b 32g s4", - "bpw": 2.1699131944444443, - "total_bits": 59993759.99999999, - "err": 0.08544200658798218, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:3b/0.75:2b 32g s4", - "bpw": 2.377320601851852, - "total_bits": 65728160.00000001, - "err": 0.07529199868440628, - "qparams": { - "group_size": 32, - "bits": [ - 3, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.25:4b/0.75:2b 32g s4", - "bpw": 2.6291724537037036, - "total_bits": 72691360.0, - "err": 0.07018117606639862, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 2 - ], - "bits_prop": [ - 0.25, - 0.75 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4", - "bpw": 2.732876157407407, - "total_bits": 75558560.0, - "err": 0.06185039505362511, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3, - 2 - ], - "bits_prop": [ - 0.1, - 0.4, - 0.5 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:4b/0.9:3b 32g s4", - "bpw": 3.2291724537037036, - "total_bits": 89280160.0, - "err": 0.039426952600479126, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.2:6b/0.8:3b 32g s4", - "bpw": 3.7476909722222222, - "total_bits": 103616160.0, - "err": 0.034815605729818344, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 3 - ], - "bits_prop": [ - 0.2, - 0.8 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 128g s4", - "bpw": 3.031599537037037, - "total_bits": 83817664.0, - "err": 0.0497053824365139, - "qparams": { - "group_size": 128, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:3b 32g s4", - "bpw": 3.12546875, - "total_bits": 86412960.0, - "err": 0.04491685330867767, - "qparams": { - "group_size": 32, - "bits": [ - 3 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:4b/0.95:3b 32g s4", - "bpw": 3.1699131944444443, - "total_bits": 87641760.0, - "err": 0.04106815531849861, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.05, - 0.95 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:4b/0.6:3b 32g s4", - "bpw": 3.52546875, - "total_bits": 97472160.0, - "err": 0.03367484360933304, - "qparams": { - "group_size": 32, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.6:4b/0.4:3b 64g s4", - "bpw": 3.6701417824074074, - "total_bits": 101472080.0, - "err": 0.03177451714873314, - "qparams": { - "group_size": 64, - "bits": [ - 4, - 3 - ], - "bits_prop": [ - 0.6, - 0.4 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 128g s4", - "bpw": 4.031599537037037, - "total_bits": 111465664.0, - "err": 0.025475122034549713, - "qparams": { - "group_size": 128, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:4b 32g s4", - "bpw": 4.12546875, - "total_bits": 114060959.99999999, - "err": 0.02190939150750637, - "qparams": { - "group_size": 32, - "bits": [ - 4 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:5b/0.9:4b 32g s4", - "bpw": 4.229172453703704, - "total_bits": 116928160.00000001, - "err": 0.0194783303886652, - "qparams": { - "group_size": 32, - "bits": [ - 5, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:4b 32g s4", - "bpw": 4.332876157407408, - "total_bits": 119795360.00000001, - "err": 0.0188737940043211, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 4 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:5b 128g s4", - "bpw": 5.031599537037037, - "total_bits": 139113664.0, - "err": 0.012976275756955147, - "qparams": { - "group_size": 128, - "bits": [ - 5 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:6b/0.9:5b 32g s4", - "bpw": 5.229172453703704, - "total_bits": 144576160.0, - "err": 0.010946308262646198, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4", - "bpw": 5.303246527777778, - "total_bits": 146624160.0, - "err": 0.01077985018491745, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.05, - 0.05, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.4:6b/0.6:5b 32g s4", - "bpw": 5.52546875, - "total_bits": 152768160.0, - "err": 0.009921814315021038, - "qparams": { - "group_size": 32, - "bits": [ - 6, - 5 - ], - "bits_prop": [ - 0.4, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4", - "bpw": 5.732876157407407, - "total_bits": 158502560.0, - "err": 0.009556741453707218, - "qparams": { - "group_size": 32, - "bits": [ - 8, - 6, - 5 - ], - "bits_prop": [ - 0.1, - 0.3, - 0.6 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 128g s4", - "bpw": 6.031599537037037, - "total_bits": 166761664.0, - "err": 0.007380172144621611, - "qparams": { - "group_size": 128, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:6b 32g s4", - "bpw": 6.12546875, - "total_bits": 169356960.0, - "err": 0.008058863691985607, - "qparams": { - "group_size": 32, - "bits": [ - 6 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - }, - { - "desc": "0.1:8b/0.9:6b 128g s4", - "bpw": 6.239006944444444, - "total_bits": 172496064.0, - "err": 0.006586990784853697, - "qparams": { - "group_size": 128, - "bits": [ - 8, - 6 - ], - "bits_prop": [ - 0.1, - 0.9 - ], - "scale_bits": 4 - } - }, - { - "desc": "1.0:8b 32g s4", - "bpw": 8.12546875, - "total_bits": 224652960.0, - "err": 0.006431732792407274, - "qparams": { - "group_size": 32, - "bits": [ - 8 - ], - "bits_prop": [ - 1.0 - ], - "scale_bits": 4 - } - } - ] - } - ], - "last_module_idx": 54, - "base_perplexity": 11.946749208375252 + "measurement": { + "model.layers.0.self_attn": [ + { + "accuracy": 0.9177240133285522, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9217327833175659, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9360260963439941, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9545134902000427, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9605870246887207, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9651926159858704, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9716001749038696, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9770992398262024, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.980593204498291, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9811851978302002, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9825040102005005, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9833828210830688, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9847335815429688, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9856932759284973, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.99090176820755, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9911166429519653, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9924720525741577, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9930375814437866, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951440691947937, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.0.mlp": [ + { + "accuracy": 0.9365872740745544, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9379965662956238, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9532368779182434, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9582359194755554, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9715542197227478, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9732175469398499, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9801041483879089, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9847777485847473, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9861656427383423, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9855791926383972, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9869512915611267, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923688769340515, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923235774040222, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950206279754639, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956074953079224, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965313673019409, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969714879989624, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.1.self_attn": [ + { + "accuracy": 0.9129329919815063, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9170666933059692, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9275577068328857, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9523346424102783, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9577993154525757, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9586471319198608, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9717002511024475, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9727781414985657, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9755765199661255, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9765754342079163, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9784319996833801, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9795116782188416, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9808445572853088, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9819432497024536, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9880783557891846, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9889917969703674, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9900076389312744, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9918657541275024, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940392374992371, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.1.mlp": [ + { + "accuracy": 0.8965533971786499, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.8988513350486755, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9194796085357666, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.925621747970581, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9502027630805969, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9537826776504517, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9626827836036682, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9742567539215088, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9768008589744568, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9746943712234497, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9777461290359497, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9871538281440735, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9879993200302124, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9925115704536438, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929497838020325, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9943098425865173, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959841966629028, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.2.self_attn": [ + { + "accuracy": 0.9176256656646729, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9214654564857483, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9327325224876404, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9556199908256531, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9605461955070496, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.961024284362793, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9764515161514282, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9770482778549194, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9791052341461182, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9799665212631226, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9802554249763489, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9815042018890381, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9834611415863037, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9846301674842834, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9907756447792053, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.991156280040741, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9931634068489075, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935560822486877, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958291053771973, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.2.mlp": [ + { + "accuracy": 0.904305100440979, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9106094241142273, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9213430881500244, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9244486689567566, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9601025581359863, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9621535539627075, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9674994945526123, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9778563380241394, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9797559976577759, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9804362654685974, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9824299812316895, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.98982173204422, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9902949929237366, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935250878334045, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9939770698547363, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9947821497917175, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958178400993347, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.3.self_attn": [ + { + "accuracy": 0.9432005286216736, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9445347785949707, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9540787935256958, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.964008629322052, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9678102135658264, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9698260426521301, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9740789532661438, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9765285849571228, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9798696637153625, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9803239703178406, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9849774837493896, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9865783452987671, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9864001870155334, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9880905151367188, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922593832015991, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935751557350159, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9932655692100525, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964281916618347, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974014163017273, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.3.mlp": [ + { + "accuracy": 0.910824179649353, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9138386249542236, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9278630018234253, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9319697022438049, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9567090272903442, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9597564339637756, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9659252166748047, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.977672815322876, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9797810912132263, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9779419302940369, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9806693196296692, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9888927936553955, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.989818811416626, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936943650245667, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940344095230103, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950266480445862, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969090223312378, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.4.self_attn": [ + { + "accuracy": 0.9314757585525513, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9325720071792603, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9502220153808594, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9595361351966858, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9654639959335327, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.96713787317276, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9722180962562561, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9741367101669312, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.977820098400116, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9783415794372559, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9828382134437561, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9847133159637451, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.984296977519989, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9862185716629028, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9912256002426147, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9926730990409851, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922121167182922, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.995746374130249, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964984655380249, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.4.mlp": [ + { + "accuracy": 0.8992438316345215, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9017688035964966, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9177355766296387, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9224463701248169, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9497400522232056, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9534088373184204, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9602686166763306, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9740990996360779, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9765526652336121, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9744430184364319, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9776059985160828, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9871250987052917, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9883031845092773, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9927500486373901, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9931350350379944, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9942410588264465, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965601563453674, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.5.self_attn": [ + { + "accuracy": 0.9312288761138916, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9332429766654968, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9466700553894043, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9577977657318115, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9638015627861023, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9652858376502991, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9716836810112, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9738436937332153, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9774815440177917, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.97789067029953, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9820955991744995, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9838878512382507, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9839197993278503, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9857984781265259, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9909078478813171, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923155307769775, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9921332001686096, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.99552321434021, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9966526627540588, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.5.mlp": [ + { + "accuracy": 0.9235719442367554, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9276253581047058, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9318480491638184, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9333574175834656, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9689197540283203, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9733375310897827, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9763890504837036, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9778809547424316, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9822179675102234, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9878314137458801, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9887891411781311, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9934679865837097, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935567378997803, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956947565078735, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961302876472473, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9966464042663574, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9973414540290833, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.6.self_attn": [ + { + "accuracy": 0.9703402519226074, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9710289239883423, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9760013222694397, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9803446531295776, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9847868084907532, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9852306246757507, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9894976019859314, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9901916980743408, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9913412928581238, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916229248046875, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9925192594528198, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9932081699371338, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936240315437317, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9942994713783264, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964954853057861, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969252347946167, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972065091133118, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980089664459229, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987316131591797, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.6.mlp": [ + { + "accuracy": 0.9520189166069031, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.953398585319519, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9604594111442566, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9625735878944397, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9761260151863098, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9778745770454407, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9809656143188477, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9875475168228149, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9886812567710876, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9876111149787903, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9889976978302002, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936081767082214, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940297603607178, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960625767707825, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962632656097412, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9966981410980225, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975510835647583, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.7.self_attn": [ + { + "accuracy": 0.9654545187950134, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9665160179138184, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9731841683387756, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9787471294403076, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9829807877540588, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9831844568252563, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.990562379360199, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.991041898727417, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9917930960655212, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9921669960021973, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9916236400604248, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9922457337379456, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935548901557922, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940813779830933, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964919686317444, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.996796727180481, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9975430965423584, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9978916645050049, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987915754318237, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.7.mlp": [ + { + "accuracy": 0.9469283223152161, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.948310136795044, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9564090967178345, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9588730335235596, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9734205007553101, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9753491878509521, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9788756966590881, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9862276911735535, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9875152111053467, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9864411950111389, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9881131649017334, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9931445121765137, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9937621355056763, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960827827453613, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963220953941345, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9968886971473694, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980957508087158, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.8.self_attn": [ + { + "accuracy": 0.9616004228591919, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9633464813232422, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9695567488670349, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9759595990180969, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9811438322067261, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9813098907470703, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9898256063461304, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9901334643363953, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9909374117851257, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9913524389266968, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9906476140022278, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9913772344589233, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.992758572101593, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933127164840698, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960657358169556, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.996425449848175, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9973448514938354, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977043867111206, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9987177848815918, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.8.mlp": [ + { + "accuracy": 0.9434968829154968, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9449386596679688, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9536237716674805, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9562851786613464, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9716530442237854, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9737120270729065, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9775218367576599, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9853124022483826, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9866890907287598, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9855623245239258, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9873430728912354, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9927249550819397, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933896064758301, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958850145339966, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961060881614685, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9967235922813416, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.998026430606842, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.9.self_attn": [ + { + "accuracy": 0.9553995132446289, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9571937322616577, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9640243649482727, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9726303219795227, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9779067039489746, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9781882762908936, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9881770014762878, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9884291291236877, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9893790483474731, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9899032711982727, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9891637563705444, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.989920973777771, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9914891719818115, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.992179811000824, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9953780174255371, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957698583602905, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.996836245059967, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972187876701355, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983914494514465, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.9.mlp": [ + { + "accuracy": 0.9422038197517395, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9436619877815247, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.952694833278656, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9555050134658813, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.970986008644104, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9731055498123169, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9770951271057129, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9849473834037781, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9863730669021606, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9852272868156433, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9870408177375793, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.992548942565918, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993217945098877, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.99577796459198, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960180521011353, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9966662526130676, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979795813560486, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.10.self_attn": [ + { + "accuracy": 0.9493499398231506, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9503210186958313, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9570291638374329, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9641897678375244, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9751027226448059, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9755076766014099, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9866898059844971, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9869493842124939, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.988017737865448, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9887108206748962, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9876888394355774, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9885847568511963, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9899047613143921, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9906918406486511, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.994533896446228, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9949538111686707, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964499473571777, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9967092871665955, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982918500900269, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.10.mlp": [ + { + "accuracy": 0.9367719888687134, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9383903741836548, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9478542804718018, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9508419036865234, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.968203604221344, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9705210328102112, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9747180938720703, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9834887385368347, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9850257039070129, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9837997555732727, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.985788881778717, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9918237328529358, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9925611019134521, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9953529238700867, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956453442573547, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963309168815613, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.997803807258606, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.11.self_attn": [ + { + "accuracy": 0.9467344880104065, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9483067393302917, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.957198441028595, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9659615159034729, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9739027619361877, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9739984273910522, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9854788184165955, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.985880434513092, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9870201349258423, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9877007007598877, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9869197607040405, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9880322217941284, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9897122383117676, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9905611872673035, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9944283962249756, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9948906302452087, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962124228477478, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9966403245925903, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982412457466125, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.11.mlp": [ + { + "accuracy": 0.9329814910888672, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.934683084487915, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.944854199886322, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9480962157249451, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9662190079689026, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9686933159828186, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9732176065444946, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9824193716049194, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9840959906578064, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9827898144721985, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9849174618721008, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.991330623626709, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9921249747276306, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950955510139465, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.995372474193573, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961143732070923, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.997665524482727, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.12.self_attn": [ + { + "accuracy": 0.9424950480461121, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9441617727279663, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9532767534255981, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9631575345993042, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9715968370437622, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9720828533172607, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9845325946807861, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9849380850791931, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9861252307891846, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9868570566177368, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.985941469669342, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9868372082710266, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9886179566383362, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9895172119140625, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9938750267028809, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9943591356277466, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959490895271301, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963870048522949, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.998094916343689, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.12.mlp": [ + { + "accuracy": 0.928490161895752, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.930306077003479, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9413250088691711, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9448633193969727, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9639359712600708, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9665801525115967, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9715113639831543, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.981207013130188, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9829950332641602, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9816311001777649, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9838912487030029, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9907392263412476, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9915673136711121, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9947534203529358, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950428009033203, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958513975143433, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974758625030518, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.13.self_attn": [ + { + "accuracy": 0.939139723777771, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9412834048271179, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9504225850105286, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.961006760597229, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9697221517562866, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9703242778778076, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9834838509559631, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9840066432952881, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.985402524471283, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9861466884613037, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9849755764007568, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9860799908638, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9879854917526245, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9889242649078369, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9935319423675537, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940752387046814, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956828355789185, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961600303649902, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979859590530396, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.13.mlp": [ + { + "accuracy": 0.9238684177398682, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.925829291343689, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9373544454574585, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9411356449127197, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.961612343788147, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9643986821174622, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9696171879768372, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9799341559410095, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9818359613418579, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9804301261901855, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9828150272369385, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9900929927825928, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9909546971321106, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9943125247955322, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9946519732475281, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9954921007156372, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971616268157959, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.14.self_attn": [ + { + "accuracy": 0.934455156326294, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9367115497589111, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9466638565063477, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9578491449356079, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9673210978507996, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9675611257553101, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9822997450828552, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9826387166976929, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9842594861984253, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9850491285324097, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9836490750312805, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.984866738319397, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9869013428688049, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.98798006772995, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9929359555244446, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993538498878479, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9953737854957581, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958711266517639, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9977965354919434, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.14.mlp": [ + { + "accuracy": 0.9177756905555725, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.919827938079834, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9325910210609436, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9367133378982544, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9585274457931519, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9615681171417236, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9673069715499878, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9783830046653748, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9804465770721436, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9788675308227539, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9814786314964294, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9893354177474976, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9902874827384949, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9939250946044922, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9943070411682129, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.995246410369873, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970898032188416, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.15.self_attn": [ + { + "accuracy": 0.9359668493270874, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9374649524688721, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9472782015800476, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.958419680595398, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9680261015892029, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9685199856758118, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9826374053955078, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.983090877532959, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9844105839729309, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9852817058563232, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9840314984321594, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9851486682891846, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9870496988296509, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9880648255348206, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993003785610199, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936627149581909, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9953904747962952, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9959225654602051, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.997851550579071, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.15.mlp": [ + { + "accuracy": 0.9138731956481934, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9160382151603699, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9294994473457336, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9338262677192688, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9565568566322327, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9597150087356567, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9657794833183289, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9773882031440735, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9795449376106262, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9778644442558289, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9805852174758911, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9888207316398621, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9898406267166138, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993646502494812, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940595030784607, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950528740882874, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970109462738037, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.16.self_attn": [ + { + "accuracy": 0.936841607093811, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9385818243026733, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9482191801071167, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9590591788291931, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9683783650398254, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9687911868095398, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9830402135848999, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9835072159767151, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9847487211227417, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9855918884277344, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9842639565467834, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9854537844657898, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9874510765075684, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9884675145149231, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9932616949081421, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9939048290252686, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9955838322639465, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961603283882141, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9979615807533264, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.16.mlp": [ + { + "accuracy": 0.9097747206687927, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9119685888290405, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9261963963508606, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9307481050491333, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9544898271560669, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9578119516372681, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9642164707183838, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9763457179069519, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9786041378974915, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9768284559249878, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9796774983406067, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9883002042770386, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9893888235092163, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933673739433289, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9937951564788818, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9948485493659973, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969077110290527, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.17.self_attn": [ + { + "accuracy": 0.936918318271637, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9390618205070496, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9487131834030151, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9596605896949768, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9684903621673584, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9689411520957947, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9830051064491272, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9835382699966431, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.985001266002655, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9857423901557922, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9843840599060059, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9856324791908264, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9875731468200684, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9886074662208557, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933009743690491, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9939743280410767, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9956124424934387, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962557554244995, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9980000257492065, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.17.mlp": [ + { + "accuracy": 0.907934308052063, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9100937843322754, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9248316884040833, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9294832944869995, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9535717964172363, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9569474458694458, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9635227918624878, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9758857488632202, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9782149195671082, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9763720631599426, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9792813062667847, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9880854487419128, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9892072081565857, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9932934641838074, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936941862106323, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.994781494140625, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969048500061035, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.18.self_attn": [ + { + "accuracy": 0.9417886137962341, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9440310001373291, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9531508088111877, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.963138222694397, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9709886312484741, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9713147878646851, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9843199253082275, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9848012924194336, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9860446453094482, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9868277907371521, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9857326149940491, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9867305159568787, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9886807799339294, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9895712733268738, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993880033493042, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9944884181022644, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.995948076248169, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9965692758560181, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982197880744934, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.18.mlp": [ + { + "accuracy": 0.9072564840316772, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9094761610031128, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9241164922714233, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9287437200546265, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9532315731048584, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9566372632980347, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9632073640823364, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9757593274116516, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9780802130699158, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9762300252914429, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.979153573513031, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9880441427230835, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9892010688781738, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9933152198791504, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9936983585357666, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9947880506515503, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9969792366027832, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.19.self_attn": [ + { + "accuracy": 0.9419925212860107, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9431938529014587, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9526045918464661, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9615346789360046, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9709963798522949, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9713408946990967, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9842761754989624, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9847032427787781, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.986029326915741, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9867346286773682, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9856119751930237, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9867487549781799, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9887361526489258, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9896883368492126, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9938585758209229, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9944806694984436, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958698153495789, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.996433436870575, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.997995913028717, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.19.mlp": [ + { + "accuracy": 0.9091904759407043, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9113208651542664, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9255373477935791, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9300055503845215, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.954219400882721, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9575490951538086, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9639028906822205, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9762916564941406, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9785608053207397, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9767391085624695, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9796088337898254, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9883003830909729, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9894593954086304, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9934748411178589, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9938567280769348, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9949146509170532, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970992803573608, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.20.self_attn": [ + { + "accuracy": 0.9486165642738342, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.95090651512146, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9591884613037109, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.96675044298172, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9748371839523315, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9753091931343079, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9862366914749146, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9867591857910156, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.987819492816925, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9883294701576233, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9875011444091797, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.98850017786026, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9901918172836304, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9909708499908447, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9946736097335815, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951024651527405, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9963586330413818, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9967754483222961, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9981701970100403, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.20.mlp": [ + { + "accuracy": 0.9127780199050903, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9148420095443726, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9283568263053894, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9325607419013977, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.956044614315033, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9592171907424927, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9652310013771057, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9772341251373291, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9794236421585083, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9776419401168823, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9804047346115112, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9887553453445435, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9898816347122192, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993736982345581, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9940869808197021, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950855374336243, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971990585327148, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.21.self_attn": [ + { + "accuracy": 0.9539671540260315, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.955508291721344, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9637141823768616, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.970506489276886, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9774361252784729, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9775776267051697, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9877943992614746, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9881541728973389, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.989065408706665, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9896013140678406, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.988885223865509, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9897370934486389, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9913874864578247, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9920946359634399, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9953165054321289, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957361221313477, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9967994689941406, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972267746925354, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.998496949672699, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.21.mlp": [ + { + "accuracy": 0.9138404130935669, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9159210324287415, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9288933873176575, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9329104423522949, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9565706849098206, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9597253203392029, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.965484619140625, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9775335788726807, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.979685366153717, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9779269099235535, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9806569814682007, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9889184236526489, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9900363087654114, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9938618540763855, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9941734075546265, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951335191726685, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972708821296692, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.22.self_attn": [ + { + "accuracy": 0.9498860836029053, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9516811370849609, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9599934220314026, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9675408601760864, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.975214421749115, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.975623369216919, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9866546392440796, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.987027645111084, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.988040030002594, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9886026978492737, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9878209829330444, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9887345433235168, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9905011057853699, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9912907481193542, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9948492646217346, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9952542781829834, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964652061462402, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.996894359588623, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9983038902282715, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.22.mlp": [ + { + "accuracy": 0.9151159524917603, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9171680808067322, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9296190738677979, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9334802031517029, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9571576118469238, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9603219032287598, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9658223390579224, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9777891039848328, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9799360632896423, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.978218138217926, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9809140563011169, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9890093803405762, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9901112914085388, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9938163161277771, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9941934943199158, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9951046705245972, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971760511398315, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.23.self_attn": [ + { + "accuracy": 0.9577724933624268, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9591618180274963, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9664594531059265, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9726923108100891, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9794216156005859, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9796939492225647, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9888426065444946, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.989107072353363, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9900698065757751, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9905381202697754, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9898517727851868, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9906265735626221, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.992112398147583, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9927366971969604, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957330226898193, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960434436798096, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970316886901855, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9973990321159363, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9985262751579285, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.23.mlp": [ + { + "accuracy": 0.9153725504875183, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9174773097038269, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9294458627700806, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9331294298171997, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9573197364807129, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9605779051780701, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9658567309379578, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9778885245323181, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9800094366073608, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9783627986907959, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9810419678688049, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9890649914741516, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9900888204574585, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993817925453186, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9941460490226746, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9950054883956909, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.997000515460968, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.24.self_attn": [ + { + "accuracy": 0.9590924978256226, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9607964754104614, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9673714637756348, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9735159873962402, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9798654317855835, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9801721572875977, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9890934228897095, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9894006848335266, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9901682138442993, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9906661510467529, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9902133941650391, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9908740520477295, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9923003315925598, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9928999543190002, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9958410859107971, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9961578845977783, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.997101366519928, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9974644184112549, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9986010789871216, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.24.mlp": [ + { + "accuracy": 0.9115006327629089, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9144500494003296, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9261060357093811, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9296735525131226, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9563577175140381, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9590162634849548, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9641901850700378, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9768136143684387, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9784690737724304, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9775051474571228, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9802490472793579, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9885525703430176, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9893867373466492, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993202269077301, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993727445602417, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9945425987243652, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9964268803596497, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.25.self_attn": [ + { + "accuracy": 0.9577867984771729, + "total_bits": 86947200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9599529504776001, + "total_bits": 90224000, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9680209755897522, + "total_bits": 98709200, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9741918444633484, + "total_bits": 108832800, + "q_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9806426763534546, + "total_bits": 129080000, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9808921217918396, + "total_bits": 129196400, + "q_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64, + "3": 64 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9901913404464722, + "total_bits": 165124800, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9904928207397461, + "total_bits": 165446000, + "q_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.991197943687439, + "total_bits": 167052000, + "q_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 64 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9915329217910767, + "total_bits": 168979200, + "q_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.990847110748291, + "total_bits": 170156400, + "q_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9915829300880432, + "total_bits": 171148000, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9928789734840393, + "total_bits": 179721600, + "q_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 64 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 64, + "4": 64 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.993245005607605, + "total_bits": 182291200, + "q_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "5": 32 + }, + "bits": [ + 5 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9960154891014099, + "total_bits": 220011200, + "q_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9962084293365479, + "total_bits": 223251200, + "q_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9971372485160828, + "total_bits": 247044800, + "q_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9972817897796631, + "total_bits": 271379200, + "q_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 32 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "6": 32 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9982848167419434, + "total_bits": 328964800, + "q_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "k_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "v_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "o_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.layers.25.mlp": [ + { + "accuracy": 0.934510350227356, + "total_bits": 185754976, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9364456534385681, + "total_bits": 192390496, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "6": 32, + "3": 64, + "2": 64 + }, + "bits": [ + 6, + 3, + 2 + ], + "bits_prop": [ + 0.05, + 0.2, + 0.75 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9440828561782837, + "total_bits": 214392160, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "3": 32 + }, + "bits": [ + 5, + 3 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.94654381275177, + "total_bits": 240606560, + "gate_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "3": 64, + "2": 64 + }, + "bits": [ + 3, + 2 + ], + "bits_prop": [ + 0.3, + 0.7 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9676538705825806, + "total_bits": 272658944, + "gate_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 128, + "3": 128 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128, + "3": 128 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9699434041976929, + "total_bits": 278832160, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32, + "3": 32 + }, + "bits": [ + 8, + 4, + 3 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9734201431274414, + "total_bits": 299814944, + "gate_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32, + "3": 32 + }, + "bits": [ + 4, + 3 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9829437732696533, + "total_bits": 342838544, + "gate_proj": { + "group_size": { + "4": 128 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 128 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9844988584518433, + "total_bits": 347900960, + "gate_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "4": 32 + }, + "bits": [ + 4 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "4": 32 + }, + "bits": [ + 8, + 4 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9833990931510925, + "total_bits": 354169344, + "gate_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 128, + "4": 128 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 128, + "4": 128 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.985383152961731, + "total_bits": 360342560, + "gate_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "5": 32, + "4": 32 + }, + "bits": [ + 5, + 4 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "5": 32, + "4": 32 + }, + "bits": [ + 8, + 5, + 4 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9913974404335022, + "total_bits": 435679744, + "gate_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128, + "5": 128 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128, + "5": 128 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9920594692230225, + "total_bits": 441852960, + "gate_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 32, + "5": 32 + }, + "bits": [ + 6, + 5 + ], + "bits_prop": [ + 0.25, + 0.75 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 32, + "5": 32 + }, + "bits": [ + 8, + 6, + 5 + ], + "bits_prop": [ + 0.05, + 0.1, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9947201013565063, + "total_bits": 503263744, + "gate_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "6": 128 + }, + "bits": [ + 6 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 32, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.05, + 0.95 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9952026009559631, + "total_bits": 522550304, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.15, + 0.85 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9957330822944641, + "total_bits": 568835104, + "gate_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128, + "6": 128 + }, + "bits": [ + 8, + 6 + ], + "bits_prop": [ + 0.1, + 0.9 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + }, + { + "accuracy": 0.9970856308937073, + "total_bits": 666156064, + "gate_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "up_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + }, + "down_proj": { + "group_size": { + "8": 128 + }, + "bits": [ + 8 + ], + "bits_prop": [ + 1 + ], + "scale_bits": 4 + } + } + ], + "model.norm.norm": null, + "lm_head.linear": null + }, + "last_module_idx": 54 } \ No newline at end of file