Upload folder using huggingface_hub
Browse files- onnx/decoder_model.onnx +2 -2
- onnx/decoder_model_bnb4.onnx +2 -2
- onnx/decoder_model_fp16.onnx +2 -2
- onnx/decoder_model_int8.onnx +2 -2
- onnx/decoder_model_merged.onnx +2 -2
- onnx/decoder_model_merged_bnb4.onnx +2 -2
- onnx/decoder_model_merged_fp16.onnx +2 -2
- onnx/decoder_model_merged_int8.onnx +2 -2
- onnx/decoder_model_merged_q4.onnx +2 -2
- onnx/decoder_model_merged_quantized.onnx +2 -2
- onnx/decoder_model_merged_uint8.onnx +2 -2
- onnx/decoder_model_q4.onnx +2 -2
- onnx/decoder_model_quantized.onnx +2 -2
- onnx/decoder_model_uint8.onnx +2 -2
- onnx/decoder_with_past_model.onnx +2 -2
- onnx/decoder_with_past_model_bnb4.onnx +2 -2
- onnx/decoder_with_past_model_fp16.onnx +2 -2
- onnx/decoder_with_past_model_int8.onnx +2 -2
- onnx/decoder_with_past_model_q4.onnx +2 -2
- onnx/decoder_with_past_model_quantized.onnx +2 -2
- onnx/decoder_with_past_model_uint8.onnx +2 -2
- onnx/encoder_model.onnx +2 -2
- onnx/encoder_model_bnb4.onnx +2 -2
- onnx/encoder_model_fp16.onnx +2 -2
- onnx/encoder_model_int8.onnx +2 -2
- onnx/encoder_model_q4.onnx +2 -2
- onnx/encoder_model_quantized.onnx +2 -2
- onnx/encoder_model_uint8.onnx +2 -2
- quantize_config.json +15 -39
onnx/decoder_model.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:07472ac2d328746ad674fedd553d4c6e730736ae29922b48d7645645d8faaa28
|
| 3 |
+
size 614942826
|
onnx/decoder_model_bnb4.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d8bd696de42e2ca45cda04c06ba8969d6d45d7c9ffcb279607e6d05db018c7c1
|
| 3 |
+
size 225677628
|
onnx/decoder_model_fp16.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:688a399c3219c9f44df05704547c1d2db98c0530c6bb03517269bb8010f6f0da
|
| 3 |
+
size 308032115
|
onnx/decoder_model_int8.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a70dbe29053c190cf026eb8b642c3cd809d2ef4f41e3ab7706fbaeeabf722820
|
| 3 |
+
size 156094479
|
onnx/decoder_model_merged.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fc28024e3729fdfc7a7b0049d1625eb19fc196ef084a5f4752662f900045966f
|
| 3 |
+
size 615656199
|
onnx/decoder_model_merged_bnb4.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:54fa831708d468820b9ad9fcae0f05846a0c028ac3193fff0bb00ddd32a00396
|
| 3 |
+
size 226405065
|
onnx/decoder_model_merged_fp16.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2760285aeee07df1527d76272d1a56c4e4fe2630b65577f9a07314ee9c6813ab
|
| 3 |
+
size 308512504
|
onnx/decoder_model_merged_int8.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5771fb241c8e2cd947dcbf5883f2478ec06a56154a121293b2539e30089e25c1
|
| 3 |
+
size 316017116
|
onnx/decoder_model_merged_q4.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1934effac669aef0a035b1bf8fe1e5fc543251766a33137b05021240f4028cc0
|
| 3 |
+
size 233481225
|
onnx/decoder_model_merged_quantized.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5771fb241c8e2cd947dcbf5883f2478ec06a56154a121293b2539e30089e25c1
|
| 3 |
+
size 316017116
|
onnx/decoder_model_merged_uint8.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6123b85661fff1d1f192a022d8df8357d43561667deeb5e2c4501856ececa9df
|
| 3 |
+
size 316017078
|
onnx/decoder_model_q4.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2e6da8822582177f263caec2255fc95db40cf3690fe4251cf337d34fc2ed6dd6
|
| 3 |
+
size 232754652
|
onnx/decoder_model_quantized.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a70dbe29053c190cf026eb8b642c3cd809d2ef4f41e3ab7706fbaeeabf722820
|
| 3 |
+
size 156094479
|
onnx/decoder_model_uint8.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:faaf1269df3de051d5de4f7a9bd9a782bf26b4875618def667c3d28fd2998697
|
| 3 |
+
size 156094540
|
onnx/decoder_with_past_model.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e06355092467d1960517c176e17aaebc608289c36a4049a77b38336308397572
|
| 3 |
+
size 558202022
|
onnx/decoder_with_past_model_bnb4.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:79570130f348928028d7cb17aa4bf2566e29a3a9e20e1548d2a09f8aa2dc4d05
|
| 3 |
+
size 217593584
|
onnx/decoder_with_past_model_fp16.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3958c6a90f7d96563aeef1fc1870de329965f7645cd4b4ef8aec5624346f6e50
|
| 3 |
+
size 279593889
|
onnx/decoder_with_past_model_int8.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e9d4e1973cd015cb0afecb783d383ce320ed9d12546ab3a7b40c375e72296b8b
|
| 3 |
+
size 141766108
|
onnx/decoder_with_past_model_q4.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5017a6bca7d1d7f3e734007551cf34d921cecf6c87d5fb49958eecc6811f5a75
|
| 3 |
+
size 223786040
|
onnx/decoder_with_past_model_quantized.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e9d4e1973cd015cb0afecb783d383ce320ed9d12546ab3a7b40c375e72296b8b
|
| 3 |
+
size 141766108
|
onnx/decoder_with_past_model_uint8.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:610839f10e574a2fa235c02893d193ff683d4f3b2702e87decadc97479417dc7
|
| 3 |
+
size 141766156
|
onnx/encoder_model.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c0908c1ef2326f5487d1f77e56c6925f910ae56c87fa269ed885682ac851e588
|
| 3 |
+
size 352791798
|
onnx/encoder_model_bnb4.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:74fce21b8ad3e1ebecb2c18e949f0f647263dc8164d495049c8abbc6e49352f9
|
| 3 |
+
size 60840144
|
onnx/encoder_model_fp16.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8b05b14e66bb76dfe4bacfc954d985f55cb8011662539899360c57d55932ebb5
|
| 3 |
+
size 176483570
|
onnx/encoder_model_int8.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e98bd3f4470ac909a4f052baa95de56345f36c7df284e958a1cae3136142a9db
|
| 3 |
+
size 92201939
|
onnx/encoder_model_q4.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ca337b9031573c999b4710c945704d0e93f6caea8d239a44fd134f32412c66fe
|
| 3 |
+
size 66148032
|
onnx/encoder_model_quantized.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e77c1c1bd471e63665783145e94019a94c2d22b9e38acf8ef7f74a1f82131cdc
|
| 3 |
+
size 92201972
|
onnx/encoder_model_uint8.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e77c1c1bd471e63665783145e94019a94c2d22b9e38acf8ef7f74a1f82131cdc
|
| 3 |
+
size 92201972
|
quantize_config.json
CHANGED
|
@@ -20,38 +20,30 @@
|
|
| 20 |
],
|
| 21 |
"weight_type": "QUInt8"
|
| 22 |
},
|
| 23 |
-
"
|
| 24 |
"op_types": [
|
| 25 |
"Add",
|
| 26 |
"Concat",
|
| 27 |
"Constant",
|
| 28 |
-
"ConstantOfShape",
|
| 29 |
"Div",
|
| 30 |
-
"Equal",
|
| 31 |
"Erf",
|
| 32 |
-
"Expand",
|
| 33 |
"Gather",
|
| 34 |
-
"If",
|
| 35 |
-
"Less",
|
| 36 |
"MatMul",
|
| 37 |
"Mul",
|
| 38 |
"Pow",
|
| 39 |
-
"Range",
|
| 40 |
"ReduceMean",
|
| 41 |
"Reshape",
|
| 42 |
"Shape",
|
| 43 |
"Slice",
|
| 44 |
"Softmax",
|
| 45 |
"Sqrt",
|
| 46 |
-
"Squeeze",
|
| 47 |
"Sub",
|
| 48 |
"Transpose",
|
| 49 |
-
"Unsqueeze"
|
| 50 |
-
"Where"
|
| 51 |
],
|
| 52 |
"weight_type": "QInt8"
|
| 53 |
},
|
| 54 |
-
"
|
| 55 |
"op_types": [
|
| 56 |
"Add",
|
| 57 |
"Cast",
|
|
@@ -82,10 +74,9 @@
|
|
| 82 |
],
|
| 83 |
"weight_type": "QInt8"
|
| 84 |
},
|
| 85 |
-
"
|
| 86 |
"op_types": [
|
| 87 |
"Add",
|
| 88 |
-
"Cast",
|
| 89 |
"Concat",
|
| 90 |
"Constant",
|
| 91 |
"ConstantOfShape",
|
|
@@ -94,6 +85,7 @@
|
|
| 94 |
"Erf",
|
| 95 |
"Expand",
|
| 96 |
"Gather",
|
|
|
|
| 97 |
"Less",
|
| 98 |
"MatMul",
|
| 99 |
"Mul",
|
|
@@ -137,38 +129,30 @@
|
|
| 137 |
],
|
| 138 |
"weight_type": "QInt8"
|
| 139 |
},
|
| 140 |
-
"
|
| 141 |
"op_types": [
|
| 142 |
"Add",
|
| 143 |
"Concat",
|
| 144 |
"Constant",
|
| 145 |
-
"ConstantOfShape",
|
| 146 |
"Div",
|
| 147 |
-
"Equal",
|
| 148 |
"Erf",
|
| 149 |
-
"Expand",
|
| 150 |
"Gather",
|
| 151 |
-
"If",
|
| 152 |
-
"Less",
|
| 153 |
"MatMul",
|
| 154 |
"Mul",
|
| 155 |
"Pow",
|
| 156 |
-
"Range",
|
| 157 |
"ReduceMean",
|
| 158 |
"Reshape",
|
| 159 |
"Shape",
|
| 160 |
"Slice",
|
| 161 |
"Softmax",
|
| 162 |
"Sqrt",
|
| 163 |
-
"Squeeze",
|
| 164 |
"Sub",
|
| 165 |
"Transpose",
|
| 166 |
-
"Unsqueeze"
|
| 167 |
-
"Where"
|
| 168 |
],
|
| 169 |
"weight_type": "QInt8"
|
| 170 |
},
|
| 171 |
-
"
|
| 172 |
"op_types": [
|
| 173 |
"Add",
|
| 174 |
"Cast",
|
|
@@ -199,10 +183,9 @@
|
|
| 199 |
],
|
| 200 |
"weight_type": "QInt8"
|
| 201 |
},
|
| 202 |
-
"
|
| 203 |
"op_types": [
|
| 204 |
"Add",
|
| 205 |
-
"Cast",
|
| 206 |
"Concat",
|
| 207 |
"Constant",
|
| 208 |
"ConstantOfShape",
|
|
@@ -211,6 +194,7 @@
|
|
| 211 |
"Erf",
|
| 212 |
"Expand",
|
| 213 |
"Gather",
|
|
|
|
| 214 |
"Less",
|
| 215 |
"MatMul",
|
| 216 |
"Mul",
|
|
@@ -254,38 +238,30 @@
|
|
| 254 |
],
|
| 255 |
"weight_type": "QUInt8"
|
| 256 |
},
|
| 257 |
-
"
|
| 258 |
"op_types": [
|
| 259 |
"Add",
|
| 260 |
"Concat",
|
| 261 |
"Constant",
|
| 262 |
-
"ConstantOfShape",
|
| 263 |
"Div",
|
| 264 |
-
"Equal",
|
| 265 |
"Erf",
|
| 266 |
-
"Expand",
|
| 267 |
"Gather",
|
| 268 |
-
"If",
|
| 269 |
-
"Less",
|
| 270 |
"MatMul",
|
| 271 |
"Mul",
|
| 272 |
"Pow",
|
| 273 |
-
"Range",
|
| 274 |
"ReduceMean",
|
| 275 |
"Reshape",
|
| 276 |
"Shape",
|
| 277 |
"Slice",
|
| 278 |
"Softmax",
|
| 279 |
"Sqrt",
|
| 280 |
-
"Squeeze",
|
| 281 |
"Sub",
|
| 282 |
"Transpose",
|
| 283 |
-
"Unsqueeze"
|
| 284 |
-
"Where"
|
| 285 |
],
|
| 286 |
"weight_type": "QUInt8"
|
| 287 |
},
|
| 288 |
-
"
|
| 289 |
"op_types": [
|
| 290 |
"Add",
|
| 291 |
"Cast",
|
|
@@ -316,10 +292,9 @@
|
|
| 316 |
],
|
| 317 |
"weight_type": "QUInt8"
|
| 318 |
},
|
| 319 |
-
"
|
| 320 |
"op_types": [
|
| 321 |
"Add",
|
| 322 |
-
"Cast",
|
| 323 |
"Concat",
|
| 324 |
"Constant",
|
| 325 |
"ConstantOfShape",
|
|
@@ -328,6 +303,7 @@
|
|
| 328 |
"Erf",
|
| 329 |
"Expand",
|
| 330 |
"Gather",
|
|
|
|
| 331 |
"Less",
|
| 332 |
"MatMul",
|
| 333 |
"Mul",
|
|
|
|
| 20 |
],
|
| 21 |
"weight_type": "QUInt8"
|
| 22 |
},
|
| 23 |
+
"decoder_with_past_model": {
|
| 24 |
"op_types": [
|
| 25 |
"Add",
|
| 26 |
"Concat",
|
| 27 |
"Constant",
|
|
|
|
| 28 |
"Div",
|
|
|
|
| 29 |
"Erf",
|
|
|
|
| 30 |
"Gather",
|
|
|
|
|
|
|
| 31 |
"MatMul",
|
| 32 |
"Mul",
|
| 33 |
"Pow",
|
|
|
|
| 34 |
"ReduceMean",
|
| 35 |
"Reshape",
|
| 36 |
"Shape",
|
| 37 |
"Slice",
|
| 38 |
"Softmax",
|
| 39 |
"Sqrt",
|
|
|
|
| 40 |
"Sub",
|
| 41 |
"Transpose",
|
| 42 |
+
"Unsqueeze"
|
|
|
|
| 43 |
],
|
| 44 |
"weight_type": "QInt8"
|
| 45 |
},
|
| 46 |
+
"decoder_model": {
|
| 47 |
"op_types": [
|
| 48 |
"Add",
|
| 49 |
"Cast",
|
|
|
|
| 74 |
],
|
| 75 |
"weight_type": "QInt8"
|
| 76 |
},
|
| 77 |
+
"decoder_model_merged": {
|
| 78 |
"op_types": [
|
| 79 |
"Add",
|
|
|
|
| 80 |
"Concat",
|
| 81 |
"Constant",
|
| 82 |
"ConstantOfShape",
|
|
|
|
| 85 |
"Erf",
|
| 86 |
"Expand",
|
| 87 |
"Gather",
|
| 88 |
+
"If",
|
| 89 |
"Less",
|
| 90 |
"MatMul",
|
| 91 |
"Mul",
|
|
|
|
| 129 |
],
|
| 130 |
"weight_type": "QInt8"
|
| 131 |
},
|
| 132 |
+
"decoder_with_past_model": {
|
| 133 |
"op_types": [
|
| 134 |
"Add",
|
| 135 |
"Concat",
|
| 136 |
"Constant",
|
|
|
|
| 137 |
"Div",
|
|
|
|
| 138 |
"Erf",
|
|
|
|
| 139 |
"Gather",
|
|
|
|
|
|
|
| 140 |
"MatMul",
|
| 141 |
"Mul",
|
| 142 |
"Pow",
|
|
|
|
| 143 |
"ReduceMean",
|
| 144 |
"Reshape",
|
| 145 |
"Shape",
|
| 146 |
"Slice",
|
| 147 |
"Softmax",
|
| 148 |
"Sqrt",
|
|
|
|
| 149 |
"Sub",
|
| 150 |
"Transpose",
|
| 151 |
+
"Unsqueeze"
|
|
|
|
| 152 |
],
|
| 153 |
"weight_type": "QInt8"
|
| 154 |
},
|
| 155 |
+
"decoder_model": {
|
| 156 |
"op_types": [
|
| 157 |
"Add",
|
| 158 |
"Cast",
|
|
|
|
| 183 |
],
|
| 184 |
"weight_type": "QInt8"
|
| 185 |
},
|
| 186 |
+
"decoder_model_merged": {
|
| 187 |
"op_types": [
|
| 188 |
"Add",
|
|
|
|
| 189 |
"Concat",
|
| 190 |
"Constant",
|
| 191 |
"ConstantOfShape",
|
|
|
|
| 194 |
"Erf",
|
| 195 |
"Expand",
|
| 196 |
"Gather",
|
| 197 |
+
"If",
|
| 198 |
"Less",
|
| 199 |
"MatMul",
|
| 200 |
"Mul",
|
|
|
|
| 238 |
],
|
| 239 |
"weight_type": "QUInt8"
|
| 240 |
},
|
| 241 |
+
"decoder_with_past_model": {
|
| 242 |
"op_types": [
|
| 243 |
"Add",
|
| 244 |
"Concat",
|
| 245 |
"Constant",
|
|
|
|
| 246 |
"Div",
|
|
|
|
| 247 |
"Erf",
|
|
|
|
| 248 |
"Gather",
|
|
|
|
|
|
|
| 249 |
"MatMul",
|
| 250 |
"Mul",
|
| 251 |
"Pow",
|
|
|
|
| 252 |
"ReduceMean",
|
| 253 |
"Reshape",
|
| 254 |
"Shape",
|
| 255 |
"Slice",
|
| 256 |
"Softmax",
|
| 257 |
"Sqrt",
|
|
|
|
| 258 |
"Sub",
|
| 259 |
"Transpose",
|
| 260 |
+
"Unsqueeze"
|
|
|
|
| 261 |
],
|
| 262 |
"weight_type": "QUInt8"
|
| 263 |
},
|
| 264 |
+
"decoder_model": {
|
| 265 |
"op_types": [
|
| 266 |
"Add",
|
| 267 |
"Cast",
|
|
|
|
| 292 |
],
|
| 293 |
"weight_type": "QUInt8"
|
| 294 |
},
|
| 295 |
+
"decoder_model_merged": {
|
| 296 |
"op_types": [
|
| 297 |
"Add",
|
|
|
|
| 298 |
"Concat",
|
| 299 |
"Constant",
|
| 300 |
"ConstantOfShape",
|
|
|
|
| 303 |
"Erf",
|
| 304 |
"Expand",
|
| 305 |
"Gather",
|
| 306 |
+
"If",
|
| 307 |
"Less",
|
| 308 |
"MatMul",
|
| 309 |
"Mul",
|