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 sha256:39369dbf19a46c493928ce171c211b3272c32114107c2c76e7dc9adff138a965
|
| 3 |
+
size 614939754
|
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 sha256:635600c22498325cfc8d7fcb1ce38ef81e98815203b0fbcce2b4facc38a30d29
|
| 3 |
+
size 225674556
|
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 sha256:07c3aa1eda7954b9f78d11a3de54b04ac3809de6883f39fc8799b45d3baf614d
|
| 3 |
+
size 308030579
|
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:dc25f7af82662346025b6a1ffe3563c446d03c7a660114c95389f08876779ca6
|
| 3 |
+
size 156093710
|
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:cd6b3cd148c301475e8147f76d114154aec3a8b17a9d56e0e467a682f4f690a9
|
| 3 |
+
size 615653127
|
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:db44d5cffc48ce3808c17189114dbb28a45bd1b68877ba19855e296cf9b7d6df
|
| 3 |
+
size 226401993
|
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:ae064b9d757b3bd8471112265916f93bcaf36f86fc440537f00a9ca491dfa85a
|
| 3 |
+
size 308510968
|
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 sha256:70aa8c1075d211833a39bde34fee1261335c46a24f9a8f18849ce873e7bab4ca
|
| 3 |
+
size 316013302
|
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 sha256:59bbae503bc0623a328c6b25ec17c7cc28bd7df3629d4934062a52988149c8e1
|
| 3 |
+
size 233478153
|
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 sha256:70aa8c1075d211833a39bde34fee1261335c46a24f9a8f18849ce873e7bab4ca
|
| 3 |
+
size 316013302
|
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 sha256:8d812d40489c4f3fe61769f2b77ddfec2ec742802787aa87e1a50248f0c946a4
|
| 3 |
+
size 316013232
|
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 sha256:68287c54b4651f31ae1d81fbcfe373bb1a5b1d297499035fef08ae1598a7e0a2
|
| 3 |
+
size 232751580
|
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:dc25f7af82662346025b6a1ffe3563c446d03c7a660114c95389f08876779ca6
|
| 3 |
+
size 156093710
|
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:e417094883343a2b937a7b89cce5da9caf4e74271e3b731b5e9ed402bb604a49
|
| 3 |
+
size 156093766
|
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:a7ff4c9cc0d01eb0eb116f36b5feccd6cea81ae8832b0cfb75e6942538691666
|
| 3 |
+
size 558198950
|
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 sha256:9ada2ccad0dd9248e447208673a76500cf2b733e330bf47da52e9ff6c198b23d
|
| 3 |
+
size 217590512
|
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:d0448b0777e57069e6809567b0eaa069627424eb15cafff66905bd665fe95fef
|
| 3 |
+
size 279592353
|
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:d21775c5d49bbc52dfc31b63f8e35eb7b43556df98ce6c322e837153b4fbfb1e
|
| 3 |
+
size 141765339
|
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 sha256:6b416bdcb5d065e0cad1addbd162391b9e5a8400045586371e2079481aa50dd7
|
| 3 |
+
size 223782968
|
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:d21775c5d49bbc52dfc31b63f8e35eb7b43556df98ce6c322e837153b4fbfb1e
|
| 3 |
+
size 141765339
|
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 sha256:1b53717c64712c0c0d064ec18d310432cad7a30b05edbba246ee6de8817d90ea
|
| 3 |
+
size 141765385
|
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:d36ca46845a050a1bef76c282a79267916f0b5d4a37df6bac341f26bb36fb8eb
|
| 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 sha256:39f173ebd2209a585542d93e1df9389e5ace56a72968d8b3e134270923e511c6
|
| 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 sha256:2a5003d27c94278f8871a20e94b0aa9fcb2ebd05eae77a5e9890ca8c81e2eeed
|
| 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:d6f505ed2363b9da16b8964cb726b7abfa4779b6671c47c922a1a945012047a5
|
| 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:bed4f3072a57015ee46254a8f34b64c3d3b3b4248974ad02a5a4189dc0449e0b
|
| 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:ddb820fc7fa12f7de8037e8c5f1afdf291ba9f788dc37497cc1da61f94b453cb
|
| 3 |
+
size 92201982
|
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:ddb820fc7fa12f7de8037e8c5f1afdf291ba9f788dc37497cc1da61f94b453cb
|
| 3 |
+
size 92201982
|
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",
|