Xenova HF staff commited on
Commit
0503028
1 Parent(s): 7128154

Upload folder using huggingface_hub

Browse files
onnx/decoder_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:df437feb64deaca88d6a70d890a9c84e8545f6137a38f013397786c545d45317
3
- size 348748482
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd0493374cd0748e5181d6c68b8248f440fd6c14bed742fd9bed7414830cd174
3
+ size 348748678
onnx/decoder_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:af348bcedcc1184fc65ab61b30e57c9753afea8ed3e185b6cc3708fcce8bcf42
3
- size 233408272
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2da75ef1cb876d8c49c02d696b9858354fd68dbf46ad725ebe8ee5688351e716
3
+ size 233408468
onnx/decoder_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7199796c9f585abbd0cb67c5ea367f02f77a8a35a7f7a8ae5b821135a5916ad5
3
- size 174476542
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9009096e8dcbb66340d6a5010c5e39b6fa68a146b13fd74ebb8a89f68905326a
3
+ size 174477159
onnx/decoder_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6d32b189956b4d759a425dede94f59b52b86cfd83d3b561206758cf577c0ffdd
3
- size 88829125
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e650ed661602cff1e06a9ab94a212399983f6c9ac3fac11cc4afe924b5d6e12
3
+ size 88828421
onnx/decoder_model_merged.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a35a3df6bea2abc8854858676a544387c53a9a253ebb2a8715f83ceaea298733
3
- size 348886174
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b53ee5a5ccf91dad271e6de1aa38189b2326d67fd1fdedcb9a391cd51333e0e
3
+ size 348872491
onnx/decoder_model_merged_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b54b4f7382d116f33fc25bbe4363d3d2fb50c645f5f43b22e7d659c0489a6c49
3
- size 233548804
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d981bfb3710d94c11749ff3cb6a1762be393e9021d01b0c77d312376fdeaac9
3
+ size 233534625
onnx/decoder_model_merged_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:74cd046c0708061f8f7662c12ddf79c15f4d805900c65a63dec3fc5b9dfb4d53
3
- size 174570531
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aa6d2c9f9ec665b90cf447e92652e8626f8622d9593834e52cadc135c99e99dd
3
+ size 174558131
onnx/decoder_model_merged_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ece7dd62879925f0137954f1020d584160730d4f9c57772f6015b1e5f1ade04d
3
- size 301381015
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0ff9cc1f5ec7150f5085aa15044d923fbdf392c92f86916a87d005505283682f
3
+ size 301363513
onnx/decoder_model_merged_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9648ee657db1abb96f075ea192f57c8dd038bf4e2b9c8adc5cf2d839a4757975
3
- size 235645614
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:09985c682350705128c03670eb194050df8451c51ef8c5fbb72ae76e3e0cb4d9
3
+ size 235631471
onnx/decoder_model_merged_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ece7dd62879925f0137954f1020d584160730d4f9c57772f6015b1e5f1ade04d
3
- size 301381015
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0ff9cc1f5ec7150f5085aa15044d923fbdf392c92f86916a87d005505283682f
3
+ size 301363513
onnx/decoder_model_merged_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:cba8272f8cddda92e649b25fb428c90e5247618884c8b1a78b3316e4bddb4f86
3
- size 301381026
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5d0a1c88d4326f0075ea7866eeb1a7d48ac3489d28b2253eb97406482209e604
3
+ size 301363506
onnx/decoder_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b0ebf0ca3a2fc01652c9d8e92fce2b6df3cb254781ee203020c7126f5b021451
3
- size 235505262
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:927ed02d5e97a0a8b1dbe5afcf24e5134aaf32f18d3770d6b04ff996d621e0e6
3
+ size 235505458
onnx/decoder_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6d32b189956b4d759a425dede94f59b52b86cfd83d3b561206758cf577c0ffdd
3
- size 88829125
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e650ed661602cff1e06a9ab94a212399983f6c9ac3fac11cc4afe924b5d6e12
3
+ size 88828421
onnx/decoder_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c42fa83eea1cfc58c7bd5b9a2bc07bab56e7592eaeb7290f95d1ac59614c8aef
3
- size 88829136
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96cfe88ac99a7e6d14ccc4638c1a805a272d6e0e2eb5a08005e6e3b72e0c13ea
3
+ size 88828432
onnx/decoder_with_past_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fd0493374cd0748e5181d6c68b8248f440fd6c14bed742fd9bed7414830cd174
3
- size 348748678
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:367bbf8a43da1bca0969c2eaa23f013137e9ad8134cb898d5c60f027decabc21
3
+ size 331943836
onnx/decoder_with_past_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2da75ef1cb876d8c49c02d696b9858354fd68dbf46ad725ebe8ee5688351e716
3
- size 233408468
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e1180c1dfd37842406c8f373f27cd11fb908ff8ab16d9c8dafb80c91db83ed0
3
+ size 231020922
onnx/decoder_with_past_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9009096e8dcbb66340d6a5010c5e39b6fa68a146b13fd74ebb8a89f68905326a
3
- size 174477159
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3bc5d3dc9a46f16f6a3ee18cac9905bd1ce2cf9935b3b50f6f1d222428c22580
3
+ size 166057794
onnx/decoder_with_past_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2e650ed661602cff1e06a9ab94a212399983f6c9ac3fac11cc4afe924b5d6e12
3
- size 88828421
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e4e92ad7b3cf6f0c64810bc35cf52f02729b6a77a65ddcc649579442162989f3
3
+ size 84594714
onnx/decoder_with_past_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:927ed02d5e97a0a8b1dbe5afcf24e5134aaf32f18d3770d6b04ff996d621e0e6
3
- size 235505458
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:665f17ccd2595943a84f3fd94f0c575539545cd785d659c10f0ec22504777a4b
3
+ size 232855800
onnx/decoder_with_past_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2e650ed661602cff1e06a9ab94a212399983f6c9ac3fac11cc4afe924b5d6e12
3
- size 88828421
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e4e92ad7b3cf6f0c64810bc35cf52f02729b6a77a65ddcc649579442162989f3
3
+ size 84594714
onnx/decoder_with_past_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:96cfe88ac99a7e6d14ccc4638c1a805a272d6e0e2eb5a08005e6e3b72e0c13ea
3
- size 88828432
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d53342c6d9e8adb6bdc786b80e5fa37872baf31c6c33ce6ad14c9e93b912574
3
+ size 84594723
onnx/encoder_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:bad13f6f0bacac78bf39dc7e76df368938d928b943cea55b6874af124472786e
3
- size 1229214768
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87ce60aae236f0306d5689f0a26d3cfb4ed93548716ce1d9ecca5d9dbfbe9ead
3
+ size 1229214934
onnx/encoder_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:17806e3ab1efbe16e2fd174e64a8e9e8cb1aa49516565211bc2bae04a9105a0f
3
- size 191147202
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:affe1219278db31e47e26835c872bdfdaa35db62357b7632639a98fff11d7849
3
+ size 191147368
onnx/encoder_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:cd7f947718da39385829305d4d4c3b5dce9a3b354e998c331cee70f2a896fca3
3
- size 614781819
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:778d55314e55fe9b67b2c7971cd03c5fcfe08a795cf63323de7cb994a28e2c5c
3
+ size 614782019
onnx/encoder_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a28005007015dcbf4df21b1bc0b23d8a3878b46538a9bc9f7ea206ec67bdabb2
3
- size 313220312
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8948bf97b094f560bb843294a0aeb9e65588f02d88febc41ef7c589a4881d281
3
+ size 313220478
onnx/encoder_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fb7cbcafb8d34c8476cd6eab1c1c82e20e640ff436a4154de06911bb9fe2d29b
3
- size 210020394
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be332e57b058bf3edebaaa135e97511a0ac0ead2d90a960755357e588c0c4411
3
+ size 210020560
onnx/encoder_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2392aa94deb47dc7788137c93cc55a54cdc1325feb4f428a59b53bb7250a771e
3
- size 313220381
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3a0619e8051e8dab1e65e2b653617bacb003dbfc45130f03f0c8ebce2d3d565
3
+ size 313220547
onnx/encoder_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2392aa94deb47dc7788137c93cc55a54cdc1325feb4f428a59b53bb7250a771e
3
- size 313220381
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3a0619e8051e8dab1e65e2b653617bacb003dbfc45130f03f0c8ebce2d3d565
3
+ size 313220547
quantize_config.json CHANGED
@@ -20,38 +20,30 @@
20
  ],
21
  "weight_type": "QUInt8"
22
  },
23
- "decoder_model_merged": {
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
- "decoder_with_past_model": {
55
  "op_types": [
56
  "Add",
57
  "Cast",
@@ -82,10 +74,9 @@
82
  ],
83
  "weight_type": "QInt8"
84
  },
85
- "decoder_model": {
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
- "decoder_model_merged": {
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
- "decoder_with_past_model": {
172
  "op_types": [
173
  "Add",
174
  "Cast",
@@ -199,10 +183,9 @@
199
  ],
200
  "weight_type": "QInt8"
201
  },
202
- "decoder_model": {
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
- "decoder_model_merged": {
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
- "decoder_with_past_model": {
289
  "op_types": [
290
  "Add",
291
  "Cast",
@@ -316,10 +292,9 @@
316
  ],
317
  "weight_type": "QUInt8"
318
  },
319
- "decoder_model": {
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",