Xenova HF staff commited on
Commit
c13c9af
1 Parent(s): bfab718

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 sha256:4c39e3a26f4628e6c67ef63363ce1d2e3a97ff1e6fe0ad6a479afb793792cfb6
3
- size 118422937
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6f9271b9f93784cab472008e8336d6811ccf117dc1477b88b7458d62b8adae5d
3
+ size 118423329
onnx/decoder_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:05baee847ffcf370e3ce86f3d4a919c2e4b6adc963036e0bbeec7d4ceb642559
3
- size 85988836
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c1a435f6cb49c1a41f0a2223fd0bb042f9695add26c8b9311c442124a0b3b1e
3
+ size 85989228
onnx/decoder_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f8341619817b5b6a9b6c718678229d55df6c42bf54d2539dc9cd0fe18f198352
3
- size 59404437
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:67c871b69cb6311626fc3bb188e7ba42185e898ac22f4183380b7a6895b742b3
3
+ size 59405700
onnx/decoder_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2d29d0731a7ff1da66f62ceefbb9b7f8aaa6ac688b061aeb8a13a63e2818febf
3
- size 30497398
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f36e7365307c5d68a703e97ab98a1b6639139081976794fad150b95d0ef3359
3
+ size 30495974
onnx/decoder_model_merged.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5001ab0c2c6378874025a66e8f0a9f3e8ebc9e70a496baf1eab80c171c479b4c
3
- size 118686508
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a6f57dadddd3dbff6d785fbd14a18cedc6ed38753638dc820e1eeb20cb672ba8
3
+ size 118664208
onnx/decoder_model_merged_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:bb8ae2aae30dcc511bcfba6989596dcab97969e60dbcd66df3df40c1d43e36a4
3
- size 86258087
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1134495cde770610187883ebb0e049c9c4467a5979f64e382ee64c432825ddf1
3
+ size 86234792
onnx/decoder_model_merged_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:cc516c6287324348c1579dabb375ef450e2ca11a8768050e3c4c8aa2116f006c
3
- size 59585567
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72b8537d2500bbee0ad8518deab0c3526680f734dea14d22e05100a1d85d7cc0
3
+ size 59565877
onnx/decoder_model_merged_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8a75d52e4b2d9c781f148d073f8b0af55d14c6bdd561be3ac5a352807592a42e
3
- size 110388428
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ef8d12d2e403fb06a77529c25d9882b3f383ec0607eda560f72b0a9dbf7997a0
3
+ size 110358643
onnx/decoder_model_merged_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b5cc7b352b3edcb8430b43d52cf886473648e86afae6fa7e256d68c88bd5cb37
3
- size 86847303
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f8941dbc39544b5930ca3c317fc072fe2947fe3fbc46fddaf741ecc2324799c3
3
+ size 86824080
onnx/decoder_model_merged_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8a75d52e4b2d9c781f148d073f8b0af55d14c6bdd561be3ac5a352807592a42e
3
- size 110388428
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ef8d12d2e403fb06a77529c25d9882b3f383ec0607eda560f72b0a9dbf7997a0
3
+ size 110358643
onnx/decoder_model_merged_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d33c9929fa0a0520cd7f05e9148354d914d940d1f05b2225bdf3651aa7d781ed
3
- size 110388458
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d6d76bf090cf1cff1ec019de26d9d30329288a613482341112aa7dfb2353fa1
3
+ size 110358664
onnx/decoder_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8adbf3db3d28be15d60d13368a70c2b3663b65bc04bffc83c94bd89cec8eab38
3
- size 86578412
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a1199c130ff940dfeb92fdbe75a503369fd31b7a3b591d8b0b860105628ca53f
3
+ size 86578804
onnx/decoder_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2d29d0731a7ff1da66f62ceefbb9b7f8aaa6ac688b061aeb8a13a63e2818febf
3
- size 30497398
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f36e7365307c5d68a703e97ab98a1b6639139081976794fad150b95d0ef3359
3
+ size 30495974
onnx/decoder_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f533aed3eb10fc3d58c06e0a5bc1d193b3b40f21ff1a80f43b563009c78a28c0
3
- size 30497428
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:34a121d1033d809f9c9c8cb8d5c7a7aac9d4ead8b8b687b34858d66a06a7962f
3
+ size 30496004
onnx/decoder_with_past_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6f9271b9f93784cab472008e8336d6811ccf117dc1477b88b7458d62b8adae5d
3
- size 118423329
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e7e3810428bc74904c71d509e908999d95defe3b22647ea3227acb8a9cbae27
3
+ size 113666902
onnx/decoder_with_past_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7c1a435f6cb49c1a41f0a2223fd0bb042f9695add26c8b9311c442124a0b3b1e
3
- size 85989228
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8dbc6334048f9d3e9c930871e213a4b8cd34c66cec0c9503602f2bad2f681b41
3
+ size 85286601
onnx/decoder_with_past_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:67c871b69cb6311626fc3bb188e7ba42185e898ac22f4183380b7a6895b742b3
3
- size 59405700
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:affbc6efb7b24bd2366dc885d022f36c5fec553197ac1f8575bc4afec2485579
3
+ size 57000402
onnx/decoder_with_past_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9f36e7365307c5d68a703e97ab98a1b6639139081976794fad150b95d0ef3359
3
- size 30495974
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8cde562b508354d55a3c98617aac323b2b765c250b15699e941c6b6773b4bf64
3
+ size 29258130
onnx/decoder_with_past_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a1199c130ff940dfeb92fdbe75a503369fd31b7a3b591d8b0b860105628ca53f
3
- size 86578804
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5facbc4ed7067a82e39d32d0c0408717e18a1fee8f24463533cd87545290d2d4
3
+ size 85802497
onnx/decoder_with_past_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9f36e7365307c5d68a703e97ab98a1b6639139081976794fad150b95d0ef3359
3
- size 30495974
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8cde562b508354d55a3c98617aac323b2b765c250b15699e941c6b6773b4bf64
3
+ size 29258130
onnx/decoder_with_past_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:34a121d1033d809f9c9c8cb8d5c7a7aac9d4ead8b8b687b34858d66a06a7962f
3
- size 30496004
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd1952a26882889273d2e339427b4c67bafd21625177962d393349de181530fe
3
+ size 29258153
onnx/encoder_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e2a1e49f1ace0db786d15aff2cc3de5991e9a6023198984881729fc684be0f43
3
- size 32894268
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8944ec04111e040305cdb07ac97f50363024255bf0bdc67ab93ecb76645f31ae
3
+ size 32894434
onnx/encoder_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:73828a3de37d6ba34b8e21ac01c111cf255b8df6b1c63ebd15aa27004da7cc70
3
- size 8567727
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a4b5391078304c8b62e3a73dfd8ac069511cbe0a8a30f2db4a06914a95e63a4
3
+ size 8567893
onnx/encoder_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:baa1dc4270baae150b5ce1703c9ac71ca631c77c8bfcbf424d2f153dc62d5446
3
- size 16477669
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:83fb3aa52145a42351b6b8cc3f520895fcfd297bf960a017fe9e91dee9484e9b
3
+ size 16477869
onnx/encoder_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0773d779bea267aea5c2bfdb14613e8343185b31ac31e1c5fd58b03122e49d87
3
- size 10083488
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dbcb11fb083f3794f9422311fe2be56c2d91792e79674e241185aad01f1c5389
3
+ size 10083654
onnx/encoder_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e4f954ad2f43b2ea7b30bb81d52589d3695c5383f23ab0ab05532b55e13ff9fe
3
- size 9009943
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e3e90a0256c385b03e6436290ffbf48eeee0515aaac9b00224efb97f3aded463
3
+ size 9010109
onnx/encoder_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1971b88bb1a9cea0e26e723503a23d672911a2204a891257c27f10ac163d2f8d
3
- size 10083501
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dbc40fd63079402b8e63aecd4a21bd5a52f00ff6524729bca022ea74961786aa
3
+ size 10083667
onnx/encoder_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1971b88bb1a9cea0e26e723503a23d672911a2204a891257c27f10ac163d2f8d
3
- size 10083501
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dbc40fd63079402b8e63aecd4a21bd5a52f00ff6524729bca022ea74961786aa
3
+ size 10083667
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",