Xenova HF staff commited on
Commit
02800e6
1 Parent(s): 62a58be

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:a2ec9a9fa75e8924f2e92c37be4307063e8d68f2494ef2a576b4a98a94325b31
3
- size 312126454
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c8a78cc1a8b4fe1a67c02274c91f9a9470e540698dc7c5b9eb6c31c66814761
3
+ size 312126846
onnx/decoder_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ebf17d8218892fd85a467a76379147041af38f41fe43bab042e1aa92996d992c
3
- size 182371400
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b7e59527529f05a531a5d4531b01d5e275d95fda9aac73b6c397f5f6184cfaea
3
+ size 182371792
onnx/decoder_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f6391e28038a41712b32543201bce7356723e7c68f7f3b32fcbaf73ead4a8e1e
3
- size 156256207
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5c46e9e1f70736f61ff775f3681ef22ceb737bf90f5b5aff07d48afe9a5d8f9
3
+ size 156257470
onnx/decoder_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3a4b1a01cb5ee31bc0201ae88ce279f5bec7c663635f251cfc5d750414a904a9
3
- size 79520048
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b097c6a8ccd7133991491e78712ee14b66127b1aa2de4388854e3d0b5cfc2e12
3
+ size 79518624
onnx/decoder_model_merged.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2482f2e3737d869429a4c1cfa7cffbfcaa392b1aae472a84e75e03c9874fab89
3
- size 312390039
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:369432673394a080eca06509bd333179f94f81494b8a1781074362691ba14b8f
3
+ size 312367757
onnx/decoder_model_merged_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d9e4318e70236f60fb5d6d6c56c71660cf3575476389a5f998d004b8a5b1e0eb
3
- size 182640665
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9acc8cfe9b05e2f099d3f6f52856e5beeaec98e928f04d5d73c6f197795069ce
3
+ size 182617391
onnx/decoder_model_merged_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:661e4feeb7dfe22b5ce653fa87863470631ae8c330f565f991924aa16793044c
3
- size 156437337
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb9e0dd95bc1b9a63650ebb3af81d178254f7e9d3ba78610acf52a29617525f7
3
+ size 156417665
onnx/decoder_model_merged_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:128b382b33be99be84195be4b2e7cc78739f1219168dd2eb6d16a8b8bd329fea
3
- size 239072646
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:35cb2e3f953dec8fdbec8fc15fc42a71ef45392668d7e7250cd05bbf2eed5f86
3
+ size 239042891
onnx/decoder_model_merged_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:02d9dc39c5c52348ac6a80c21945833c33ef88d7a71e46b1e4728c56c53349c1
3
- size 184999313
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f4e255b202f65ce2589c27af3b67bec5be7d57d2d8741820ade2f4bbdb3f8377
3
+ size 184976111
onnx/decoder_model_merged_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:128b382b33be99be84195be4b2e7cc78739f1219168dd2eb6d16a8b8bd329fea
3
- size 239072646
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:35cb2e3f953dec8fdbec8fc15fc42a71ef45392668d7e7250cd05bbf2eed5f86
3
+ size 239042891
onnx/decoder_model_merged_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:57800d3564f6eb787b279225f4a026d1c38e62d3cdcd657b030b15589f0153d6
3
- size 239072669
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4abc60988a2d76d0f386ab26216884d3b70f774677088c553d6c21a6912dda3e
3
+ size 239042896
onnx/decoder_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7b4d46ffe5fa822932fa077f2d022756ae1490f41bd87608eb763f214394ff21
3
- size 184730408
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:da0d760d02d93b04ad0256d00f028495f1097ad8798be7765b35deb6b88a8d45
3
+ size 184730800
onnx/decoder_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3a4b1a01cb5ee31bc0201ae88ce279f5bec7c663635f251cfc5d750414a904a9
3
- size 79520048
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b097c6a8ccd7133991491e78712ee14b66127b1aa2de4388854e3d0b5cfc2e12
3
+ size 79518624
onnx/decoder_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:26519ffe3e1f68afb30aaa8754ef16c654d1f250815395953b7cbcd51c0a9115
3
- size 79520071
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a68a7be36a273c1f2e82ec4ce01efa00749e01b7fd6abceb42a6b9ed57c8c2a8
3
+ size 79518647
onnx/decoder_with_past_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7c8a78cc1a8b4fe1a67c02274c91f9a9470e540698dc7c5b9eb6c31c66814761
3
- size 312126846
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:76695421c4b8c7b5ee1a7fe765195864c5a466524860d57128a0af511161e21e
3
+ size 293208489
onnx/decoder_with_past_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b7e59527529f05a531a5d4531b01d5e275d95fda9aac73b6c397f5f6184cfaea
3
- size 182371792
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:067b3102ef5b42a8b202abb64fb872fd2032611f7ad486c28419af6eeb2a66e8
3
+ size 179672355
onnx/decoder_with_past_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b5c46e9e1f70736f61ff775f3681ef22ceb737bf90f5b5aff07d48afe9a5d8f9
3
- size 156257470
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5c5b070538447245c170435c52228c590e11cfa65fef0f49250ed8c7466a8fb
3
+ size 146771221
onnx/decoder_with_past_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b097c6a8ccd7133991491e78712ee14b66127b1aa2de4388854e3d0b5cfc2e12
3
- size 79518624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:efbfd1e61d8c5eea4b6c0a01691a4450a7b83b51b6d26df4130df9bff664def1
3
+ size 74735701
onnx/decoder_with_past_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:da0d760d02d93b04ad0256d00f028495f1097ad8798be7765b35deb6b88a8d45
3
- size 184730800
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb4ac1f7fc080fd37ca9b7dfb878a9556cf81fcd104f8a3a6b4223098d1bf49e
3
+ size 181736507
onnx/decoder_with_past_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b097c6a8ccd7133991491e78712ee14b66127b1aa2de4388854e3d0b5cfc2e12
3
- size 79518624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:efbfd1e61d8c5eea4b6c0a01691a4450a7b83b51b6d26df4130df9bff664def1
3
+ size 74735701
onnx/decoder_with_past_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a68a7be36a273c1f2e82ec4ce01efa00749e01b7fd6abceb42a6b9ed57c8c2a8
3
- size 79518647
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:276b5a738692d99cdd94f376c0f42fac06c116d2adfc52348aa45e4540f8c3cc
3
+ size 74735718
onnx/encoder_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:edeb5c7ad5cdff61b5e5b311057ed17313e9caf03aa6bca642b1932a0b2959a9
3
- size 352791632
 
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 sha256:8829cb7f4370d6323a438d0d1c245cb31111ecd55cabed133ebc442c551a78b2
3
- size 60839978
 
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:a5d42a57e52601db5ed5d6eb6cef9a967f8490b331f646622e9382aaf71912ce
3
- size 176483370
 
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 sha256:8e5d2c12bd86b800acec15f7ced7d332dcc33aa0bf2a2beb60fc7e76d06ea481
3
- size 92201773
 
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 sha256:4199841ba69fc77b537c0868a92e7bf2871bf392f6f0ac40a98b286e39073d1d
3
- size 66147866
 
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 sha256:5e91f357af43fd96b9b674cf1970810d062e72a5e3602687bde0b4a873d72764
3
- size 92201816
 
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 sha256:5e91f357af43fd96b9b674cf1970810d062e72a5e3602687bde0b4a873d72764
3
- size 92201816
 
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
- "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",