Xenova HF staff commited on
Commit
43102e2
1 Parent(s): f23880f

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 sha256:41d67ea337f7a6ef56feb6d53dc87e1941a3385268614afcf67481a32d3517d5
3
- size 208327950
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:21307ff74dc502fe5a4aacf80f2d316fc67976b5ee5810cc209ab85a9a443abd
3
+ size 208328538
onnx/decoder_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:cc80c7f4935199c487674bf1ec30dff4d202eb43aab6a9860de8832bb33dc67b
3
- size 121829857
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e239382eabc45af9b237404ffffa11dd210ad037d6f9ac7234a64ecb04c7338
3
+ size 121830445
onnx/decoder_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ebfa5a94504b32f50101a16be83d5e0276a31f959d4fdf1c1531ae074d1bd596
3
- size 104447760
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4db2c1dabe0d79f459131188a215d52e088a894e2ba601ce6375396b53359822
3
+ size 104449669
onnx/decoder_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:262888fb4d90f8b0ef9e5800215736e7195b15c4489f521076ac6fc3877d9712
3
- size 53364695
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19afc4dcada1c72c980dda2091cea031172d42156be50e802c95ca128a5344a5
3
+ size 53362551
onnx/decoder_model_merged.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5d2545e048b04cfa631c546eb2c3f5b3bbf3bdec2782a3fa639ec3f7cb959640
3
- size 208717713
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4f9ed411c882d9c82466a98fd5fb95225916d671cf9668ddf51fb39aa88fd8ab
3
+ size 208686733
onnx/decoder_model_merged_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c13b1073ddbab2cddd1fbe6232a7d97a43b2f90a1da2f403a6b9e81efcdfe22f
3
- size 122228140
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d508ff3f2d7dcc9d6dce588b42eeafbbfdf4f7c92ef6580da6a4ea00eff9cdf8
3
+ size 122195672
onnx/decoder_model_merged_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:95c830a0ca7b81ccb7ba7604605e0b28aaf5b2d3877503e77d792ef2484e78a2
3
- size 104716300
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06dba593de2aef1169545f662feccf20e6a0e5f525870beb560987799a28c93f
3
+ size 104689258
onnx/decoder_model_merged_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ec01317e837072c39502f9521415e9fc3340cec6604ba0fd86d6e859a07d347d
3
- size 159920321
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4471de9d8803a415be0d0a3882d61104aaa7a46bba72c3b5dc9b27616e1bb8c5
3
+ size 159878259
onnx/decoder_model_merged_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d384ffb5a2a4e1bc318351649ca4b451d0e93e532b809a2e87af6d59f3a0393c
3
- size 123799984
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b1f37f4b5a5328c965f0d8f8f32e4a9e6f0a04c2634d2171c051c1332212dcc7
3
+ size 123767624
onnx/decoder_model_merged_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ec01317e837072c39502f9521415e9fc3340cec6604ba0fd86d6e859a07d347d
3
- size 159920321
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4471de9d8803a415be0d0a3882d61104aaa7a46bba72c3b5dc9b27616e1bb8c5
3
+ size 159878259
onnx/decoder_model_merged_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:06bd9a155d0a4f74e9c7ead7703a53af06a9359f73adda408e18405210c8d0ec
3
- size 159920349
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b6db5750d0b51a267fb4a85b53a4f134313866302e346ec957c8ad1ae45310b8
3
+ size 159878224
onnx/decoder_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ca34aaa1e523e582c80633580ca98182a11597f923912c8704d6efb504c2925c
3
- size 123402241
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:508603ff2a878851b11d1aef446475985611c2989d920e95db94ef4ea361a04e
3
+ size 123402829
onnx/decoder_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:262888fb4d90f8b0ef9e5800215736e7195b15c4489f521076ac6fc3877d9712
3
- size 53364695
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19afc4dcada1c72c980dda2091cea031172d42156be50e802c95ca128a5344a5
3
+ size 53362551
onnx/decoder_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4228982ec3a8623d88aa2acf2694ccfff99ca15bc144e29dade374b439b9e2e7
3
- size 53364723
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d25b88971cd97d70c54e1fc0fb4538633b9fd8f48ea98f012ffbff2fb55eda2a
3
+ size 53362579
onnx/decoder_with_past_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:21307ff74dc502fe5a4aacf80f2d316fc67976b5ee5810cc209ab85a9a443abd
3
- size 208328538
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c4ea98c410bf332c61804a37452eef363b1e6ea35175e43d195b931abfa0fba1
3
+ size 195689406
onnx/decoder_with_past_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6e239382eabc45af9b237404ffffa11dd210ad037d6f9ac7234a64ecb04c7338
3
- size 121830445
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4221d1c90a619dcd40e63f3e8ce98c1994674fd4a090c1aece6aec8233c0af39
3
+ size 120002869
onnx/decoder_with_past_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4db2c1dabe0d79f459131188a215d52e088a894e2ba601ce6375396b53359822
3
- size 104449669
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9fc5b225e776439048d570b39fec8af863f80af1cc9fab62cc39b407e9117aa5
3
+ size 98092909
onnx/decoder_with_past_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:19afc4dcada1c72c980dda2091cea031172d42156be50e802c95ca128a5344a5
3
- size 53362551
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bdd92860d0ed7dff2aca623963378cbba1b617bfae127356db1c8aa8baa930ef
3
+ size 50131672
onnx/decoder_with_past_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:508603ff2a878851b11d1aef446475985611c2989d920e95db94ef4ea361a04e
3
- size 123402829
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d864ca26509968b00d92e6823c4db7ac460c106f9228a21bc5199e9893fe4126
3
+ size 121378741
onnx/decoder_with_past_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:19afc4dcada1c72c980dda2091cea031172d42156be50e802c95ca128a5344a5
3
- size 53362551
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bdd92860d0ed7dff2aca623963378cbba1b617bfae127356db1c8aa8baa930ef
3
+ size 50131672
onnx/decoder_with_past_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d25b88971cd97d70c54e1fc0fb4538633b9fd8f48ea98f012ffbff2fb55eda2a
3
- size 53362579
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e1154b873a3cadc45a9a88044ab4d0f39303cbb3315dadd852a161d1e791769
3
+ size 50131695
onnx/encoder_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:cfe19000dd7cb4503b24699f3ec7fe9782e7964f72d2a0f9f26704e0d4e56af9
3
- size 82451564
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7fcea817bb2be4d86729b521e5a7fcbec28fa743edfed67e882b33ff15852540
3
+ size 82451730
onnx/encoder_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3c236d004b4c6b579e04bc919c85ec3f5b95bec88399a9543ff98e3242d9b27b
3
- size 17576595
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c104148ee3ec3ae79653a99340e372eb5ab7a15181cc6808811b2f2c699c0ff2
3
+ size 17576761
onnx/encoder_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a714353428970cf18470bccad9255a36c7563f25e9afbd63019f0551d3e89109
3
- size 41270531
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d5a450fd2162fbd313685cda39911ac0af144c11419160849ce6cf551081f2b
3
+ size 41270731
onnx/encoder_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1c8a766757371f1d84290d8c2d3253ad544a85b55ef8acf4fab38c888789e5d9
3
- size 23139286
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:da5b7763e8e39bc23320fbc10eeed875c3be5793e94f5296ae0c84860bdebff5
3
+ size 23139452
onnx/encoder_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f3ecc03f1fbb121edbde89ba6178939a14701400289b704cc5404996e3f12374
3
- size 18755955
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb43b6e36f683ee4afd174f7cd7053794798e9225f38c036aea864e28618e138
3
+ size 18756121
onnx/encoder_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e95035a32dc5248a56eabd7ec739f28a74d968771afd4f8d33e38d03091afb52
3
- size 23139303
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e0de70ab1791a8fe02f22af2d5df0e18c9732bdf018cce8c1028dd9f1df8c91
3
+ size 23139469
onnx/encoder_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e95035a32dc5248a56eabd7ec739f28a74d968771afd4f8d33e38d03091afb52
3
- size 23139303
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e0de70ab1791a8fe02f22af2d5df0e18c9732bdf018cce8c1028dd9f1df8c91
3
+ size 23139469
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",