Xenova HF staff commited on
Commit
5b146eb
1 Parent(s): fce5939

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 sha256:80fac563336bea3574cd5d8736c6420ea4510bffef32265605c8b0df75f514f7
3
- size 118421401
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f17cb6e082755647bbd0c525600e375ee8da9072cc3214be89783c8eced22ee0
3
+ size 118421793
onnx/decoder_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f7a72bfd996da5864b45dd4a937c13565ff72433f9eef718b9bdd3f6b2c6d932
3
- size 85987300
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a5e4795d2a06d65e55b0ebfc9578dce62978ea4baaf29315cd7e43ddb59835fd
3
+ size 85987692
onnx/decoder_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:030d1a72da73123948ff6de4552e919a639999f4fc9f02b44064d624680946a3
3
- size 59403669
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1420bbf3b19b6a350b4ff1b7c9cbed2ac23e4465a8eedf2bf9ae40b23017ffba
3
+ size 59404932
onnx/decoder_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:763b40325e399cf503b2550191c415ea1bab051189f9657b7108b1e926e7e306
3
- size 30497015
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac5f0854305a46cf47ea5db14ddcf7c307ace84480255b785b1c51616b3e6a45
3
+ size 30495591
onnx/decoder_model_merged.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:698476770007285b00d3ada564c2741f6c7c2468d50736ce925626da2a09445f
3
- size 118684972
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b7fcea2416195cfcf2c781ed235c5251309c6cb7a61f696d4d8893b4502f828
3
+ size 118662672
onnx/decoder_model_merged_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:906f04bcc1d3d50188d3166836818211c603e8e535512916015108e3ea76f259
3
- size 86256551
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca5be5a92a2c08f58fe86ea8ccde501d0e981160aa8b105287174a6c7565e597
3
+ size 86233256
onnx/decoder_model_merged_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:bbae935a08431da2dbb181349aaaa4cc87005e430c7d8446765ce02193231fec
3
- size 59584799
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dbe0a55ae5384c941fd17173f99b71c8a1b88c2f26fd8b419025af55c8e1f63b
3
+ size 59565109
onnx/decoder_model_merged_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:646e214ed05e31729dd9855d59bb32b7df785f9d6a84e0c2f802d9f96ee83358
3
- size 110386509
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3f3cb0c497318bcc6a3f0f56d299e31d40dcf6ad7708fa464ac821e2bf67d89c
3
+ size 110356751
onnx/decoder_model_merged_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b7d3aa6168920e8cf4e935c5bbca9530c38e32006b13c0e8285d3b291a2d9928
3
- size 86845767
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96b6241b2dc99848d0d9f3a85e6cbe84040fefabb63a61ae4ff6ee6f0cad224a
3
+ size 86822544
onnx/decoder_model_merged_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:646e214ed05e31729dd9855d59bb32b7df785f9d6a84e0c2f802d9f96ee83358
3
- size 110386509
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3f3cb0c497318bcc6a3f0f56d299e31d40dcf6ad7708fa464ac821e2bf67d89c
3
+ size 110356751
onnx/decoder_model_merged_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:60cb6157d68ca6ba265055f2e5667a5a17c15108a46e7d9b2beac1182758c74e
3
- size 110386531
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be450180ea73042fe2c05a45eba4deb0a95f5ffb07baa59c15e4a5f59874a6fa
3
+ size 110356737
onnx/decoder_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:69f6108b49e59bba728b2cabb59717b602debee36bac4ff6c7357c5f5e465443
3
- size 86576876
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5a1bdf1444e48fb0f5419f7ad4123a4fede2c570656e1905dffd62682b6bb3f9
3
+ size 86577268
onnx/decoder_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:763b40325e399cf503b2550191c415ea1bab051189f9657b7108b1e926e7e306
3
- size 30497015
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac5f0854305a46cf47ea5db14ddcf7c307ace84480255b785b1c51616b3e6a45
3
+ size 30495591
onnx/decoder_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4cf2d4c8efa2bbb759bab1d1f829e6a7c91bf3ad5a8d59dda6d38396660cf57f
3
- size 30497037
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:825e8ef54fc274d167c6a91cc84cc216c3d589b01b68adc730f8bfc2460c2f0e
3
+ size 30495613
onnx/decoder_with_past_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f17cb6e082755647bbd0c525600e375ee8da9072cc3214be89783c8eced22ee0
3
- size 118421793
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:40d55aa4354d0364c0e9fe4e9794d00452f5dca644274ba2ee12494fa3cbea53
3
+ size 113665366
onnx/decoder_with_past_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a5e4795d2a06d65e55b0ebfc9578dce62978ea4baaf29315cd7e43ddb59835fd
3
- size 85987692
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:df663fb5a0aa4f2ea6004a566e5d322b336e80a2bed1c8c6a4e37631728dd03a
3
+ size 85285065
onnx/decoder_with_past_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1420bbf3b19b6a350b4ff1b7c9cbed2ac23e4465a8eedf2bf9ae40b23017ffba
3
- size 59404932
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b4cf5d8a1d6b53ab9cc7e7eb252bcffcc3758351b2d7269316793101639bbe9a
3
+ size 56999634
onnx/decoder_with_past_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ac5f0854305a46cf47ea5db14ddcf7c307ace84480255b785b1c51616b3e6a45
3
- size 30495591
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6597233e5dff7b3defd55b5863719fe212c1977d1780ea9d7fa98c094751c2b6
3
+ size 29257747
onnx/decoder_with_past_model_q4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:5a1bdf1444e48fb0f5419f7ad4123a4fede2c570656e1905dffd62682b6bb3f9
3
- size 86577268
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a9876aef9500c6731312f348c630a7503a0ac0e3efedadeade9cf4abf35acf4a
3
+ size 85800961
onnx/decoder_with_past_model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ac5f0854305a46cf47ea5db14ddcf7c307ace84480255b785b1c51616b3e6a45
3
- size 30495591
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6597233e5dff7b3defd55b5863719fe212c1977d1780ea9d7fa98c094751c2b6
3
+ size 29257747
onnx/decoder_with_past_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:825e8ef54fc274d167c6a91cc84cc216c3d589b01b68adc730f8bfc2460c2f0e
3
- size 30495613
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:779e9f5398125afa1b7e102b4bbc577ca9ab58bc39ddfe3db7b5aa9284483987
3
+ size 29257765
onnx/encoder_model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9fcbbb4e63569a64be1306b60e2cee7d7fb39eeb215d491a12358dfc5303da02
3
- size 32894268
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e19140280ff8ae5299e4a82566e6c33e809f45b635428f7674ac821dd64f2ccc
3
+ size 32894434
onnx/encoder_model_bnb4.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:926e365c4f704a3a34f33c59514a7f452814c4cdae6f59eb65811c3d662f8d81
3
- size 8567727
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:45894efc50d1a3404968e81e0cbc2bbf2d8d46c1f15b3fc06d52f53054149a70
3
+ size 8567893
onnx/encoder_model_fp16.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4e9eab41f0a2ac9d25c59c8ee16ae39233525f5bfba9be7f6ee98aa0b293b212
3
- size 16477669
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a212ae8bef18cbc3ac0a305d9f381ca588d768114d2ab2d0dd4296211555f69
3
+ size 16477869
onnx/encoder_model_int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:83f03b1606750d6e1499773d4472c0d6295b06813fe75f4c718c882b70771e74
3
- size 10083488
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb49545e8000c1008d77421ff34e96f538eb563df31c8c562ebf2b4cf461714b
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:dd9343b0237dc0713265d9b1b2f679edf9c01fd3d6ce4f8507d45c4bd4c2bafe
3
- size 9009943
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:59adf4d9deae157666a4d2b0de1f22624a252cc58af663b69ffdc3b0b1937e7f
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:f90b18c7705df4a7f2861ee7f3108fbe954a3a3f8df8abfa4505e5a71403bd0c
3
- size 10083504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e8279fe0e48026ceed826b00bc866d697ae3727b4c68f1881beb6127aa55cafb
3
+ size 10083670
onnx/encoder_model_uint8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f90b18c7705df4a7f2861ee7f3108fbe954a3a3f8df8abfa4505e5a71403bd0c
3
- size 10083504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e8279fe0e48026ceed826b00bc866d697ae3727b4c68f1881beb6127aa55cafb
3
+ size 10083670
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",