Upload folder using huggingface_hub
Browse files- onnx/decoder_model_merged.onnx +2 -2
- onnx/decoder_model_merged_bnb4.onnx +2 -2
- onnx/decoder_model_merged_fp16.onnx +2 -2
- onnx/decoder_model_merged_int8.onnx +2 -2
- onnx/decoder_model_merged_q4.onnx +2 -2
- onnx/decoder_model_merged_quantized.onnx +2 -2
- onnx/decoder_model_merged_uint8.onnx +2 -2
- onnx/encoder_model.onnx +2 -2
- onnx/encoder_model_bnb4.onnx +2 -2
- onnx/encoder_model_fp16.onnx +2 -2
- onnx/encoder_model_int8.onnx +2 -2
- onnx/encoder_model_q4.onnx +2 -2
- onnx/encoder_model_quantized.onnx +2 -2
- onnx/encoder_model_uint8.onnx +2 -2
- quantize_config.json +21 -3
onnx/decoder_model_merged.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f5d26282d379e1475e9b336135db7cbaf8d7a1c92ca7f5321bda1a3ce5683697
|
3 |
+
size 615484675
|
onnx/decoder_model_merged_bnb4.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:44e0c9d1bebbaed3c61717ed95bc413f87126f2dff93a2d092e59215155fd047
|
3 |
+
size 226233541
|
onnx/decoder_model_merged_fp16.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:8def6797a31494f1244b270da883e6c7706cca649e049ee346a4f651f7037f35
|
3 |
+
size 308793929
|
onnx/decoder_model_merged_int8.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f3d1ed5ff2aa312fcb2daa68696a40fdf2def05ef18ee4b9da68f39406db31e8
|
3 |
+
size 156970445
|
onnx/decoder_model_merged_q4.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:13a95fc6f5525e91085bdbc81a42fc4a98f4e1cd9a342d589f5d5be992bcfce7
|
3 |
+
size 233309701
|
onnx/decoder_model_merged_quantized.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f3d1ed5ff2aa312fcb2daa68696a40fdf2def05ef18ee4b9da68f39406db31e8
|
3 |
+
size 156970445
|
onnx/decoder_model_merged_uint8.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d90691c379d3b0d5db19be2b4e96dfd3a7da283aed1a761ccc70965999ca8e4f
|
3 |
+
size 156970501
|
onnx/encoder_model.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:6c86e6e15983eec5c26af63fd6bebbe32b00a95090cf57bf630f97da4ac9d8a2
|
3 |
+
size 352858701
|
onnx/encoder_model_bnb4.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:99fb37353c3ac23ef96cd2d8c7f8af0a4b4b8b27df6f40ec85e548519d753654
|
3 |
+
size 60907047
|
onnx/encoder_model_fp16.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:cb4eec0577bf11ba9384bf9a589807688b3027fe796905578b49653b64c68eac
|
3 |
+
size 176653325
|
onnx/encoder_model_int8.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:480a75f446bd3f06699e7bf2acac85b73934a466a7576f9da089fd6c4505c3b6
|
3 |
+
size 92371696
|
onnx/encoder_model_q4.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:01c7662620fe3b2f4dfb520fc8229bc0a29ef39f3da2e371af80ba8529b09f57
|
3 |
+
size 66214935
|
onnx/encoder_model_quantized.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3af93b150edfc33f98d68c54d6d44acdb0510cc36e365ff20bbd1e9bd6b9218c
|
3 |
+
size 92371739
|
onnx/encoder_model_uint8.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3af93b150edfc33f98d68c54d6d44acdb0510cc36e365ff20bbd1e9bd6b9218c
|
3 |
+
size 92371739
|
quantize_config.json
CHANGED
@@ -5,18 +5,23 @@
|
|
5 |
"encoder_model": {
|
6 |
"op_types": [
|
7 |
"Add",
|
|
|
|
|
8 |
"Conv",
|
9 |
"Div",
|
10 |
"Erf",
|
|
|
11 |
"MatMul",
|
12 |
"Mul",
|
13 |
"Pow",
|
14 |
"ReduceMean",
|
15 |
"Reshape",
|
|
|
16 |
"Softmax",
|
17 |
"Sqrt",
|
18 |
"Sub",
|
19 |
-
"Transpose"
|
|
|
20 |
],
|
21 |
"weight_type": "QUInt8"
|
22 |
},
|
@@ -77,6 +82,7 @@
|
|
77 |
"decoder_model_merged": {
|
78 |
"op_types": [
|
79 |
"Add",
|
|
|
80 |
"Concat",
|
81 |
"Constant",
|
82 |
"ConstantOfShape",
|
@@ -114,18 +120,23 @@
|
|
114 |
"encoder_model": {
|
115 |
"op_types": [
|
116 |
"Add",
|
|
|
|
|
117 |
"Conv",
|
118 |
"Div",
|
119 |
"Erf",
|
|
|
120 |
"MatMul",
|
121 |
"Mul",
|
122 |
"Pow",
|
123 |
"ReduceMean",
|
124 |
"Reshape",
|
|
|
125 |
"Softmax",
|
126 |
"Sqrt",
|
127 |
"Sub",
|
128 |
-
"Transpose"
|
|
|
129 |
],
|
130 |
"weight_type": "QInt8"
|
131 |
},
|
@@ -186,6 +197,7 @@
|
|
186 |
"decoder_model_merged": {
|
187 |
"op_types": [
|
188 |
"Add",
|
|
|
189 |
"Concat",
|
190 |
"Constant",
|
191 |
"ConstantOfShape",
|
@@ -223,18 +235,23 @@
|
|
223 |
"encoder_model": {
|
224 |
"op_types": [
|
225 |
"Add",
|
|
|
|
|
226 |
"Conv",
|
227 |
"Div",
|
228 |
"Erf",
|
|
|
229 |
"MatMul",
|
230 |
"Mul",
|
231 |
"Pow",
|
232 |
"ReduceMean",
|
233 |
"Reshape",
|
|
|
234 |
"Softmax",
|
235 |
"Sqrt",
|
236 |
"Sub",
|
237 |
-
"Transpose"
|
|
|
238 |
],
|
239 |
"weight_type": "QUInt8"
|
240 |
},
|
@@ -295,6 +312,7 @@
|
|
295 |
"decoder_model_merged": {
|
296 |
"op_types": [
|
297 |
"Add",
|
|
|
298 |
"Concat",
|
299 |
"Constant",
|
300 |
"ConstantOfShape",
|
|
|
5 |
"encoder_model": {
|
6 |
"op_types": [
|
7 |
"Add",
|
8 |
+
"Concat",
|
9 |
+
"Constant",
|
10 |
"Conv",
|
11 |
"Div",
|
12 |
"Erf",
|
13 |
+
"Gather",
|
14 |
"MatMul",
|
15 |
"Mul",
|
16 |
"Pow",
|
17 |
"ReduceMean",
|
18 |
"Reshape",
|
19 |
+
"Shape",
|
20 |
"Softmax",
|
21 |
"Sqrt",
|
22 |
"Sub",
|
23 |
+
"Transpose",
|
24 |
+
"Unsqueeze"
|
25 |
],
|
26 |
"weight_type": "QUInt8"
|
27 |
},
|
|
|
82 |
"decoder_model_merged": {
|
83 |
"op_types": [
|
84 |
"Add",
|
85 |
+
"Cast",
|
86 |
"Concat",
|
87 |
"Constant",
|
88 |
"ConstantOfShape",
|
|
|
120 |
"encoder_model": {
|
121 |
"op_types": [
|
122 |
"Add",
|
123 |
+
"Concat",
|
124 |
+
"Constant",
|
125 |
"Conv",
|
126 |
"Div",
|
127 |
"Erf",
|
128 |
+
"Gather",
|
129 |
"MatMul",
|
130 |
"Mul",
|
131 |
"Pow",
|
132 |
"ReduceMean",
|
133 |
"Reshape",
|
134 |
+
"Shape",
|
135 |
"Softmax",
|
136 |
"Sqrt",
|
137 |
"Sub",
|
138 |
+
"Transpose",
|
139 |
+
"Unsqueeze"
|
140 |
],
|
141 |
"weight_type": "QInt8"
|
142 |
},
|
|
|
197 |
"decoder_model_merged": {
|
198 |
"op_types": [
|
199 |
"Add",
|
200 |
+
"Cast",
|
201 |
"Concat",
|
202 |
"Constant",
|
203 |
"ConstantOfShape",
|
|
|
235 |
"encoder_model": {
|
236 |
"op_types": [
|
237 |
"Add",
|
238 |
+
"Concat",
|
239 |
+
"Constant",
|
240 |
"Conv",
|
241 |
"Div",
|
242 |
"Erf",
|
243 |
+
"Gather",
|
244 |
"MatMul",
|
245 |
"Mul",
|
246 |
"Pow",
|
247 |
"ReduceMean",
|
248 |
"Reshape",
|
249 |
+
"Shape",
|
250 |
"Softmax",
|
251 |
"Sqrt",
|
252 |
"Sub",
|
253 |
+
"Transpose",
|
254 |
+
"Unsqueeze"
|
255 |
],
|
256 |
"weight_type": "QUInt8"
|
257 |
},
|
|
|
312 |
"decoder_model_merged": {
|
313 |
"op_types": [
|
314 |
"Add",
|
315 |
+
"Cast",
|
316 |
"Concat",
|
317 |
"Constant",
|
318 |
"ConstantOfShape",
|