Upload folder using huggingface_hub
Browse files- config.json +298 -0
- generation_config.json +12 -0
- onnx/decoder_model.onnx +3 -0
- onnx/decoder_model_bnb4.onnx +3 -0
- onnx/decoder_model_fp16.onnx +3 -0
- onnx/decoder_model_int8.onnx +3 -0
- onnx/decoder_model_merged.onnx +3 -0
- onnx/decoder_model_merged_bnb4.onnx +3 -0
- onnx/decoder_model_merged_fp16.onnx +3 -0
- onnx/decoder_model_merged_int8.onnx +3 -0
- onnx/decoder_model_merged_q4.onnx +3 -0
- onnx/decoder_model_merged_q4f16.onnx +3 -0
- onnx/decoder_model_merged_quantized.onnx +3 -0
- onnx/decoder_model_merged_uint8.onnx +3 -0
- onnx/decoder_model_q4.onnx +3 -0
- onnx/decoder_model_q4f16.onnx +3 -0
- onnx/decoder_model_quantized.onnx +3 -0
- onnx/decoder_model_uint8.onnx +3 -0
- onnx/decoder_with_past_model.onnx +3 -0
- onnx/decoder_with_past_model_bnb4.onnx +3 -0
- onnx/decoder_with_past_model_fp16.onnx +3 -0
- onnx/decoder_with_past_model_int8.onnx +3 -0
- onnx/decoder_with_past_model_q4.onnx +3 -0
- onnx/decoder_with_past_model_q4f16.onnx +3 -0
- onnx/decoder_with_past_model_quantized.onnx +3 -0
- onnx/decoder_with_past_model_uint8.onnx +3 -0
- onnx/encoder_model.onnx +3 -0
- onnx/encoder_model_bnb4.onnx +3 -0
- onnx/encoder_model_fp16.onnx +3 -0
- onnx/encoder_model_int8.onnx +3 -0
- onnx/encoder_model_q4.onnx +3 -0
- onnx/encoder_model_q4f16.onnx +3 -0
- onnx/encoder_model_quantized.onnx +3 -0
- onnx/encoder_model_uint8.onnx +3 -0
- quantize_config.json +18 -0
config.json
ADDED
@@ -0,0 +1,298 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_attn_implementation_autoset": true,
|
3 |
+
"_name_or_path": "efficient-speech/lite-whisper-large-v3-turbo-fast",
|
4 |
+
"activation_dropout": 0.0,
|
5 |
+
"activation_function": "gelu",
|
6 |
+
"apply_spec_augment": false,
|
7 |
+
"architectures": [
|
8 |
+
"LiteWhisperForConditionalGeneration"
|
9 |
+
],
|
10 |
+
"attention_dropout": 0.0,
|
11 |
+
"auto_map": {
|
12 |
+
"AutoConfig": "efficient-speech/lite-whisper-large-v3-turbo-fast--configuration_lite_whisper.LiteWhisperConfig",
|
13 |
+
"AutoModel": "efficient-speech/lite-whisper-large-v3-turbo-fast--modeling_lite_whisper.LiteWhisperForConditionalGeneration"
|
14 |
+
},
|
15 |
+
"begin_suppress_tokens": [
|
16 |
+
220,
|
17 |
+
50256
|
18 |
+
],
|
19 |
+
"bos_token_id": 50257,
|
20 |
+
"classifier_proj_size": 256,
|
21 |
+
"d_model": 1280,
|
22 |
+
"decoder_attention_heads": 20,
|
23 |
+
"decoder_ffn_dim": 5120,
|
24 |
+
"decoder_layerdrop": 0.0,
|
25 |
+
"decoder_layers": 4,
|
26 |
+
"decoder_start_token_id": 50258,
|
27 |
+
"dropout": 0.0,
|
28 |
+
"encoder_attention_heads": 20,
|
29 |
+
"encoder_ffn_dim": 5120,
|
30 |
+
"encoder_layerdrop": 0.0,
|
31 |
+
"encoder_layers": 32,
|
32 |
+
"eos_token_id": 50257,
|
33 |
+
"init_std": 0.02,
|
34 |
+
"is_encoder_decoder": true,
|
35 |
+
"low_rank_config": [
|
36 |
+
{
|
37 |
+
"fc1": 320,
|
38 |
+
"fc2": 256,
|
39 |
+
"k_proj": 32,
|
40 |
+
"out_proj": 32,
|
41 |
+
"q_proj": 32,
|
42 |
+
"v_proj": 48
|
43 |
+
},
|
44 |
+
{
|
45 |
+
"fc1": 176,
|
46 |
+
"fc2": 160,
|
47 |
+
"k_proj": 48,
|
48 |
+
"out_proj": 80,
|
49 |
+
"q_proj": 48,
|
50 |
+
"v_proj": 112
|
51 |
+
},
|
52 |
+
{
|
53 |
+
"fc1": 128,
|
54 |
+
"fc2": 64,
|
55 |
+
"k_proj": 32,
|
56 |
+
"out_proj": 112,
|
57 |
+
"q_proj": 32,
|
58 |
+
"v_proj": 112
|
59 |
+
},
|
60 |
+
{
|
61 |
+
"fc1": 160,
|
62 |
+
"fc2": 64,
|
63 |
+
"k_proj": 48,
|
64 |
+
"out_proj": 160,
|
65 |
+
"q_proj": 48,
|
66 |
+
"v_proj": 176
|
67 |
+
},
|
68 |
+
{
|
69 |
+
"fc1": 176,
|
70 |
+
"fc2": 96,
|
71 |
+
"k_proj": 48,
|
72 |
+
"out_proj": 112,
|
73 |
+
"q_proj": 48,
|
74 |
+
"v_proj": 176
|
75 |
+
},
|
76 |
+
{
|
77 |
+
"fc1": 160,
|
78 |
+
"fc2": 176,
|
79 |
+
"k_proj": 32,
|
80 |
+
"out_proj": 112,
|
81 |
+
"q_proj": 32,
|
82 |
+
"v_proj": 160
|
83 |
+
},
|
84 |
+
{
|
85 |
+
"fc1": 224,
|
86 |
+
"fc2": 192,
|
87 |
+
"k_proj": 64,
|
88 |
+
"out_proj": 192,
|
89 |
+
"q_proj": 64,
|
90 |
+
"v_proj": 304
|
91 |
+
},
|
92 |
+
{
|
93 |
+
"fc1": 224,
|
94 |
+
"fc2": 368,
|
95 |
+
"k_proj": 64,
|
96 |
+
"out_proj": 192,
|
97 |
+
"q_proj": 80,
|
98 |
+
"v_proj": 272
|
99 |
+
},
|
100 |
+
{
|
101 |
+
"fc1": 288,
|
102 |
+
"fc2": 448,
|
103 |
+
"k_proj": 64,
|
104 |
+
"out_proj": 224,
|
105 |
+
"q_proj": 80,
|
106 |
+
"v_proj": 272
|
107 |
+
},
|
108 |
+
{
|
109 |
+
"fc1": 288,
|
110 |
+
"fc2": 464,
|
111 |
+
"k_proj": 80,
|
112 |
+
"out_proj": 224,
|
113 |
+
"q_proj": 96,
|
114 |
+
"v_proj": 304
|
115 |
+
},
|
116 |
+
{
|
117 |
+
"fc1": 320,
|
118 |
+
"fc2": 480,
|
119 |
+
"k_proj": 48,
|
120 |
+
"out_proj": 240,
|
121 |
+
"q_proj": 48,
|
122 |
+
"v_proj": 288
|
123 |
+
},
|
124 |
+
{
|
125 |
+
"fc1": 304,
|
126 |
+
"fc2": 480,
|
127 |
+
"k_proj": 64,
|
128 |
+
"out_proj": 224,
|
129 |
+
"q_proj": 64,
|
130 |
+
"v_proj": 320
|
131 |
+
},
|
132 |
+
{
|
133 |
+
"fc1": 336,
|
134 |
+
"fc2": 544,
|
135 |
+
"k_proj": 48,
|
136 |
+
"out_proj": 224,
|
137 |
+
"q_proj": 64,
|
138 |
+
"v_proj": 368
|
139 |
+
},
|
140 |
+
{
|
141 |
+
"fc1": 400,
|
142 |
+
"fc2": 656,
|
143 |
+
"k_proj": 48,
|
144 |
+
"out_proj": 272,
|
145 |
+
"q_proj": 48,
|
146 |
+
"v_proj": 304
|
147 |
+
},
|
148 |
+
{
|
149 |
+
"fc1": 464,
|
150 |
+
"fc2": 608,
|
151 |
+
"k_proj": 64,
|
152 |
+
"out_proj": 288,
|
153 |
+
"q_proj": 64,
|
154 |
+
"v_proj": 464
|
155 |
+
},
|
156 |
+
{
|
157 |
+
"fc1": 352,
|
158 |
+
"fc2": 720,
|
159 |
+
"k_proj": 32,
|
160 |
+
"out_proj": 352,
|
161 |
+
"q_proj": 48,
|
162 |
+
"v_proj": 400
|
163 |
+
},
|
164 |
+
{
|
165 |
+
"fc1": 384,
|
166 |
+
"fc2": 752,
|
167 |
+
"k_proj": 48,
|
168 |
+
"out_proj": 224,
|
169 |
+
"q_proj": 48,
|
170 |
+
"v_proj": 288
|
171 |
+
},
|
172 |
+
{
|
173 |
+
"fc1": 480,
|
174 |
+
"fc2": 784,
|
175 |
+
"k_proj": 80,
|
176 |
+
"out_proj": 304,
|
177 |
+
"q_proj": 96,
|
178 |
+
"v_proj": 384
|
179 |
+
},
|
180 |
+
{
|
181 |
+
"fc1": 512,
|
182 |
+
"fc2": 848,
|
183 |
+
"k_proj": 64,
|
184 |
+
"out_proj": 272,
|
185 |
+
"q_proj": 64,
|
186 |
+
"v_proj": 320
|
187 |
+
},
|
188 |
+
{
|
189 |
+
"fc1": 560,
|
190 |
+
"fc2": 784,
|
191 |
+
"k_proj": 64,
|
192 |
+
"out_proj": 272,
|
193 |
+
"q_proj": 64,
|
194 |
+
"v_proj": 416
|
195 |
+
},
|
196 |
+
{
|
197 |
+
"fc1": 512,
|
198 |
+
"fc2": 16,
|
199 |
+
"k_proj": 64,
|
200 |
+
"out_proj": 352,
|
201 |
+
"q_proj": 64,
|
202 |
+
"v_proj": 432
|
203 |
+
},
|
204 |
+
{
|
205 |
+
"fc1": 704,
|
206 |
+
"fc2": 992,
|
207 |
+
"k_proj": 96,
|
208 |
+
"out_proj": 368,
|
209 |
+
"q_proj": 112,
|
210 |
+
"v_proj": 512
|
211 |
+
},
|
212 |
+
{
|
213 |
+
"fc1": 656,
|
214 |
+
"fc2": 1024,
|
215 |
+
"k_proj": 112,
|
216 |
+
"out_proj": 496,
|
217 |
+
"q_proj": 112,
|
218 |
+
"v_proj": 576
|
219 |
+
},
|
220 |
+
{
|
221 |
+
"fc1": 752,
|
222 |
+
"k_proj": 96,
|
223 |
+
"out_proj": 496,
|
224 |
+
"q_proj": 112,
|
225 |
+
"v_proj": 528
|
226 |
+
},
|
227 |
+
{
|
228 |
+
"fc1": 800,
|
229 |
+
"k_proj": 144,
|
230 |
+
"out_proj": 512,
|
231 |
+
"q_proj": 160,
|
232 |
+
"v_proj": 624
|
233 |
+
},
|
234 |
+
{
|
235 |
+
"fc1": 816,
|
236 |
+
"k_proj": 128,
|
237 |
+
"out_proj": 464,
|
238 |
+
"q_proj": 144
|
239 |
+
},
|
240 |
+
{
|
241 |
+
"fc1": 912,
|
242 |
+
"k_proj": 160,
|
243 |
+
"out_proj": 576,
|
244 |
+
"q_proj": 176
|
245 |
+
},
|
246 |
+
{
|
247 |
+
"fc1": 976,
|
248 |
+
"k_proj": 160,
|
249 |
+
"out_proj": 560,
|
250 |
+
"q_proj": 192
|
251 |
+
},
|
252 |
+
{
|
253 |
+
"fc1": 976,
|
254 |
+
"k_proj": 176,
|
255 |
+
"out_proj": 576,
|
256 |
+
"q_proj": 192
|
257 |
+
},
|
258 |
+
{
|
259 |
+
"fc1": 976,
|
260 |
+
"k_proj": 240,
|
261 |
+
"out_proj": 624,
|
262 |
+
"q_proj": 272,
|
263 |
+
"v_proj": 608
|
264 |
+
},
|
265 |
+
{
|
266 |
+
"fc1": 960,
|
267 |
+
"k_proj": 272,
|
268 |
+
"out_proj": 640,
|
269 |
+
"q_proj": 320,
|
270 |
+
"v_proj": 640
|
271 |
+
},
|
272 |
+
{
|
273 |
+
"fc1": 912,
|
274 |
+
"fc2": 896,
|
275 |
+
"k_proj": 272,
|
276 |
+
"q_proj": 320,
|
277 |
+
"v_proj": 592
|
278 |
+
}
|
279 |
+
],
|
280 |
+
"mask_feature_length": 10,
|
281 |
+
"mask_feature_min_masks": 0,
|
282 |
+
"mask_feature_prob": 0.0,
|
283 |
+
"mask_time_length": 10,
|
284 |
+
"mask_time_min_masks": 2,
|
285 |
+
"mask_time_prob": 0.05,
|
286 |
+
"max_source_positions": 1500,
|
287 |
+
"max_target_positions": 448,
|
288 |
+
"median_filter_width": 7,
|
289 |
+
"model_type": "lite-whisper",
|
290 |
+
"num_hidden_layers": 32,
|
291 |
+
"num_mel_bins": 128,
|
292 |
+
"pad_token_id": 50257,
|
293 |
+
"scale_embedding": false,
|
294 |
+
"transformers_version": "4.48.3",
|
295 |
+
"use_cache": true,
|
296 |
+
"use_weighted_layer_sum": false,
|
297 |
+
"vocab_size": 51866
|
298 |
+
}
|
generation_config.json
ADDED
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_from_model_config": true,
|
3 |
+
"begin_suppress_tokens": [
|
4 |
+
220,
|
5 |
+
50256
|
6 |
+
],
|
7 |
+
"bos_token_id": 50257,
|
8 |
+
"decoder_start_token_id": 50258,
|
9 |
+
"eos_token_id": 50257,
|
10 |
+
"pad_token_id": 50257,
|
11 |
+
"transformers_version": "4.48.3"
|
12 |
+
}
|
onnx/decoder_model.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:8eb5333eafa0143c92cc32643a4c6d5b05142fb02ea83b893717a30155c57010
|
3 |
+
size 687782706
|
onnx/decoder_model_bnb4.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2d22ce250dcbdbb76d2d7d1b08d5c5aefa02f95dd60737b64bf624a5fbb19437
|
3 |
+
size 327340989
|
onnx/decoder_model_fp16.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:bb940c9c4d8ba82a8cbe40595979420cb4ea0ccb485385171a9abf7421017e8d
|
3 |
+
size 343968493
|
onnx/decoder_model_int8.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e86e86f9623268ff79cbf176f96650f57017a3d22c58f147546c85c305ce95fb
|
3 |
+
size 437927549
|
onnx/decoder_model_merged.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:10baac2f0ddf0b9793c01f5059da1a4634bbc9458ff31f6d9411af993a78a59a
|
3 |
+
size 687963426
|
onnx/decoder_model_merged_bnb4.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e9635d84d161b1600ea764bbdefa108fcea2897cac83a8ac3e328df8bbd393be
|
3 |
+
size 327526397
|
onnx/decoder_model_merged_fp16.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1eabb33d307b641cd8f817464fdeddc83c50059139eca25d95017b026f8d24bc
|
3 |
+
size 661764234
|
onnx/decoder_model_merged_int8.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:278fb404b28b0362fccc2a8f1df4e284a5332117e028d90b64358629be91d188
|
3 |
+
size 438149217
|
onnx/decoder_model_merged_q4.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ab58bfa5cc9da9682413a3b404d84ec3f5037d1c34d8a3d1719354617c826cec
|
3 |
+
size 334079385
|
onnx/decoder_model_merged_q4f16.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:231219a8e1b553a86dfecd12f43ef624f67937ef86b03a27bbdc19790eebe545
|
3 |
+
size 379152361
|
onnx/decoder_model_merged_quantized.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:278fb404b28b0362fccc2a8f1df4e284a5332117e028d90b64358629be91d188
|
3 |
+
size 438149217
|
onnx/decoder_model_merged_uint8.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:a633751b728a17730ea870b71e1cdee4fd813ddd4fa5913fad2bfbf09d1a0b6b
|
3 |
+
size 438149238
|
onnx/decoder_model_q4.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:6f9d42edcd8ac943f671ac2533710f588826fc6e9e2df3323fdee53b1b835776
|
3 |
+
size 333894265
|
onnx/decoder_model_q4f16.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3c19dbc0e9893180d4ebe52a2876e9e477cd62ab66e06fa5fca1e84f30bc4904
|
3 |
+
size 193241747
|
onnx/decoder_model_quantized.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e86e86f9623268ff79cbf176f96650f57017a3d22c58f147546c85c305ce95fb
|
3 |
+
size 437927549
|
onnx/decoder_model_uint8.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:12ec6dc33d88d954b2a96a9ef032de7ea2cb70c8645a608b2267ae13ce499a41
|
3 |
+
size 437927570
|
onnx/decoder_with_past_model.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:dd14258616b60e351b56d95604e1ea814598da1ab70e7d83de81e5cabd67ff92
|
3 |
+
size 635300182
|
onnx/decoder_with_past_model_bnb4.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1f2c2dcd795803c5b4d52d4cd75bfdaeb50a71818e8f627c0e6274eee61dd9b5
|
3 |
+
size 319913217
|
onnx/decoder_with_past_model_fp16.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:03f1c82ecc508471d4ef8b11f7e778407fa874d1fcbe9555b77edc34b79ee1b4
|
3 |
+
size 317714162
|
onnx/decoder_with_past_model_int8.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1768c4d4aa26ba933d62330dfec1425f1a9098963ede3b3b55bffedb5b61ba5b
|
3 |
+
size 424758658
|
onnx/decoder_with_past_model_q4.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c61f60c5af815fd9078fdfe8d3251e8a321289f944cacf0cc94c9e8f3a325095
|
3 |
+
size 325647357
|
onnx/decoder_with_past_model_q4f16.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:314139e11aa3c8d293a967d130a0f9c3c8a1d71dcd5964bc4d6ec29ea22badfa
|
3 |
+
size 185827736
|
onnx/decoder_with_past_model_quantized.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1768c4d4aa26ba933d62330dfec1425f1a9098963ede3b3b55bffedb5b61ba5b
|
3 |
+
size 424758658
|
onnx/decoder_with_past_model_uint8.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:839a031db61e540501eb2a544a40452d3d933dbd0f9e4fbae5f2cb36474f7772
|
3 |
+
size 424758674
|
onnx/encoder_model.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e8c82a0a478f5ea3c1ba1716305d4d865d6ee9adb01b80ddad8ae18073919f5a
|
3 |
+
size 1260624040
|
onnx/encoder_model_bnb4.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ce584eda8f6949c3fa883e1f56c200725652df89fec67891a13b4f94d87e476c
|
3 |
+
size 204761607
|
onnx/encoder_model_fp16.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ca9f3de72ac90cafacf18ef92ff2379973bf9b09a91cd1206bc963611e453517
|
3 |
+
size 630536397
|
onnx/encoder_model_int8.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d50ed04d0b11da08f296f1869dfe73d4be5973948c500d48bf1b56277f0246c8
|
3 |
+
size 323283647
|
onnx/encoder_model_q4.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:403f5274c75ef9d7a640057f5daf99acb0614da9b28e109ab7c0206b65739f4c
|
3 |
+
size 225647102
|
onnx/encoder_model_q4f16.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:8357a970639b6b994d1050c18d237b28eaebbe212352c6e17bf2e5bd6ec08414
|
3 |
+
size 190550837
|
onnx/encoder_model_quantized.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:9d431f00bcbbd6c96f643184eb780c3a4fa9ac4ef5f74ce21e9dd1facc78d7cd
|
3 |
+
size 323283809
|
onnx/encoder_model_uint8.onnx
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:9d431f00bcbbd6c96f643184eb780c3a4fa9ac4ef5f74ce21e9dd1facc78d7cd
|
3 |
+
size 323283809
|
quantize_config.json
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"modes": [
|
3 |
+
"fp16",
|
4 |
+
"q8",
|
5 |
+
"int8",
|
6 |
+
"uint8",
|
7 |
+
"q4",
|
8 |
+
"q4f16",
|
9 |
+
"bnb4"
|
10 |
+
],
|
11 |
+
"per_channel": false,
|
12 |
+
"reduce_range": false,
|
13 |
+
"block_size": null,
|
14 |
+
"is_symmetric": true,
|
15 |
+
"accuracy_level": null,
|
16 |
+
"quant_type": 1,
|
17 |
+
"op_block_list": null
|
18 |
+
}
|