Hugging Face's logo

Intel
/
Qwen2-57B-A14B-Instruct-int4-inc

Safetensors
qwen2_moe
4-bit precision
intel/auto-round
Model card Files
xet
Community
wenhuach commited on
Commit
730c79e
·
1 Parent(s): 679bddf

replace with autogptq format

Browse files

Signed-off-by: wenhuach <[email protected]>

Files changed (13) hide show
  1. README.md +0 -161
  2. config.json +216 -350
  3. generation_config.json +1 -1
  4. model-00001-of-00007.safetensors +0 -3
  5. model-00002-of-00007.safetensors +0 -3
  6. model-00003-of-00007.safetensors +0 -3
  7. model-00005-of-00007.safetensors +0 -3
  8. model-00006-of-00007.safetensors +0 -3
  9. model-00007-of-00007.safetensors +0 -3
  10. model-00004-of-00007.safetensors → model.safetensors +2 -2
  11. model.safetensors.index.json +0 -0
  12. quantization_config.json +0 -360
  13. quantize_config.json +229 -0
README.md DELETED
@@ -1,161 +0,0 @@
1
- ---
2
- license: apache-2.0
3
- datasets:
4
- - NeelNanda/pile-10k
5
-
6
- ---
7
-
8
- ## Model Details
9
-
10
- This model is an int4 model with group_size 128 of [Qwen/Qwen2-57B-A14B-Instruct](https://huggingface.co/Qwen/Qwen2-57B-A14B-Instruct) generated by [intel/auto-round](https://github.com/intel/auto-round), auto-round is needed to run this model
11
-
12
- ## How To Use
13
-
14
- ### INT4 CPU/CUDA Inference
15
-
16
-
17
-
18
- ```python
19
- ##git clone https://github.com/intel/auto-round.git
20
- ##cd auto-round && pip install -vvv --no-build-isolation -e .
21
- from auto_round import AutoHfQuantizer ##must import
22
- import torch
23
- from transformers import AutoModelForCausalLM,AutoTokenizer
24
- quantized_model_dir = "Intel/Qwen2-57B-A14B-Instruct-int4-inc"
25
- tokenizer = AutoTokenizer.from_pretrained(quantized_model_dir)
26
-
27
- model = AutoModelForCausalLM.from_pretrained(
28
- quantized_model_dir,
29
- torch_dtype=torch.float16,
30
- device_map="auto",
31
- )
32
- prompt = "There is a girl who likes adventure,"
33
- messages = [
34
- {"role": "system", "content": "You are a helpful assistant."},
35
- {"role": "user", "content": prompt}
36
- ]
37
-
38
- tokenizer = AutoTokenizer.from_pretrained(quantized_model_dir)
39
- text = tokenizer.apply_chat_template(
40
- messages,
41
- tokenize=False,
42
- add_generation_prompt=True
43
- )
44
- model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
45
-
46
- generated_ids = model.generate(
47
- model_inputs.input_ids,
48
- max_new_tokens=50, ##change this to align with the official usage
49
- do_sample=False ##change this to align with the official usage
50
- )
51
- generated_ids = [
52
- output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
53
- ]
54
-
55
- response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
56
- print(response)
57
-
58
- ##prompt = "请介绍一下阿里巴巴公司"
59
- ##阿里巴巴集团是一家中国跨国科技公司，成立于1999年，总部位于杭州。阿里巴巴的业务涵盖了电子商务、零售、金融、物流、云计算等多个领域，是全球最大的电子商务公司之一。\n 阿里巴巴旗下拥有淘宝网、天猫、
60
-
61
- ##prompt = "9.8大还是9.11大"
62
- ##9.8和9.11都是小数，但是9.8比9.11大。在数学中，小数的大小取决于它们的数值，数值越大则越“大”。在这个情况下，9.8的
63
-
64
- ##prompt = "Once upon a time,"
65
- ##there was a kingdom far, far away. In this kingdom, there lived a beautiful princess who had hair as golden as the sun and eyes as blue as the sea. The princess was kind and gentle, and everyone in the kingdom loved her dearly.
66
-
67
- ##prompt = "There is a girl who likes adventure,"
68
- ##That's great to hear! Adventure can be a wonderful way to explore new places, learn new things, and challenge yourself in exciting ways. If you're looking for ideas on how to embark on an adventure, here are a few suggestions: 1.
69
- ```
70
-
71
-
72
-
73
- ### Evaluate the model
74
-
75
- pip3 install lm-eval==0.4.2
76
-
77
- ```bash
78
- git clone https://github.com/intel/auto-round
79
- cd auto-round/examples/language-modeling
80
- python3 eval_042/evluation.py --model_name "Intel/Qwen2-57B-A14B-Instruct-int4-inc" --eval_bs 16 --tasks lambada_openai,hellaswag,piqa,winogrande,truthfulqa_mc1,openbookqa,boolq,arc_easy,arc_challenge,mmlu,gsm8k,cmmlu,ceval-valid --trust_remote_code
81
- ```
82
-
83
- | Metric | BF16 | INT4-AutoRound | [official GPTQ](https://huggingface.co/Qwen/Qwen2-57B-A14B-Instruct-GPTQ-Int4) |
84
- | :---------------------- | :------ | :-------------- | :------------------------------------------------------------ |
85
- | Avg | 0.7040 | 0.7043 | 0.6990 |
86
- | mmlu | 0.7438 | 0.7408 | 0.7409 |
87
- | cmmlu | 0.8505 | 0.8448 | 0.8475 |
88
- | ceval-valid | 0.8767 | 0.8611 | 0.8507 |
89
- | gsm8k 5 shots (strict) | 0.7627 | 0.7657 | 0.7597 |
90
- | lambada_openai | 0.7452 | 0.7444 | 0.7524 |
91
- | hellaswag | 0.6517 | 0.6475 | 0.6471 |
92
- | winogrande | 0.7245 | 0.7285 | 0.7198 |
93
- | piqa | 0.8058 | 0.8058 | 0.8041 |
94
- | truthfulqa_mc1 | 0.4345 | 0.4321 | 0.4272 |
95
- | openbookqa | 0.3400 | 0.3560 | 0.3300 |
96
- | boolq | 0.8835 | 0.8844 | 0.8810 |
97
- | arc_easy | 0.8035 | 0.8051 | 0.8001 |
98
- | arc_challenge | 0.5299 | 0.5392 | 0.5265 |
99
-
100
-
101
-
102
- ## Reproduce
103
-
104
- Here is the sample command to reproduce the model.
105
-
106
- ```bash
107
- git clone https://github.com/intel/auto-round
108
- cd auto-round/examples/language-modeling
109
- pip install -r requirements.txt
110
- python3 main.py \
111
- --model_name Qwen/Qwen2-57B-A14B-Instruct \
112
- --device 0 \
113
- --group_size 128 \
114
- --nsamples 512 \
115
- --bits 4 \
116
- --iter 1000 \
117
- --disable_eval \
118
- --fp_layers "shared_expert_gate,gate" \
119
- --deployment_device 'auto_round' \
120
- --output_dir "./tmp_autoround"
121
- ```
122
-
123
- we found the output of model.layers.3.mlp.shared_expert.down_proj could be up to ~50k if adding chat template and will cause some backend like exllamav2 oeverflow. so after quantizing the model, please manually add this to config.json
124
-
125
- ~~~bash
126
- "extra_config": {
127
- "model.layers.3.mlp.shared_expert.down_proj": {
128
- "clip": true
129
- },
130
- }
131
-
132
- ~~~
133
-
134
-
135
-
136
- ## Ethical Considerations and Limitations
137
-
138
- The model can produce factually incorrect output, and should not be relied on to produce factually accurate information. Because of the limitations of the pretrained model and the finetuning datasets, it is possible that this model could generate lewd, biased or otherwise offensive outputs.
139
-
140
- Therefore, before deploying any applications of the model, developers should perform safety testing.
141
-
142
- ## Caveats and Recommendations
143
-
144
- Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model.
145
-
146
- Here are a couple of useful links to learn more about Intel's AI software:
147
-
148
- * Intel Neural Compressor [link](https://github.com/intel/neural-compressor)
149
- * Intel Extension for Transformers [link](https://github.com/intel/intel-extension-for-transformers)
150
-
151
- ## Disclaimer
152
-
153
- The license on this model does not constitute legal advice. We are not responsible for the actions of third parties who use this model. Please consult an attorney before using this model for commercial purposes.
154
-
155
-
156
-
157
- ## Cite
158
-
159
- @article{cheng2023optimize, title={Optimize weight rounding via signed gradient descent for the quantization of llms}, author={Cheng, Wenhua and Zhang, Weiwei and Shen, Haihao and Cai, Yiyang and He, Xin and Lv, Kaokao and Liu, Yi}, journal={arXiv preprint arXiv:2309.05516}, year={2023} }
160
-
161
- [arxiv](https://arxiv.org/abs/2309.05516) [github](https://github.com/intel/auto-round)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
config.json CHANGED
@@ -25,375 +25,241 @@
25
  "output_router_logits": false,
26
  "quantization_config": {
27
  "amp": true,
28
- "autoround_version": "0.3.0.dev",
29
- "backend": "auto_round:exllamav2",
30
  "bits": 4,
 
31
  "data_type": "int",
32
- "dataset": "NeelNanda/pile-10k",
33
  "enable_minmax_tuning": true,
 
34
  "enable_quanted_input": true,
35
- "extra_config": {
36
- "model.layers.3.mlp.shared_expert.down_proj": {
37
- "clip": true
38
- },
39
- "model.layers.0.mlp.gate": {
40
- "bits": 32,
41
- "data_type": "bfloat",
42
- "group_size": null,
43
- "sym": null
44
- },
45
- "model.layers.0.mlp.shared_expert_gate": {
46
- "bits": 32,
47
- "data_type": "bfloat",
48
- "group_size": null,
49
- "sym": null
50
- },
51
- "model.layers.1.mlp.gate": {
52
- "bits": 32,
53
- "data_type": "bfloat",
54
- "group_size": null,
55
- "sym": null
56
- },
57
- "model.layers.1.mlp.shared_expert_gate": {
58
- "bits": 32,
59
- "data_type": "bfloat",
60
- "group_size": null,
61
- "sym": null
62
- },
63
- "model.layers.10.mlp.gate": {
64
- "bits": 32,
65
- "data_type": "bfloat",
66
- "group_size": null,
67
- "sym": null
68
- },
69
- "model.layers.10.mlp.shared_expert_gate": {
70
- "bits": 32,
71
- "data_type": "bfloat",
72
- "group_size": null,
73
- "sym": null
74
- },
75
- "model.layers.11.mlp.gate": {
76
- "bits": 32,
77
- "data_type": "bfloat",
78
- "group_size": null,
79
- "sym": null
80
- },
81
- "model.layers.11.mlp.shared_expert_gate": {
82
- "bits": 32,
83
- "data_type": "bfloat",
84
- "group_size": null,
85
- "sym": null
86
- },
87
- "model.layers.12.mlp.gate": {
88
- "bits": 32,
89
- "data_type": "bfloat",
90
- "group_size": null,
91
- "sym": null
92
- },
93
- "model.layers.12.mlp.shared_expert_gate": {
94
- "bits": 32,
95
- "data_type": "bfloat",
96
- "group_size": null,
97
- "sym": null
98
- },
99
- "model.layers.13.mlp.gate": {
100
- "bits": 32,
101
- "data_type": "bfloat",
102
- "group_size": null,
103
- "sym": null
104
- },
105
- "model.layers.13.mlp.shared_expert_gate": {
106
- "bits": 32,
107
- "data_type": "bfloat",
108
- "group_size": null,
109
- "sym": null
110
- },
111
- "model.layers.14.mlp.gate": {
112
- "bits": 32,
113
- "data_type": "bfloat",
114
- "group_size": null,
115
- "sym": null
116
- },
117
- "model.layers.14.mlp.shared_expert_gate": {
118
- "bits": 32,
119
- "data_type": "bfloat",
120
- "group_size": null,
121
- "sym": null
122
- },
123
- "model.layers.15.mlp.gate": {
124
- "bits": 32,
125
- "data_type": "bfloat",
126
- "group_size": null,
127
- "sym": null
128
- },
129
- "model.layers.15.mlp.shared_expert_gate": {
130
- "bits": 32,
131
- "data_type": "bfloat",
132
- "group_size": null,
133
- "sym": null
134
- },
135
- "model.layers.16.mlp.gate": {
136
- "bits": 32,
137
- "data_type": "bfloat",
138
- "group_size": null,
139
- "sym": null
140
- },
141
- "model.layers.16.mlp.shared_expert_gate": {
142
- "bits": 32,
143
- "data_type": "bfloat",
144
- "group_size": null,
145
- "sym": null
146
- },
147
- "model.layers.17.mlp.gate": {
148
- "bits": 32,
149
- "data_type": "bfloat",
150
- "group_size": null,
151
- "sym": null
152
- },
153
- "model.layers.17.mlp.shared_expert_gate": {
154
- "bits": 32,
155
- "data_type": "bfloat",
156
- "group_size": null,
157
- "sym": null
158
- },
159
- "model.layers.18.mlp.gate": {
160
- "bits": 32,
161
- "data_type": "bfloat",
162
- "group_size": null,
163
- "sym": null
164
- },
165
- "model.layers.18.mlp.shared_expert_gate": {
166
- "bits": 32,
167
- "data_type": "bfloat",
168
- "group_size": null,
169
- "sym": null
170
- },
171
- "model.layers.19.mlp.gate": {
172
- "bits": 32,
173
- "data_type": "bfloat",
174
- "group_size": null,
175
- "sym": null
176
- },
177
- "model.layers.19.mlp.shared_expert_gate": {
178
- "bits": 32,
179
- "data_type": "bfloat",
180
- "group_size": null,
181
- "sym": null
182
- },
183
- "model.layers.2.mlp.gate": {
184
- "bits": 32,
185
- "data_type": "bfloat",
186
- "group_size": null,
187
- "sym": null
188
- },
189
- "model.layers.2.mlp.shared_expert_gate": {
190
- "bits": 32,
191
- "data_type": "bfloat",
192
- "group_size": null,
193
- "sym": null
194
- },
195
- "model.layers.20.mlp.gate": {
196
- "bits": 32,
197
- "data_type": "bfloat",
198
- "group_size": null,
199
- "sym": null
200
- },
201
- "model.layers.20.mlp.shared_expert_gate": {
202
- "bits": 32,
203
- "data_type": "bfloat",
204
- "group_size": null,
205
- "sym": null
206
- },
207
- "model.layers.21.mlp.gate": {
208
- "bits": 32,
209
- "data_type": "bfloat",
210
- "group_size": null,
211
- "sym": null
212
- },
213
- "model.layers.21.mlp.shared_expert_gate": {
214
- "bits": 32,
215
- "data_type": "bfloat",
216
- "group_size": null,
217
- "sym": null
218
- },
219
- "model.layers.22.mlp.gate": {
220
- "bits": 32,
221
- "data_type": "bfloat",
222
- "group_size": null,
223
- "sym": null
224
- },
225
- "model.layers.22.mlp.shared_expert_gate": {
226
- "bits": 32,
227
- "data_type": "bfloat",
228
- "group_size": null,
229
- "sym": null
230
- },
231
- "model.layers.23.mlp.gate": {
232
- "bits": 32,
233
- "data_type": "bfloat",
234
- "group_size": null,
235
- "sym": null
236
- },
237
- "model.layers.23.mlp.shared_expert_gate": {
238
- "bits": 32,
239
- "data_type": "bfloat",
240
- "group_size": null,
241
- "sym": null
242
- },
243
- "model.layers.24.mlp.gate": {
244
- "bits": 32,
245
- "data_type": "bfloat",
246
- "group_size": null,
247
- "sym": null
248
- },
249
- "model.layers.24.mlp.shared_expert_gate": {
250
- "bits": 32,
251
- "data_type": "bfloat",
252
- "group_size": null,
253
- "sym": null
254
- },
255
- "model.layers.25.mlp.gate": {
256
- "bits": 32,
257
- "data_type": "bfloat",
258
- "group_size": null,
259
- "sym": null
260
- },
261
- "model.layers.25.mlp.shared_expert_gate": {
262
- "bits": 32,
263
- "data_type": "bfloat",
264
- "group_size": null,
265
- "sym": null
266
- },
267
- "model.layers.26.mlp.gate": {
268
- "bits": 32,
269
- "data_type": "bfloat",
270
- "group_size": null,
271
- "sym": null
272
- },
273
- "model.layers.26.mlp.shared_expert_gate": {
274
- "bits": 32,
275
- "data_type": "bfloat",
276
- "group_size": null,
277
- "sym": null
278
- },
279
- "model.layers.27.mlp.gate": {
280
- "bits": 32,
281
- "data_type": "bfloat",
282
- "group_size": null,
283
- "sym": null
284
- },
285
- "model.layers.27.mlp.shared_expert_gate": {
286
- "bits": 32,
287
- "data_type": "bfloat",
288
- "group_size": null,
289
- "sym": null
290
- },
291
- "model.layers.3.mlp.gate": {
292
- "bits": 32,
293
- "data_type": "bfloat",
294
- "group_size": null,
295
- "sym": null
296
- },
297
- "model.layers.3.mlp.shared_expert_gate": {
298
- "bits": 32,
299
- "data_type": "bfloat",
300
- "group_size": null,
301
- "sym": null
302
- },
303
- "model.layers.4.mlp.gate": {
304
- "bits": 32,
305
- "data_type": "bfloat",
306
- "group_size": null,
307
- "sym": null
308
- },
309
- "model.layers.4.mlp.shared_expert_gate": {
310
- "bits": 32,
311
- "data_type": "bfloat",
312
- "group_size": null,
313
- "sym": null
314
- },
315
- "model.layers.5.mlp.gate": {
316
- "bits": 32,
317
- "data_type": "bfloat",
318
- "group_size": null,
319
- "sym": null
320
- },
321
- "model.layers.5.mlp.shared_expert_gate": {
322
- "bits": 32,
323
- "data_type": "bfloat",
324
- "group_size": null,
325
- "sym": null
326
- },
327
- "model.layers.6.mlp.gate": {
328
- "bits": 32,
329
- "data_type": "bfloat",
330
- "group_size": null,
331
- "sym": null
332
- },
333
- "model.layers.6.mlp.shared_expert_gate": {
334
- "bits": 32,
335
- "data_type": "bfloat",
336
- "group_size": null,
337
- "sym": null
338
- },
339
- "model.layers.7.mlp.gate": {
340
- "bits": 32,
341
- "data_type": "bfloat",
342
- "group_size": null,
343
- "sym": null
344
- },
345
- "model.layers.7.mlp.shared_expert_gate": {
346
- "bits": 32,
347
- "data_type": "bfloat",
348
- "group_size": null,
349
- "sym": null
350
- },
351
- "model.layers.8.mlp.gate": {
352
- "bits": 32,
353
- "data_type": "bfloat",
354
- "group_size": null,
355
- "sym": null
356
- },
357
- "model.layers.8.mlp.shared_expert_gate": {
358
- "bits": 32,
359
- "data_type": "bfloat",
360
- "group_size": null,
361
- "sym": null
362
- },
363
- "model.layers.9.mlp.gate": {
364
- "bits": 32,
365
- "data_type": "bfloat",
366
- "group_size": null,
367
- "sym": null
368
- },
369
- "model.layers.9.mlp.shared_expert_gate": {
370
- "bits": 32,
371
- "data_type": "bfloat",
372
- "group_size": null,
373
- "sym": null
374
- }
375
- },
376
  "gradient_accumulate_steps": 1,
377
  "group_size": 128,
378
  "iters": 1000,
379
  "low_gpu_mem_usage": false,
380
  "lr": 0.001,
381
  "minmax_lr": 0.001,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
382
  "nsamples": 512,
383
- "quant_method": "intel/auto-round",
 
384
  "scale_dtype": "torch.float16",
385
  "seqlen": 2048,
386
- "sym": false,
387
- "train_bs": 8
 
388
  },
389
  "rms_norm_eps": 1e-06,
390
  "rope_theta": 1000000.0,
391
  "router_aux_loss_coef": 0.001,
392
  "shared_expert_intermediate_size": 20480,
393
- "sliding_window": 65536,
394
  "tie_word_embeddings": false,
395
- "torch_dtype": "bfloat16",
396
- "transformers_version": "4.41.1",
397
  "use_cache": true,
398
  "use_sliding_window": false,
399
  "vocab_size": 151936
 
25
  "output_router_logits": false,
26
  "quantization_config": {
27
  "amp": true,
28
+ "autoround_version": "0.3.1.dev",
 
29
  "bits": 4,
30
+ "damp_percent": 0.01,
31
  "data_type": "int",
32
+ "desc_act": false,
33
  "enable_minmax_tuning": true,
34
+ "enable_norm_bias_tuning": false,
35
  "enable_quanted_input": true,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
36
  "gradient_accumulate_steps": 1,
37
  "group_size": 128,
38
  "iters": 1000,
39
  "low_gpu_mem_usage": false,
40
  "lr": 0.001,
41
  "minmax_lr": 0.001,
42
+ "modules_in_block_to_quantize": [
43
+ [
44
+ "self_attn.q_proj",
45
+ "self_attn.k_proj",
46
+ "self_attn.v_proj",
47
+ "self_attn.o_proj",
48
+ "mlp.gate",
49
+ "mlp.experts.0.gate_proj",
50
+ "mlp.experts.0.up_proj",
51
+ "mlp.experts.0.down_proj",
52
+ "mlp.experts.1.gate_proj",
53
+ "mlp.experts.1.up_proj",
54
+ "mlp.experts.1.down_proj",
55
+ "mlp.experts.2.gate_proj",
56
+ "mlp.experts.2.up_proj",
57
+ "mlp.experts.2.down_proj",
58
+ "mlp.experts.3.gate_proj",
59
+ "mlp.experts.3.up_proj",
60
+ "mlp.experts.3.down_proj",
61
+ "mlp.experts.4.gate_proj",
62
+ "mlp.experts.4.up_proj",
63
+ "mlp.experts.4.down_proj",
64
+ "mlp.experts.5.gate_proj",
65
+ "mlp.experts.5.up_proj",
66
+ "mlp.experts.5.down_proj",
67
+ "mlp.experts.6.gate_proj",
68
+ "mlp.experts.6.up_proj",
69
+ "mlp.experts.6.down_proj",
70
+ "mlp.experts.7.gate_proj",
71
+ "mlp.experts.7.up_proj",
72
+ "mlp.experts.7.down_proj",
73
+ "mlp.experts.8.gate_proj",
74
+ "mlp.experts.8.up_proj",
75
+ "mlp.experts.8.down_proj",
76
+ "mlp.experts.9.gate_proj",
77
+ "mlp.experts.9.up_proj",
78
+ "mlp.experts.9.down_proj",
79
+ "mlp.experts.10.gate_proj",
80
+ "mlp.experts.10.up_proj",
81
+ "mlp.experts.10.down_proj",
82
+ "mlp.experts.11.gate_proj",
83
+ "mlp.experts.11.up_proj",
84
+ "mlp.experts.11.down_proj",
85
+ "mlp.experts.12.gate_proj",
86
+ "mlp.experts.12.up_proj",
87
+ "mlp.experts.12.down_proj",
88
+ "mlp.experts.13.gate_proj",
89
+ "mlp.experts.13.up_proj",
90
+ "mlp.experts.13.down_proj",
91
+ "mlp.experts.14.gate_proj",
92
+ "mlp.experts.14.up_proj",
93
+ "mlp.experts.14.down_proj",
94
+ "mlp.experts.15.gate_proj",
95
+ "mlp.experts.15.up_proj",
96
+ "mlp.experts.15.down_proj",
97
+ "mlp.experts.16.gate_proj",
98
+ "mlp.experts.16.up_proj",
99
+ "mlp.experts.16.down_proj",
100
+ "mlp.experts.17.gate_proj",
101
+ "mlp.experts.17.up_proj",
102
+ "mlp.experts.17.down_proj",
103
+ "mlp.experts.18.gate_proj",
104
+ "mlp.experts.18.up_proj",
105
+ "mlp.experts.18.down_proj",
106
+ "mlp.experts.19.gate_proj",
107
+ "mlp.experts.19.up_proj",
108
+ "mlp.experts.19.down_proj",
109
+ "mlp.experts.20.gate_proj",
110
+ "mlp.experts.20.up_proj",
111
+ "mlp.experts.20.down_proj",
112
+ "mlp.experts.21.gate_proj",
113
+ "mlp.experts.21.up_proj",
114
+ "mlp.experts.21.down_proj",
115
+ "mlp.experts.22.gate_proj",
116
+ "mlp.experts.22.up_proj",
117
+ "mlp.experts.22.down_proj",
118
+ "mlp.experts.23.gate_proj",
119
+ "mlp.experts.23.up_proj",
120
+ "mlp.experts.23.down_proj",
121
+ "mlp.experts.24.gate_proj",
122
+ "mlp.experts.24.up_proj",
123
+ "mlp.experts.24.down_proj",
124
+ "mlp.experts.25.gate_proj",
125
+ "mlp.experts.25.up_proj",
126
+ "mlp.experts.25.down_proj",
127
+ "mlp.experts.26.gate_proj",
128
+ "mlp.experts.26.up_proj",
129
+ "mlp.experts.26.down_proj",
130
+ "mlp.experts.27.gate_proj",
131
+ "mlp.experts.27.up_proj",
132
+ "mlp.experts.27.down_proj",
133
+ "mlp.experts.28.gate_proj",
134
+ "mlp.experts.28.up_proj",
135
+ "mlp.experts.28.down_proj",
136
+ "mlp.experts.29.gate_proj",
137
+ "mlp.experts.29.up_proj",
138
+ "mlp.experts.29.down_proj",
139
+ "mlp.experts.30.gate_proj",
140
+ "mlp.experts.30.up_proj",
141
+ "mlp.experts.30.down_proj",
142
+ "mlp.experts.31.gate_proj",
143
+ "mlp.experts.31.up_proj",
144
+ "mlp.experts.31.down_proj",
145
+ "mlp.experts.32.gate_proj",
146
+ "mlp.experts.32.up_proj",
147
+ "mlp.experts.32.down_proj",
148
+ "mlp.experts.33.gate_proj",
149
+ "mlp.experts.33.up_proj",
150
+ "mlp.experts.33.down_proj",
151
+ "mlp.experts.34.gate_proj",
152
+ "mlp.experts.34.up_proj",
153
+ "mlp.experts.34.down_proj",
154
+ "mlp.experts.35.gate_proj",
155
+ "mlp.experts.35.up_proj",
156
+ "mlp.experts.35.down_proj",
157
+ "mlp.experts.36.gate_proj",
158
+ "mlp.experts.36.up_proj",
159
+ "mlp.experts.36.down_proj",
160
+ "mlp.experts.37.gate_proj",
161
+ "mlp.experts.37.up_proj",
162
+ "mlp.experts.37.down_proj",
163
+ "mlp.experts.38.gate_proj",
164
+ "mlp.experts.38.up_proj",
165
+ "mlp.experts.38.down_proj",
166
+ "mlp.experts.39.gate_proj",
167
+ "mlp.experts.39.up_proj",
168
+ "mlp.experts.39.down_proj",
169
+ "mlp.experts.40.gate_proj",
170
+ "mlp.experts.40.up_proj",
171
+ "mlp.experts.40.down_proj",
172
+ "mlp.experts.41.gate_proj",
173
+ "mlp.experts.41.up_proj",
174
+ "mlp.experts.41.down_proj",
175
+ "mlp.experts.42.gate_proj",
176
+ "mlp.experts.42.up_proj",
177
+ "mlp.experts.42.down_proj",
178
+ "mlp.experts.43.gate_proj",
179
+ "mlp.experts.43.up_proj",
180
+ "mlp.experts.43.down_proj",
181
+ "mlp.experts.44.gate_proj",
182
+ "mlp.experts.44.up_proj",
183
+ "mlp.experts.44.down_proj",
184
+ "mlp.experts.45.gate_proj",
185
+ "mlp.experts.45.up_proj",
186
+ "mlp.experts.45.down_proj",
187
+ "mlp.experts.46.gate_proj",
188
+ "mlp.experts.46.up_proj",
189
+ "mlp.experts.46.down_proj",
190
+ "mlp.experts.47.gate_proj",
191
+ "mlp.experts.47.up_proj",
192
+ "mlp.experts.47.down_proj",
193
+ "mlp.experts.48.gate_proj",
194
+ "mlp.experts.48.up_proj",
195
+ "mlp.experts.48.down_proj",
196
+ "mlp.experts.49.gate_proj",
197
+ "mlp.experts.49.up_proj",
198
+ "mlp.experts.49.down_proj",
199
+ "mlp.experts.50.gate_proj",
200
+ "mlp.experts.50.up_proj",
201
+ "mlp.experts.50.down_proj",
202
+ "mlp.experts.51.gate_proj",
203
+ "mlp.experts.51.up_proj",
204
+ "mlp.experts.51.down_proj",
205
+ "mlp.experts.52.gate_proj",
206
+ "mlp.experts.52.up_proj",
207
+ "mlp.experts.52.down_proj",
208
+ "mlp.experts.53.gate_proj",
209
+ "mlp.experts.53.up_proj",
210
+ "mlp.experts.53.down_proj",
211
+ "mlp.experts.54.gate_proj",
212
+ "mlp.experts.54.up_proj",
213
+ "mlp.experts.54.down_proj",
214
+ "mlp.experts.55.gate_proj",
215
+ "mlp.experts.55.up_proj",
216
+ "mlp.experts.55.down_proj",
217
+ "mlp.experts.56.gate_proj",
218
+ "mlp.experts.56.up_proj",
219
+ "mlp.experts.56.down_proj",
220
+ "mlp.experts.57.gate_proj",
221
+ "mlp.experts.57.up_proj",
222
+ "mlp.experts.57.down_proj",
223
+ "mlp.experts.58.gate_proj",
224
+ "mlp.experts.58.up_proj",
225
+ "mlp.experts.58.down_proj",
226
+ "mlp.experts.59.gate_proj",
227
+ "mlp.experts.59.up_proj",
228
+ "mlp.experts.59.down_proj",
229
+ "mlp.experts.60.gate_proj",
230
+ "mlp.experts.60.up_proj",
231
+ "mlp.experts.60.down_proj",
232
+ "mlp.experts.61.gate_proj",
233
+ "mlp.experts.61.up_proj",
234
+ "mlp.experts.61.down_proj",
235
+ "mlp.experts.62.gate_proj",
236
+ "mlp.experts.62.up_proj",
237
+ "mlp.experts.62.down_proj",
238
+ "mlp.experts.63.gate_proj",
239
+ "mlp.experts.63.up_proj",
240
+ "mlp.experts.63.down_proj",
241
+ "mlp.shared_expert.gate_proj",
242
+ "mlp.shared_expert.up_proj",
243
+ "mlp.shared_expert.down_proj"
244
+ ]
245
+ ],
246
  "nsamples": 512,
247
+ "quant_block_list": null,
248
+ "quant_method": "gptq",
249
  "scale_dtype": "torch.float16",
250
  "seqlen": 2048,
251
+ "sym": true,
252
+ "train_bs": 8,
253
+ "true_sequential": false
254
  },
255
  "rms_norm_eps": 1e-06,
256
  "rope_theta": 1000000.0,
257
  "router_aux_loss_coef": 0.001,
258
  "shared_expert_intermediate_size": 20480,
259
+ "sliding_window": null,
260
  "tie_word_embeddings": false,
261
+ "torch_dtype": "float16",
262
+ "transformers_version": "4.44.2",
263
  "use_cache": true,
264
  "use_sliding_window": false,
265
  "vocab_size": 151936
generation_config.json CHANGED
@@ -10,5 +10,5 @@
10
  "temperature": 0.7,
11
  "top_k": 20,
12
  "top_p": 0.8,
13
- "transformers_version": "4.41.1"
14
  }
 
10
  "temperature": 0.7,
11
  "top_k": 20,
12
  "top_p": 0.8,
13
+ "transformers_version": "4.44.2"
14
  }
model-00001-of-00007.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:167fc297862ecd2e0033de4c717668a83a6904b31c14025c6d96298f08706e4b
3
- size 4999673648
 
 
 
 
model-00002-of-00007.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:2e484a47e41edba04660cf6862d0ee5c2cb1d8614968c35b5ec7aa1cbe7681b3
3
- size 4996756752
 
 
 
 
model-00003-of-00007.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:77db3c5e4a435e4f834011974388b3f8d177e83537c5acd5764d06db3c592462
3
- size 4996759400
 
 
 
 
model-00005-of-00007.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c825fa9b74f5e7f55362b085315c8db3c82e4679e0faae23c07eb1ffff8eed3c
3
- size 5000290584
 
 
 
 
model-00006-of-00007.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:79105d05c0ea1507176ced998d5598cb3b4f4a7678abea09d87cd87e682120b6
3
- size 4996760576
 
 
 
 
model-00007-of-00007.safetensors DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c1d6e7a53c719cb3c8938fbaad0e638d1f5149a40bcc3103539004ef994eb072
3
- size 1538228952
 
 
 
 
model-00004-of-00007.safetensors → model.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:74f499b3d5cc607ee82042f2de2bc4ace38bd51e0b947b702ac52b39e91563b8
3
- size 4996760664
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d15f21e20a022b32b6099710f772e88b8586b50af88871fb44e2422664109fa2
3
+ size 31475221720
model.safetensors.index.json DELETED
The diff for this file is too large to render. See raw diff
 
quantization_config.json DELETED
@@ -1,360 +0,0 @@
1
- {
2
- "bits": 4,
3
- "group_size": 128,
4
- "sym": false,
5
- "data_type": "int",
6
- "enable_quanted_input": true,
7
- "enable_minmax_tuning": true,
8
- "seqlen": 2048,
9
- "train_bs": 8,
10
- "scale_dtype": "torch.float16",
11
- "lr": 0.001,
12
- "minmax_lr": 0.001,
13
- "gradient_accumulate_steps": 1,
14
- "iters": 1000,
15
- "amp": true,
16
- "nsamples": 512,
17
- "low_gpu_mem_usage": false,
18
- "dataset": "NeelNanda/pile-10k",
19
- "autoround_version": "0.3.0.dev",
20
- "quant_method": "intel/auto-round",
21
- "backend": "auto_round:exllamav2",
22
- "extra_config": {
23
- "model.layers.0.mlp.shared_expert_gate": {
24
- "data_type": "bfloat",
25
- "bits": 32,
26
- "group_size": null,
27
- "sym": null
28
- },
29
- "model.layers.1.mlp.shared_expert_gate": {
30
- "data_type": "bfloat",
31
- "bits": 32,
32
- "group_size": null,
33
- "sym": null
34
- },
35
- "model.layers.2.mlp.shared_expert_gate": {
36
- "data_type": "bfloat",
37
- "bits": 32,
38
- "group_size": null,
39
- "sym": null
40
- },
41
- "model.layers.3.mlp.shared_expert_gate": {
42
- "data_type": "bfloat",
43
- "bits": 32,
44
- "group_size": null,
45
- "sym": null
46
- },
47
- "model.layers.4.mlp.shared_expert_gate": {
48
- "data_type": "bfloat",
49
- "bits": 32,
50
- "group_size": null,
51
- "sym": null
52
- },
53
- "model.layers.5.mlp.shared_expert_gate": {
54
- "data_type": "bfloat",
55
- "bits": 32,
56
- "group_size": null,
57
- "sym": null
58
- },
59
- "model.layers.6.mlp.shared_expert_gate": {
60
- "data_type": "bfloat",
61
- "bits": 32,
62
- "group_size": null,
63
- "sym": null
64
- },
65
- "model.layers.7.mlp.shared_expert_gate": {
66
- "data_type": "bfloat",
67
- "bits": 32,
68
- "group_size": null,
69
- "sym": null
70
- },
71
- "model.layers.8.mlp.shared_expert_gate": {
72
- "data_type": "bfloat",
73
- "bits": 32,
74
- "group_size": null,
75
- "sym": null
76
- },
77
- "model.layers.9.mlp.shared_expert_gate": {
78
- "data_type": "bfloat",
79
- "bits": 32,
80
- "group_size": null,
81
- "sym": null
82
- },
83
- "model.layers.10.mlp.shared_expert_gate": {
84
- "data_type": "bfloat",
85
- "bits": 32,
86
- "group_size": null,
87
- "sym": null
88
- },
89
- "model.layers.11.mlp.shared_expert_gate": {
90
- "data_type": "bfloat",
91
- "bits": 32,
92
- "group_size": null,
93
- "sym": null
94
- },
95
- "model.layers.12.mlp.shared_expert_gate": {
96
- "data_type": "bfloat",
97
- "bits": 32,
98
- "group_size": null,
99
- "sym": null
100
- },
101
- "model.layers.13.mlp.shared_expert_gate": {
102
- "data_type": "bfloat",
103
- "bits": 32,
104
- "group_size": null,
105
- "sym": null
106
- },
107
- "model.layers.14.mlp.shared_expert_gate": {
108
- "data_type": "bfloat",
109
- "bits": 32,
110
- "group_size": null,
111
- "sym": null
112
- },
113
- "model.layers.15.mlp.shared_expert_gate": {
114
- "data_type": "bfloat",
115
- "bits": 32,
116
- "group_size": null,
117
- "sym": null
118
- },
119
- "model.layers.16.mlp.shared_expert_gate": {
120
- "data_type": "bfloat",
121
- "bits": 32,
122
- "group_size": null,
123
- "sym": null
124
- },
125
- "model.layers.17.mlp.shared_expert_gate": {
126
- "data_type": "bfloat",
127
- "bits": 32,
128
- "group_size": null,
129
- "sym": null
130
- },
131
- "model.layers.18.mlp.shared_expert_gate": {
132
- "data_type": "bfloat",
133
- "bits": 32,
134
- "group_size": null,
135
- "sym": null
136
- },
137
- "model.layers.19.mlp.shared_expert_gate": {
138
- "data_type": "bfloat",
139
- "bits": 32,
140
- "group_size": null,
141
- "sym": null
142
- },
143
- "model.layers.20.mlp.shared_expert_gate": {
144
- "data_type": "bfloat",
145
- "bits": 32,
146
- "group_size": null,
147
- "sym": null
148
- },
149
- "model.layers.21.mlp.shared_expert_gate": {
150
- "data_type": "bfloat",
151
- "bits": 32,
152
- "group_size": null,
153
- "sym": null
154
- },
155
- "model.layers.22.mlp.shared_expert_gate": {
156
- "data_type": "bfloat",
157
- "bits": 32,
158
- "group_size": null,
159
- "sym": null
160
- },
161
- "model.layers.23.mlp.shared_expert_gate": {
162
- "data_type": "bfloat",
163
- "bits": 32,
164
- "group_size": null,
165
- "sym": null
166
- },
167
- "model.layers.24.mlp.shared_expert_gate": {
168
- "data_type": "bfloat",
169
- "bits": 32,
170
- "group_size": null,
171
- "sym": null
172
- },
173
- "model.layers.25.mlp.shared_expert_gate": {
174
- "data_type": "bfloat",
175
- "bits": 32,
176
- "group_size": null,
177
- "sym": null
178
- },
179
- "model.layers.26.mlp.shared_expert_gate": {
180
- "data_type": "bfloat",
181
- "bits": 32,
182
- "group_size": null,
183
- "sym": null
184
- },
185
- "model.layers.27.mlp.shared_expert_gate": {
186
- "data_type": "bfloat",
187
- "bits": 32,
188
- "group_size": null,
189
- "sym": null
190
- },
191
- "model.layers.0.mlp.gate": {
192
- "data_type": "bfloat",
193
- "bits": 32,
194
- "group_size": null,
195
- "sym": null
196
- },
197
- "model.layers.1.mlp.gate": {
198
- "data_type": "bfloat",
199
- "bits": 32,
200
- "group_size": null,
201
- "sym": null
202
- },
203
- "model.layers.2.mlp.gate": {
204
- "data_type": "bfloat",
205
- "bits": 32,
206
- "group_size": null,
207
- "sym": null
208
- },
209
- "model.layers.3.mlp.gate": {
210
- "data_type": "bfloat",
211
- "bits": 32,
212
- "group_size": null,
213
- "sym": null
214
- },
215
- "model.layers.4.mlp.gate": {
216
- "data_type": "bfloat",
217
- "bits": 32,
218
- "group_size": null,
219
- "sym": null
220
- },
221
- "model.layers.5.mlp.gate": {
222
- "data_type": "bfloat",
223
- "bits": 32,
224
- "group_size": null,
225
- "sym": null
226
- },
227
- "model.layers.6.mlp.gate": {
228
- "data_type": "bfloat",
229
- "bits": 32,
230
- "group_size": null,
231
- "sym": null
232
- },
233
- "model.layers.7.mlp.gate": {
234
- "data_type": "bfloat",
235
- "bits": 32,
236
- "group_size": null,
237
- "sym": null
238
- },
239
- "model.layers.8.mlp.gate": {
240
- "data_type": "bfloat",
241
- "bits": 32,
242
- "group_size": null,
243
- "sym": null
244
- },
245
- "model.layers.9.mlp.gate": {
246
- "data_type": "bfloat",
247
- "bits": 32,
248
- "group_size": null,
249
- "sym": null
250
- },
251
- "model.layers.10.mlp.gate": {
252
- "data_type": "bfloat",
253
- "bits": 32,
254
- "group_size": null,
255
- "sym": null
256
- },
257
- "model.layers.11.mlp.gate": {
258
- "data_type": "bfloat",
259
- "bits": 32,
260
- "group_size": null,
261
- "sym": null
262
- },
263
- "model.layers.12.mlp.gate": {
264
- "data_type": "bfloat",
265
- "bits": 32,
266
- "group_size": null,
267
- "sym": null
268
- },
269
- "model.layers.13.mlp.gate": {
270
- "data_type": "bfloat",
271
- "bits": 32,
272
- "group_size": null,
273
- "sym": null
274
- },
275
- "model.layers.14.mlp.gate": {
276
- "data_type": "bfloat",
277
- "bits": 32,
278
- "group_size": null,
279
- "sym": null
280
- },
281
- "model.layers.15.mlp.gate": {
282
- "data_type": "bfloat",
283
- "bits": 32,
284
- "group_size": null,
285
- "sym": null
286
- },
287
- "model.layers.16.mlp.gate": {
288
- "data_type": "bfloat",
289
- "bits": 32,
290
- "group_size": null,
291
- "sym": null
292
- },
293
- "model.layers.17.mlp.gate": {
294
- "data_type": "bfloat",
295
- "bits": 32,
296
- "group_size": null,
297
- "sym": null
298
- },
299
- "model.layers.18.mlp.gate": {
300
- "data_type": "bfloat",
301
- "bits": 32,
302
- "group_size": null,
303
- "sym": null
304
- },
305
- "model.layers.19.mlp.gate": {
306
- "data_type": "bfloat",
307
- "bits": 32,
308
- "group_size": null,
309
- "sym": null
310
- },
311
- "model.layers.20.mlp.gate": {
312
- "data_type": "bfloat",
313
- "bits": 32,
314
- "group_size": null,
315
- "sym": null
316
- },
317
- "model.layers.21.mlp.gate": {
318
- "data_type": "bfloat",
319
- "bits": 32,
320
- "group_size": null,
321
- "sym": null
322
- },
323
- "model.layers.22.mlp.gate": {
324
- "data_type": "bfloat",
325
- "bits": 32,
326
- "group_size": null,
327
- "sym": null
328
- },
329
- "model.layers.23.mlp.gate": {
330
- "data_type": "bfloat",
331
- "bits": 32,
332
- "group_size": null,
333
- "sym": null
334
- },
335
- "model.layers.24.mlp.gate": {
336
- "data_type": "bfloat",
337
- "bits": 32,
338
- "group_size": null,
339
- "sym": null
340
- },
341
- "model.layers.25.mlp.gate": {
342
- "data_type": "bfloat",
343
- "bits": 32,
344
- "group_size": null,
345
- "sym": null
346
- },
347
- "model.layers.26.mlp.gate": {
348
- "data_type": "bfloat",
349
- "bits": 32,
350
- "group_size": null,
351
- "sym": null
352
- },
353
- "model.layers.27.mlp.gate": {
354
- "data_type": "bfloat",
355
- "bits": 32,
356
- "group_size": null,
357
- "sym": null
358
- }
359
- }
360
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
quantize_config.json ADDED
@@ -0,0 +1,229 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bits": 4,
3
+ "group_size": 128,
4
+ "sym": true,
5
+ "data_type": "int",
6
+ "enable_quanted_input": true,
7
+ "enable_minmax_tuning": true,
8
+ "seqlen": 2048,
9
+ "train_bs": 8,
10
+ "scale_dtype": "torch.float16",
11
+ "lr": 0.001,
12
+ "minmax_lr": 0.001,
13
+ "gradient_accumulate_steps": 1,
14
+ "iters": 1000,
15
+ "amp": true,
16
+ "nsamples": 512,
17
+ "low_gpu_mem_usage": false,
18
+ "quant_block_list": null,
19
+ "enable_norm_bias_tuning": false,
20
+ "autoround_version": "0.3.1.dev",
21
+ "quant_method": "gptq",
22
+ "desc_act": false,
23
+ "true_sequential": false,
24
+ "damp_percent": 0.01,
25
+ "modules_in_block_to_quantize": [
26
+ [
27
+ "self_attn.q_proj",
28
+ "self_attn.k_proj",
29
+ "self_attn.v_proj",
30
+ "self_attn.o_proj",
31
+ "mlp.gate",
32
+ "mlp.experts.0.gate_proj",
33
+ "mlp.experts.0.up_proj",
34
+ "mlp.experts.0.down_proj",
35
+ "mlp.experts.1.gate_proj",
36
+ "mlp.experts.1.up_proj",
37
+ "mlp.experts.1.down_proj",
38
+ "mlp.experts.2.gate_proj",
39
+ "mlp.experts.2.up_proj",
40
+ "mlp.experts.2.down_proj",
41
+ "mlp.experts.3.gate_proj",
42
+ "mlp.experts.3.up_proj",
43
+ "mlp.experts.3.down_proj",
44
+ "mlp.experts.4.gate_proj",
45
+ "mlp.experts.4.up_proj",
46
+ "mlp.experts.4.down_proj",
47
+ "mlp.experts.5.gate_proj",
48
+ "mlp.experts.5.up_proj",
49
+ "mlp.experts.5.down_proj",
50
+ "mlp.experts.6.gate_proj",
51
+ "mlp.experts.6.up_proj",
52
+ "mlp.experts.6.down_proj",
53
+ "mlp.experts.7.gate_proj",
54
+ "mlp.experts.7.up_proj",
55
+ "mlp.experts.7.down_proj",
56
+ "mlp.experts.8.gate_proj",
57
+ "mlp.experts.8.up_proj",
58
+ "mlp.experts.8.down_proj",
59
+ "mlp.experts.9.gate_proj",
60
+ "mlp.experts.9.up_proj",
61
+ "mlp.experts.9.down_proj",
62
+ "mlp.experts.10.gate_proj",
63
+ "mlp.experts.10.up_proj",
64
+ "mlp.experts.10.down_proj",
65
+ "mlp.experts.11.gate_proj",
66
+ "mlp.experts.11.up_proj",
67
+ "mlp.experts.11.down_proj",
68
+ "mlp.experts.12.gate_proj",
69
+ "mlp.experts.12.up_proj",
70
+ "mlp.experts.12.down_proj",
71
+ "mlp.experts.13.gate_proj",
72
+ "mlp.experts.13.up_proj",
73
+ "mlp.experts.13.down_proj",
74
+ "mlp.experts.14.gate_proj",
75
+ "mlp.experts.14.up_proj",
76
+ "mlp.experts.14.down_proj",
77
+ "mlp.experts.15.gate_proj",
78
+ "mlp.experts.15.up_proj",
79
+ "mlp.experts.15.down_proj",
80
+ "mlp.experts.16.gate_proj",
81
+ "mlp.experts.16.up_proj",
82
+ "mlp.experts.16.down_proj",
83
+ "mlp.experts.17.gate_proj",
84
+ "mlp.experts.17.up_proj",
85
+ "mlp.experts.17.down_proj",
86
+ "mlp.experts.18.gate_proj",
87
+ "mlp.experts.18.up_proj",
88
+ "mlp.experts.18.down_proj",
89
+ "mlp.experts.19.gate_proj",
90
+ "mlp.experts.19.up_proj",
91
+ "mlp.experts.19.down_proj",
92
+ "mlp.experts.20.gate_proj",
93
+ "mlp.experts.20.up_proj",
94
+ "mlp.experts.20.down_proj",
95
+ "mlp.experts.21.gate_proj",
96
+ "mlp.experts.21.up_proj",
97
+ "mlp.experts.21.down_proj",
98
+ "mlp.experts.22.gate_proj",
99
+ "mlp.experts.22.up_proj",
100
+ "mlp.experts.22.down_proj",
101
+ "mlp.experts.23.gate_proj",
102
+ "mlp.experts.23.up_proj",
103
+ "mlp.experts.23.down_proj",
104
+ "mlp.experts.24.gate_proj",
105
+ "mlp.experts.24.up_proj",
106
+ "mlp.experts.24.down_proj",
107
+ "mlp.experts.25.gate_proj",
108
+ "mlp.experts.25.up_proj",
109
+ "mlp.experts.25.down_proj",
110
+ "mlp.experts.26.gate_proj",
111
+ "mlp.experts.26.up_proj",
112
+ "mlp.experts.26.down_proj",
113
+ "mlp.experts.27.gate_proj",
114
+ "mlp.experts.27.up_proj",
115
+ "mlp.experts.27.down_proj",
116
+ "mlp.experts.28.gate_proj",
117
+ "mlp.experts.28.up_proj",
118
+ "mlp.experts.28.down_proj",
119
+ "mlp.experts.29.gate_proj",
120
+ "mlp.experts.29.up_proj",
121
+ "mlp.experts.29.down_proj",
122
+ "mlp.experts.30.gate_proj",
123
+ "mlp.experts.30.up_proj",
124
+ "mlp.experts.30.down_proj",
125
+ "mlp.experts.31.gate_proj",
126
+ "mlp.experts.31.up_proj",
127
+ "mlp.experts.31.down_proj",
128
+ "mlp.experts.32.gate_proj",
129
+ "mlp.experts.32.up_proj",
130
+ "mlp.experts.32.down_proj",
131
+ "mlp.experts.33.gate_proj",
132
+ "mlp.experts.33.up_proj",
133
+ "mlp.experts.33.down_proj",
134
+ "mlp.experts.34.gate_proj",
135
+ "mlp.experts.34.up_proj",
136
+ "mlp.experts.34.down_proj",
137
+ "mlp.experts.35.gate_proj",
138
+ "mlp.experts.35.up_proj",
139
+ "mlp.experts.35.down_proj",
140
+ "mlp.experts.36.gate_proj",
141
+ "mlp.experts.36.up_proj",
142
+ "mlp.experts.36.down_proj",
143
+ "mlp.experts.37.gate_proj",
144
+ "mlp.experts.37.up_proj",
145
+ "mlp.experts.37.down_proj",
146
+ "mlp.experts.38.gate_proj",
147
+ "mlp.experts.38.up_proj",
148
+ "mlp.experts.38.down_proj",
149
+ "mlp.experts.39.gate_proj",
150
+ "mlp.experts.39.up_proj",
151
+ "mlp.experts.39.down_proj",
152
+ "mlp.experts.40.gate_proj",
153
+ "mlp.experts.40.up_proj",
154
+ "mlp.experts.40.down_proj",
155
+ "mlp.experts.41.gate_proj",
156
+ "mlp.experts.41.up_proj",
157
+ "mlp.experts.41.down_proj",
158
+ "mlp.experts.42.gate_proj",
159
+ "mlp.experts.42.up_proj",
160
+ "mlp.experts.42.down_proj",
161
+ "mlp.experts.43.gate_proj",
162
+ "mlp.experts.43.up_proj",
163
+ "mlp.experts.43.down_proj",
164
+ "mlp.experts.44.gate_proj",
165
+ "mlp.experts.44.up_proj",
166
+ "mlp.experts.44.down_proj",
167
+ "mlp.experts.45.gate_proj",
168
+ "mlp.experts.45.up_proj",
169
+ "mlp.experts.45.down_proj",
170
+ "mlp.experts.46.gate_proj",
171
+ "mlp.experts.46.up_proj",
172
+ "mlp.experts.46.down_proj",
173
+ "mlp.experts.47.gate_proj",
174
+ "mlp.experts.47.up_proj",
175
+ "mlp.experts.47.down_proj",
176
+ "mlp.experts.48.gate_proj",
177
+ "mlp.experts.48.up_proj",
178
+ "mlp.experts.48.down_proj",
179
+ "mlp.experts.49.gate_proj",
180
+ "mlp.experts.49.up_proj",
181
+ "mlp.experts.49.down_proj",
182
+ "mlp.experts.50.gate_proj",
183
+ "mlp.experts.50.up_proj",
184
+ "mlp.experts.50.down_proj",
185
+ "mlp.experts.51.gate_proj",
186
+ "mlp.experts.51.up_proj",
187
+ "mlp.experts.51.down_proj",
188
+ "mlp.experts.52.gate_proj",
189
+ "mlp.experts.52.up_proj",
190
+ "mlp.experts.52.down_proj",
191
+ "mlp.experts.53.gate_proj",
192
+ "mlp.experts.53.up_proj",
193
+ "mlp.experts.53.down_proj",
194
+ "mlp.experts.54.gate_proj",
195
+ "mlp.experts.54.up_proj",
196
+ "mlp.experts.54.down_proj",
197
+ "mlp.experts.55.gate_proj",
198
+ "mlp.experts.55.up_proj",
199
+ "mlp.experts.55.down_proj",
200
+ "mlp.experts.56.gate_proj",
201
+ "mlp.experts.56.up_proj",
202
+ "mlp.experts.56.down_proj",
203
+ "mlp.experts.57.gate_proj",
204
+ "mlp.experts.57.up_proj",
205
+ "mlp.experts.57.down_proj",
206
+ "mlp.experts.58.gate_proj",
207
+ "mlp.experts.58.up_proj",
208
+ "mlp.experts.58.down_proj",
209
+ "mlp.experts.59.gate_proj",
210
+ "mlp.experts.59.up_proj",
211
+ "mlp.experts.59.down_proj",
212
+ "mlp.experts.60.gate_proj",
213
+ "mlp.experts.60.up_proj",
214
+ "mlp.experts.60.down_proj",
215
+ "mlp.experts.61.gate_proj",
216
+ "mlp.experts.61.up_proj",
217
+ "mlp.experts.61.down_proj",
218
+ "mlp.experts.62.gate_proj",
219
+ "mlp.experts.62.up_proj",
220
+ "mlp.experts.62.down_proj",
221
+ "mlp.experts.63.gate_proj",
222
+ "mlp.experts.63.up_proj",
223
+ "mlp.experts.63.down_proj",
224
+ "mlp.shared_expert.gate_proj",
225
+ "mlp.shared_expert.up_proj",
226
+ "mlp.shared_expert.down_proj"
227
+ ]
228
+ ]
229
+ }