danielhanchen commited on
Commit
0dc6cb1
·
verified ·
1 Parent(s): 31624b0

Sync hf_quant_config.json with the thinkingmachines release

Browse files
Files changed (1) hide show
  1. hf_quant_config.json +578 -0
hf_quant_config.json ADDED
@@ -0,0 +1,578 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "quantization": {
3
+ "quant_algo": "NVFP4",
4
+ "kv_cache_quant_algo": "none",
5
+ "exclude_modules": [
6
+ "model.audio.encoder",
7
+ "model.audio.final_norm",
8
+ "model.llm.embed",
9
+ "model.llm.embed_norm",
10
+ "model.llm.norm",
11
+ "model.llm.unembed",
12
+ "model.visual.final_norm",
13
+ "model.visual.layers.linear_0",
14
+ "model.visual.layers.linear_1",
15
+ "model.visual.layers.linear_2",
16
+ "model.visual.layers.linear_3",
17
+ "model.visual.layers.norm_0",
18
+ "model.visual.layers.norm_1",
19
+ "model.visual.layers.norm_2",
20
+ "model.llm.layers.0.attn",
21
+ "model.llm.layers.0.attn_norm",
22
+ "model.llm.layers.0.attn_sconv",
23
+ "model.llm.layers.0.mlp.global_scale",
24
+ "model.llm.layers.0.mlp.w13_dn",
25
+ "model.llm.layers.0.mlp.w2_md",
26
+ "model.llm.layers.0.mlp_norm",
27
+ "model.llm.layers.0.mlp_sconv",
28
+ "model.llm.layers.1.attn",
29
+ "model.llm.layers.1.attn_norm",
30
+ "model.llm.layers.1.attn_sconv",
31
+ "model.llm.layers.1.mlp.global_scale",
32
+ "model.llm.layers.1.mlp.w13_dn",
33
+ "model.llm.layers.1.mlp.w2_md",
34
+ "model.llm.layers.1.mlp_norm",
35
+ "model.llm.layers.1.mlp_sconv",
36
+ "model.llm.layers.10.attn",
37
+ "model.llm.layers.10.attn_norm",
38
+ "model.llm.layers.10.attn_sconv",
39
+ "model.llm.layers.10.mlp.gate",
40
+ "model.llm.layers.10.mlp.shared_experts",
41
+ "model.llm.layers.10.mlp_norm",
42
+ "model.llm.layers.10.mlp_sconv",
43
+ "model.llm.layers.11.attn",
44
+ "model.llm.layers.11.attn_norm",
45
+ "model.llm.layers.11.attn_sconv",
46
+ "model.llm.layers.11.mlp.gate",
47
+ "model.llm.layers.11.mlp.shared_experts",
48
+ "model.llm.layers.11.mlp_norm",
49
+ "model.llm.layers.11.mlp_sconv",
50
+ "model.llm.layers.12.attn",
51
+ "model.llm.layers.12.attn_norm",
52
+ "model.llm.layers.12.attn_sconv",
53
+ "model.llm.layers.12.mlp.gate",
54
+ "model.llm.layers.12.mlp.shared_experts",
55
+ "model.llm.layers.12.mlp_norm",
56
+ "model.llm.layers.12.mlp_sconv",
57
+ "model.llm.layers.13.attn",
58
+ "model.llm.layers.13.attn_norm",
59
+ "model.llm.layers.13.attn_sconv",
60
+ "model.llm.layers.13.mlp.gate",
61
+ "model.llm.layers.13.mlp.shared_experts",
62
+ "model.llm.layers.13.mlp_norm",
63
+ "model.llm.layers.13.mlp_sconv",
64
+ "model.llm.layers.14.attn",
65
+ "model.llm.layers.14.attn_norm",
66
+ "model.llm.layers.14.attn_sconv",
67
+ "model.llm.layers.14.mlp.gate",
68
+ "model.llm.layers.14.mlp.shared_experts",
69
+ "model.llm.layers.14.mlp_norm",
70
+ "model.llm.layers.14.mlp_sconv",
71
+ "model.llm.layers.15.attn",
72
+ "model.llm.layers.15.attn_norm",
73
+ "model.llm.layers.15.attn_sconv",
74
+ "model.llm.layers.15.mlp.gate",
75
+ "model.llm.layers.15.mlp.shared_experts",
76
+ "model.llm.layers.15.mlp_norm",
77
+ "model.llm.layers.15.mlp_sconv",
78
+ "model.llm.layers.16.attn",
79
+ "model.llm.layers.16.attn_norm",
80
+ "model.llm.layers.16.attn_sconv",
81
+ "model.llm.layers.16.mlp.gate",
82
+ "model.llm.layers.16.mlp.shared_experts",
83
+ "model.llm.layers.16.mlp_norm",
84
+ "model.llm.layers.16.mlp_sconv",
85
+ "model.llm.layers.17.attn",
86
+ "model.llm.layers.17.attn_norm",
87
+ "model.llm.layers.17.attn_sconv",
88
+ "model.llm.layers.17.mlp.gate",
89
+ "model.llm.layers.17.mlp.shared_experts",
90
+ "model.llm.layers.17.mlp_norm",
91
+ "model.llm.layers.17.mlp_sconv",
92
+ "model.llm.layers.18.attn",
93
+ "model.llm.layers.18.attn_norm",
94
+ "model.llm.layers.18.attn_sconv",
95
+ "model.llm.layers.18.mlp.gate",
96
+ "model.llm.layers.18.mlp.shared_experts",
97
+ "model.llm.layers.18.mlp_norm",
98
+ "model.llm.layers.18.mlp_sconv",
99
+ "model.llm.layers.19.attn",
100
+ "model.llm.layers.19.attn_norm",
101
+ "model.llm.layers.19.attn_sconv",
102
+ "model.llm.layers.19.mlp.gate",
103
+ "model.llm.layers.19.mlp.shared_experts",
104
+ "model.llm.layers.19.mlp_norm",
105
+ "model.llm.layers.19.mlp_sconv",
106
+ "model.llm.layers.2.attn",
107
+ "model.llm.layers.2.attn_norm",
108
+ "model.llm.layers.2.attn_sconv",
109
+ "model.llm.layers.2.mlp.experts",
110
+ "model.llm.layers.2.mlp.gate",
111
+ "model.llm.layers.2.mlp.shared_experts",
112
+ "model.llm.layers.2.mlp_norm",
113
+ "model.llm.layers.2.mlp_sconv",
114
+ "model.llm.layers.20.attn",
115
+ "model.llm.layers.20.attn_norm",
116
+ "model.llm.layers.20.attn_sconv",
117
+ "model.llm.layers.20.mlp.gate",
118
+ "model.llm.layers.20.mlp.shared_experts",
119
+ "model.llm.layers.20.mlp_norm",
120
+ "model.llm.layers.20.mlp_sconv",
121
+ "model.llm.layers.21.attn",
122
+ "model.llm.layers.21.attn_norm",
123
+ "model.llm.layers.21.attn_sconv",
124
+ "model.llm.layers.21.mlp.gate",
125
+ "model.llm.layers.21.mlp.shared_experts",
126
+ "model.llm.layers.21.mlp_norm",
127
+ "model.llm.layers.21.mlp_sconv",
128
+ "model.llm.layers.22.attn",
129
+ "model.llm.layers.22.attn_norm",
130
+ "model.llm.layers.22.attn_sconv",
131
+ "model.llm.layers.22.mlp.gate",
132
+ "model.llm.layers.22.mlp.shared_experts",
133
+ "model.llm.layers.22.mlp_norm",
134
+ "model.llm.layers.22.mlp_sconv",
135
+ "model.llm.layers.23.attn",
136
+ "model.llm.layers.23.attn_norm",
137
+ "model.llm.layers.23.attn_sconv",
138
+ "model.llm.layers.23.mlp.gate",
139
+ "model.llm.layers.23.mlp.shared_experts",
140
+ "model.llm.layers.23.mlp_norm",
141
+ "model.llm.layers.23.mlp_sconv",
142
+ "model.llm.layers.24.attn",
143
+ "model.llm.layers.24.attn_norm",
144
+ "model.llm.layers.24.attn_sconv",
145
+ "model.llm.layers.24.mlp.gate",
146
+ "model.llm.layers.24.mlp.shared_experts",
147
+ "model.llm.layers.24.mlp_norm",
148
+ "model.llm.layers.24.mlp_sconv",
149
+ "model.llm.layers.25.attn",
150
+ "model.llm.layers.25.attn_norm",
151
+ "model.llm.layers.25.attn_sconv",
152
+ "model.llm.layers.25.mlp.gate",
153
+ "model.llm.layers.25.mlp.shared_experts",
154
+ "model.llm.layers.25.mlp_norm",
155
+ "model.llm.layers.25.mlp_sconv",
156
+ "model.llm.layers.26.attn",
157
+ "model.llm.layers.26.attn_norm",
158
+ "model.llm.layers.26.attn_sconv",
159
+ "model.llm.layers.26.mlp.gate",
160
+ "model.llm.layers.26.mlp.shared_experts",
161
+ "model.llm.layers.26.mlp_norm",
162
+ "model.llm.layers.26.mlp_sconv",
163
+ "model.llm.layers.27.attn",
164
+ "model.llm.layers.27.attn_norm",
165
+ "model.llm.layers.27.attn_sconv",
166
+ "model.llm.layers.27.mlp.gate",
167
+ "model.llm.layers.27.mlp.shared_experts",
168
+ "model.llm.layers.27.mlp_norm",
169
+ "model.llm.layers.27.mlp_sconv",
170
+ "model.llm.layers.28.attn",
171
+ "model.llm.layers.28.attn_norm",
172
+ "model.llm.layers.28.attn_sconv",
173
+ "model.llm.layers.28.mlp.gate",
174
+ "model.llm.layers.28.mlp.shared_experts",
175
+ "model.llm.layers.28.mlp_norm",
176
+ "model.llm.layers.28.mlp_sconv",
177
+ "model.llm.layers.29.attn",
178
+ "model.llm.layers.29.attn_norm",
179
+ "model.llm.layers.29.attn_sconv",
180
+ "model.llm.layers.29.mlp.gate",
181
+ "model.llm.layers.29.mlp.shared_experts",
182
+ "model.llm.layers.29.mlp_norm",
183
+ "model.llm.layers.29.mlp_sconv",
184
+ "model.llm.layers.3.attn",
185
+ "model.llm.layers.3.attn_norm",
186
+ "model.llm.layers.3.attn_sconv",
187
+ "model.llm.layers.3.mlp.gate",
188
+ "model.llm.layers.3.mlp.shared_experts",
189
+ "model.llm.layers.3.mlp_norm",
190
+ "model.llm.layers.3.mlp_sconv",
191
+ "model.llm.layers.30.attn",
192
+ "model.llm.layers.30.attn_norm",
193
+ "model.llm.layers.30.attn_sconv",
194
+ "model.llm.layers.30.mlp.gate",
195
+ "model.llm.layers.30.mlp.shared_experts",
196
+ "model.llm.layers.30.mlp_norm",
197
+ "model.llm.layers.30.mlp_sconv",
198
+ "model.llm.layers.31.attn",
199
+ "model.llm.layers.31.attn_norm",
200
+ "model.llm.layers.31.attn_sconv",
201
+ "model.llm.layers.31.mlp.gate",
202
+ "model.llm.layers.31.mlp.shared_experts",
203
+ "model.llm.layers.31.mlp_norm",
204
+ "model.llm.layers.31.mlp_sconv",
205
+ "model.llm.layers.32.attn",
206
+ "model.llm.layers.32.attn_norm",
207
+ "model.llm.layers.32.attn_sconv",
208
+ "model.llm.layers.32.mlp.gate",
209
+ "model.llm.layers.32.mlp.shared_experts",
210
+ "model.llm.layers.32.mlp_norm",
211
+ "model.llm.layers.32.mlp_sconv",
212
+ "model.llm.layers.33.attn",
213
+ "model.llm.layers.33.attn_norm",
214
+ "model.llm.layers.33.attn_sconv",
215
+ "model.llm.layers.33.mlp.gate",
216
+ "model.llm.layers.33.mlp.shared_experts",
217
+ "model.llm.layers.33.mlp_norm",
218
+ "model.llm.layers.33.mlp_sconv",
219
+ "model.llm.layers.34.attn",
220
+ "model.llm.layers.34.attn_norm",
221
+ "model.llm.layers.34.attn_sconv",
222
+ "model.llm.layers.34.mlp.gate",
223
+ "model.llm.layers.34.mlp.shared_experts",
224
+ "model.llm.layers.34.mlp_norm",
225
+ "model.llm.layers.34.mlp_sconv",
226
+ "model.llm.layers.35.attn",
227
+ "model.llm.layers.35.attn_norm",
228
+ "model.llm.layers.35.attn_sconv",
229
+ "model.llm.layers.35.mlp.gate",
230
+ "model.llm.layers.35.mlp.shared_experts",
231
+ "model.llm.layers.35.mlp_norm",
232
+ "model.llm.layers.35.mlp_sconv",
233
+ "model.llm.layers.36.attn",
234
+ "model.llm.layers.36.attn_norm",
235
+ "model.llm.layers.36.attn_sconv",
236
+ "model.llm.layers.36.mlp.gate",
237
+ "model.llm.layers.36.mlp.shared_experts",
238
+ "model.llm.layers.36.mlp_norm",
239
+ "model.llm.layers.36.mlp_sconv",
240
+ "model.llm.layers.37.attn",
241
+ "model.llm.layers.37.attn_norm",
242
+ "model.llm.layers.37.attn_sconv",
243
+ "model.llm.layers.37.mlp.gate",
244
+ "model.llm.layers.37.mlp.shared_experts",
245
+ "model.llm.layers.37.mlp_norm",
246
+ "model.llm.layers.37.mlp_sconv",
247
+ "model.llm.layers.38.attn",
248
+ "model.llm.layers.38.attn_norm",
249
+ "model.llm.layers.38.attn_sconv",
250
+ "model.llm.layers.38.mlp.gate",
251
+ "model.llm.layers.38.mlp.shared_experts",
252
+ "model.llm.layers.38.mlp_norm",
253
+ "model.llm.layers.38.mlp_sconv",
254
+ "model.llm.layers.39.attn",
255
+ "model.llm.layers.39.attn_norm",
256
+ "model.llm.layers.39.attn_sconv",
257
+ "model.llm.layers.39.mlp.gate",
258
+ "model.llm.layers.39.mlp.shared_experts",
259
+ "model.llm.layers.39.mlp_norm",
260
+ "model.llm.layers.39.mlp_sconv",
261
+ "model.llm.layers.4.attn",
262
+ "model.llm.layers.4.attn_norm",
263
+ "model.llm.layers.4.attn_sconv",
264
+ "model.llm.layers.4.mlp.gate",
265
+ "model.llm.layers.4.mlp.shared_experts",
266
+ "model.llm.layers.4.mlp_norm",
267
+ "model.llm.layers.4.mlp_sconv",
268
+ "model.llm.layers.40.attn",
269
+ "model.llm.layers.40.attn_norm",
270
+ "model.llm.layers.40.attn_sconv",
271
+ "model.llm.layers.40.mlp.gate",
272
+ "model.llm.layers.40.mlp.shared_experts",
273
+ "model.llm.layers.40.mlp_norm",
274
+ "model.llm.layers.40.mlp_sconv",
275
+ "model.llm.layers.41.attn",
276
+ "model.llm.layers.41.attn_norm",
277
+ "model.llm.layers.41.attn_sconv",
278
+ "model.llm.layers.41.mlp.gate",
279
+ "model.llm.layers.41.mlp.shared_experts",
280
+ "model.llm.layers.41.mlp_norm",
281
+ "model.llm.layers.41.mlp_sconv",
282
+ "model.llm.layers.42.attn",
283
+ "model.llm.layers.42.attn_norm",
284
+ "model.llm.layers.42.attn_sconv",
285
+ "model.llm.layers.42.mlp.gate",
286
+ "model.llm.layers.42.mlp.shared_experts",
287
+ "model.llm.layers.42.mlp_norm",
288
+ "model.llm.layers.42.mlp_sconv",
289
+ "model.llm.layers.43.attn",
290
+ "model.llm.layers.43.attn_norm",
291
+ "model.llm.layers.43.attn_sconv",
292
+ "model.llm.layers.43.mlp.gate",
293
+ "model.llm.layers.43.mlp.shared_experts",
294
+ "model.llm.layers.43.mlp_norm",
295
+ "model.llm.layers.43.mlp_sconv",
296
+ "model.llm.layers.44.attn",
297
+ "model.llm.layers.44.attn_norm",
298
+ "model.llm.layers.44.attn_sconv",
299
+ "model.llm.layers.44.mlp.gate",
300
+ "model.llm.layers.44.mlp.shared_experts",
301
+ "model.llm.layers.44.mlp_norm",
302
+ "model.llm.layers.44.mlp_sconv",
303
+ "model.llm.layers.45.attn",
304
+ "model.llm.layers.45.attn_norm",
305
+ "model.llm.layers.45.attn_sconv",
306
+ "model.llm.layers.45.mlp.gate",
307
+ "model.llm.layers.45.mlp.shared_experts",
308
+ "model.llm.layers.45.mlp_norm",
309
+ "model.llm.layers.45.mlp_sconv",
310
+ "model.llm.layers.46.attn",
311
+ "model.llm.layers.46.attn_norm",
312
+ "model.llm.layers.46.attn_sconv",
313
+ "model.llm.layers.46.mlp.gate",
314
+ "model.llm.layers.46.mlp.shared_experts",
315
+ "model.llm.layers.46.mlp_norm",
316
+ "model.llm.layers.46.mlp_sconv",
317
+ "model.llm.layers.47.attn",
318
+ "model.llm.layers.47.attn_norm",
319
+ "model.llm.layers.47.attn_sconv",
320
+ "model.llm.layers.47.mlp.gate",
321
+ "model.llm.layers.47.mlp.shared_experts",
322
+ "model.llm.layers.47.mlp_norm",
323
+ "model.llm.layers.47.mlp_sconv",
324
+ "model.llm.layers.48.attn",
325
+ "model.llm.layers.48.attn_norm",
326
+ "model.llm.layers.48.attn_sconv",
327
+ "model.llm.layers.48.mlp.gate",
328
+ "model.llm.layers.48.mlp.shared_experts",
329
+ "model.llm.layers.48.mlp_norm",
330
+ "model.llm.layers.48.mlp_sconv",
331
+ "model.llm.layers.49.attn",
332
+ "model.llm.layers.49.attn_norm",
333
+ "model.llm.layers.49.attn_sconv",
334
+ "model.llm.layers.49.mlp.gate",
335
+ "model.llm.layers.49.mlp.shared_experts",
336
+ "model.llm.layers.49.mlp_norm",
337
+ "model.llm.layers.49.mlp_sconv",
338
+ "model.llm.layers.5.attn",
339
+ "model.llm.layers.5.attn_norm",
340
+ "model.llm.layers.5.attn_sconv",
341
+ "model.llm.layers.5.mlp.gate",
342
+ "model.llm.layers.5.mlp.shared_experts",
343
+ "model.llm.layers.5.mlp_norm",
344
+ "model.llm.layers.5.mlp_sconv",
345
+ "model.llm.layers.50.attn",
346
+ "model.llm.layers.50.attn_norm",
347
+ "model.llm.layers.50.attn_sconv",
348
+ "model.llm.layers.50.mlp.gate",
349
+ "model.llm.layers.50.mlp.shared_experts",
350
+ "model.llm.layers.50.mlp_norm",
351
+ "model.llm.layers.50.mlp_sconv",
352
+ "model.llm.layers.51.attn",
353
+ "model.llm.layers.51.attn_norm",
354
+ "model.llm.layers.51.attn_sconv",
355
+ "model.llm.layers.51.mlp.gate",
356
+ "model.llm.layers.51.mlp.shared_experts",
357
+ "model.llm.layers.51.mlp_norm",
358
+ "model.llm.layers.51.mlp_sconv",
359
+ "model.llm.layers.52.attn",
360
+ "model.llm.layers.52.attn_norm",
361
+ "model.llm.layers.52.attn_sconv",
362
+ "model.llm.layers.52.mlp.gate",
363
+ "model.llm.layers.52.mlp.shared_experts",
364
+ "model.llm.layers.52.mlp_norm",
365
+ "model.llm.layers.52.mlp_sconv",
366
+ "model.llm.layers.53.attn",
367
+ "model.llm.layers.53.attn_norm",
368
+ "model.llm.layers.53.attn_sconv",
369
+ "model.llm.layers.53.mlp.gate",
370
+ "model.llm.layers.53.mlp.shared_experts",
371
+ "model.llm.layers.53.mlp_norm",
372
+ "model.llm.layers.53.mlp_sconv",
373
+ "model.llm.layers.54.attn",
374
+ "model.llm.layers.54.attn_norm",
375
+ "model.llm.layers.54.attn_sconv",
376
+ "model.llm.layers.54.mlp.gate",
377
+ "model.llm.layers.54.mlp.shared_experts",
378
+ "model.llm.layers.54.mlp_norm",
379
+ "model.llm.layers.54.mlp_sconv",
380
+ "model.llm.layers.55.attn",
381
+ "model.llm.layers.55.attn_norm",
382
+ "model.llm.layers.55.attn_sconv",
383
+ "model.llm.layers.55.mlp.gate",
384
+ "model.llm.layers.55.mlp.shared_experts",
385
+ "model.llm.layers.55.mlp_norm",
386
+ "model.llm.layers.55.mlp_sconv",
387
+ "model.llm.layers.56.attn",
388
+ "model.llm.layers.56.attn_norm",
389
+ "model.llm.layers.56.attn_sconv",
390
+ "model.llm.layers.56.mlp.gate",
391
+ "model.llm.layers.56.mlp.shared_experts",
392
+ "model.llm.layers.56.mlp_norm",
393
+ "model.llm.layers.56.mlp_sconv",
394
+ "model.llm.layers.57.attn",
395
+ "model.llm.layers.57.attn_norm",
396
+ "model.llm.layers.57.attn_sconv",
397
+ "model.llm.layers.57.mlp.gate",
398
+ "model.llm.layers.57.mlp.shared_experts",
399
+ "model.llm.layers.57.mlp_norm",
400
+ "model.llm.layers.57.mlp_sconv",
401
+ "model.llm.layers.58.attn",
402
+ "model.llm.layers.58.attn_norm",
403
+ "model.llm.layers.58.attn_sconv",
404
+ "model.llm.layers.58.mlp.gate",
405
+ "model.llm.layers.58.mlp.shared_experts",
406
+ "model.llm.layers.58.mlp_norm",
407
+ "model.llm.layers.58.mlp_sconv",
408
+ "model.llm.layers.59.attn",
409
+ "model.llm.layers.59.attn_norm",
410
+ "model.llm.layers.59.attn_sconv",
411
+ "model.llm.layers.59.mlp.gate",
412
+ "model.llm.layers.59.mlp.shared_experts",
413
+ "model.llm.layers.59.mlp_norm",
414
+ "model.llm.layers.59.mlp_sconv",
415
+ "model.llm.layers.6.attn",
416
+ "model.llm.layers.6.attn_norm",
417
+ "model.llm.layers.6.attn_sconv",
418
+ "model.llm.layers.6.mlp.gate",
419
+ "model.llm.layers.6.mlp.shared_experts",
420
+ "model.llm.layers.6.mlp_norm",
421
+ "model.llm.layers.6.mlp_sconv",
422
+ "model.llm.layers.60.attn",
423
+ "model.llm.layers.60.attn_norm",
424
+ "model.llm.layers.60.attn_sconv",
425
+ "model.llm.layers.60.mlp.gate",
426
+ "model.llm.layers.60.mlp.shared_experts",
427
+ "model.llm.layers.60.mlp_norm",
428
+ "model.llm.layers.60.mlp_sconv",
429
+ "model.llm.layers.61.attn",
430
+ "model.llm.layers.61.attn_norm",
431
+ "model.llm.layers.61.attn_sconv",
432
+ "model.llm.layers.61.mlp.gate",
433
+ "model.llm.layers.61.mlp.shared_experts",
434
+ "model.llm.layers.61.mlp_norm",
435
+ "model.llm.layers.61.mlp_sconv",
436
+ "model.llm.layers.62.attn",
437
+ "model.llm.layers.62.attn_norm",
438
+ "model.llm.layers.62.attn_sconv",
439
+ "model.llm.layers.62.mlp.gate",
440
+ "model.llm.layers.62.mlp.shared_experts",
441
+ "model.llm.layers.62.mlp_norm",
442
+ "model.llm.layers.62.mlp_sconv",
443
+ "model.llm.layers.63.attn",
444
+ "model.llm.layers.63.attn_norm",
445
+ "model.llm.layers.63.attn_sconv",
446
+ "model.llm.layers.63.mlp.gate",
447
+ "model.llm.layers.63.mlp.shared_experts",
448
+ "model.llm.layers.63.mlp_norm",
449
+ "model.llm.layers.63.mlp_sconv",
450
+ "model.llm.layers.64.attn",
451
+ "model.llm.layers.64.attn_norm",
452
+ "model.llm.layers.64.attn_sconv",
453
+ "model.llm.layers.64.mlp.gate",
454
+ "model.llm.layers.64.mlp.shared_experts",
455
+ "model.llm.layers.64.mlp_norm",
456
+ "model.llm.layers.64.mlp_sconv",
457
+ "model.llm.layers.65.attn",
458
+ "model.llm.layers.65.attn_norm",
459
+ "model.llm.layers.65.attn_sconv",
460
+ "model.llm.layers.65.mlp.gate",
461
+ "model.llm.layers.65.mlp.shared_experts",
462
+ "model.llm.layers.65.mlp_norm",
463
+ "model.llm.layers.65.mlp_sconv",
464
+ "model.llm.layers.7.attn",
465
+ "model.llm.layers.7.attn_norm",
466
+ "model.llm.layers.7.attn_sconv",
467
+ "model.llm.layers.7.mlp.gate",
468
+ "model.llm.layers.7.mlp.shared_experts",
469
+ "model.llm.layers.7.mlp_norm",
470
+ "model.llm.layers.7.mlp_sconv",
471
+ "model.llm.layers.8.attn",
472
+ "model.llm.layers.8.attn_norm",
473
+ "model.llm.layers.8.attn_sconv",
474
+ "model.llm.layers.8.mlp.gate",
475
+ "model.llm.layers.8.mlp.shared_experts",
476
+ "model.llm.layers.8.mlp_norm",
477
+ "model.llm.layers.8.mlp_sconv",
478
+ "model.llm.layers.9.attn",
479
+ "model.llm.layers.9.attn_norm",
480
+ "model.llm.layers.9.attn_sconv",
481
+ "model.llm.layers.9.mlp.gate",
482
+ "model.llm.layers.9.mlp.shared_experts",
483
+ "model.llm.layers.9.mlp_norm",
484
+ "model.llm.layers.9.mlp_sconv"
485
+ ],
486
+ "modelopt_quant_config": {
487
+ "quant_cfg": {
488
+ "*weight_quantizer": {
489
+ "num_bits": [
490
+ 2,
491
+ 1
492
+ ],
493
+ "block_sizes": {
494
+ "-1": 16,
495
+ "type": "dynamic",
496
+ "scale_bits": [
497
+ 4,
498
+ 3
499
+ ]
500
+ },
501
+ "axis": null,
502
+ "enable": true
503
+ },
504
+ "*input_quantizer": {
505
+ "num_bits": [
506
+ 2,
507
+ 1
508
+ ],
509
+ "block_sizes": {
510
+ "-1": 16,
511
+ "type": "dynamic",
512
+ "scale_bits": [
513
+ 4,
514
+ 3
515
+ ]
516
+ },
517
+ "axis": null,
518
+ "enable": true
519
+ },
520
+ "nn.BatchNorm1d": {
521
+ "*": {
522
+ "enable": false
523
+ }
524
+ },
525
+ "nn.BatchNorm2d": {
526
+ "*": {
527
+ "enable": false
528
+ }
529
+ },
530
+ "nn.BatchNorm3d": {
531
+ "*": {
532
+ "enable": false
533
+ }
534
+ },
535
+ "nn.LeakyReLU": {
536
+ "*": {
537
+ "enable": false
538
+ }
539
+ },
540
+ "*lm_head*": {
541
+ "enable": false
542
+ },
543
+ "*proj_out.*": {
544
+ "enable": false
545
+ },
546
+ "*block_sparse_moe.gate*": {
547
+ "enable": false
548
+ },
549
+ "*router*": {
550
+ "enable": false
551
+ },
552
+ "*mlp.gate.*": {
553
+ "enable": false
554
+ },
555
+ "*mlp.shared_expert_gate.*": {
556
+ "enable": false
557
+ },
558
+ "*linear_attn.conv1d*": {
559
+ "enable": false
560
+ },
561
+ "*mixer.conv1d*": {
562
+ "enable": false
563
+ },
564
+ "*output_layer*": {
565
+ "enable": false
566
+ },
567
+ "output.*": {
568
+ "enable": false
569
+ },
570
+ "default": {
571
+ "enable": false
572
+ }
573
+ },
574
+ "algorithm": "max"
575
+ },
576
+ "group_size": 16
577
+ }
578
+ }