Jundot HF Staff commited on
Commit
a7fbd2d
·
verified ·
1 Parent(s): 9d4e68e

Record nvfp4 quantization for the expert layers in config

Browse files
Files changed (1) hide show
  1. config.json +2054 -2
config.json CHANGED
@@ -23,12 +23,2064 @@
23
  "quantization": {
24
  "group_size": 32,
25
  "bits": 8,
26
- "mode": "mxfp8"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
27
  },
28
  "quantization_config": {
29
  "group_size": 32,
30
  "bits": 8,
31
- "mode": "mxfp8"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
32
  },
33
  "text_config": {
34
  "hidden_size": 6144,
 
23
  "quantization": {
24
  "group_size": 32,
25
  "bits": 8,
26
+ "mode": "mxfp8",
27
+ "language_model.model.layers.3.block_sparse_moe.switch_mlp.down_proj": {
28
+ "group_size": 16,
29
+ "bits": 4,
30
+ "mode": "nvfp4",
31
+ "global_scale": true
32
+ },
33
+ "language_model.model.layers.3.block_sparse_moe.switch_mlp.gate_proj": {
34
+ "group_size": 16,
35
+ "bits": 4,
36
+ "mode": "nvfp4",
37
+ "global_scale": true
38
+ },
39
+ "language_model.model.layers.3.block_sparse_moe.switch_mlp.up_proj": {
40
+ "group_size": 16,
41
+ "bits": 4,
42
+ "mode": "nvfp4",
43
+ "global_scale": true
44
+ },
45
+ "language_model.model.layers.4.block_sparse_moe.switch_mlp.down_proj": {
46
+ "group_size": 16,
47
+ "bits": 4,
48
+ "mode": "nvfp4",
49
+ "global_scale": true
50
+ },
51
+ "language_model.model.layers.4.block_sparse_moe.switch_mlp.gate_proj": {
52
+ "group_size": 16,
53
+ "bits": 4,
54
+ "mode": "nvfp4",
55
+ "global_scale": true
56
+ },
57
+ "language_model.model.layers.4.block_sparse_moe.switch_mlp.up_proj": {
58
+ "group_size": 16,
59
+ "bits": 4,
60
+ "mode": "nvfp4",
61
+ "global_scale": true
62
+ },
63
+ "language_model.model.layers.5.block_sparse_moe.switch_mlp.down_proj": {
64
+ "group_size": 16,
65
+ "bits": 4,
66
+ "mode": "nvfp4",
67
+ "global_scale": true
68
+ },
69
+ "language_model.model.layers.5.block_sparse_moe.switch_mlp.gate_proj": {
70
+ "group_size": 16,
71
+ "bits": 4,
72
+ "mode": "nvfp4",
73
+ "global_scale": true
74
+ },
75
+ "language_model.model.layers.5.block_sparse_moe.switch_mlp.up_proj": {
76
+ "group_size": 16,
77
+ "bits": 4,
78
+ "mode": "nvfp4",
79
+ "global_scale": true
80
+ },
81
+ "language_model.model.layers.6.block_sparse_moe.switch_mlp.down_proj": {
82
+ "group_size": 16,
83
+ "bits": 4,
84
+ "mode": "nvfp4",
85
+ "global_scale": true
86
+ },
87
+ "language_model.model.layers.6.block_sparse_moe.switch_mlp.gate_proj": {
88
+ "group_size": 16,
89
+ "bits": 4,
90
+ "mode": "nvfp4",
91
+ "global_scale": true
92
+ },
93
+ "language_model.model.layers.6.block_sparse_moe.switch_mlp.up_proj": {
94
+ "group_size": 16,
95
+ "bits": 4,
96
+ "mode": "nvfp4",
97
+ "global_scale": true
98
+ },
99
+ "language_model.model.layers.7.block_sparse_moe.switch_mlp.down_proj": {
100
+ "group_size": 16,
101
+ "bits": 4,
102
+ "mode": "nvfp4",
103
+ "global_scale": true
104
+ },
105
+ "language_model.model.layers.7.block_sparse_moe.switch_mlp.gate_proj": {
106
+ "group_size": 16,
107
+ "bits": 4,
108
+ "mode": "nvfp4",
109
+ "global_scale": true
110
+ },
111
+ "language_model.model.layers.7.block_sparse_moe.switch_mlp.up_proj": {
112
+ "group_size": 16,
113
+ "bits": 4,
114
+ "mode": "nvfp4",
115
+ "global_scale": true
116
+ },
117
+ "language_model.model.layers.8.block_sparse_moe.switch_mlp.down_proj": {
118
+ "group_size": 16,
119
+ "bits": 4,
120
+ "mode": "nvfp4",
121
+ "global_scale": true
122
+ },
123
+ "language_model.model.layers.8.block_sparse_moe.switch_mlp.gate_proj": {
124
+ "group_size": 16,
125
+ "bits": 4,
126
+ "mode": "nvfp4",
127
+ "global_scale": true
128
+ },
129
+ "language_model.model.layers.8.block_sparse_moe.switch_mlp.up_proj": {
130
+ "group_size": 16,
131
+ "bits": 4,
132
+ "mode": "nvfp4",
133
+ "global_scale": true
134
+ },
135
+ "language_model.model.layers.9.block_sparse_moe.switch_mlp.down_proj": {
136
+ "group_size": 16,
137
+ "bits": 4,
138
+ "mode": "nvfp4",
139
+ "global_scale": true
140
+ },
141
+ "language_model.model.layers.9.block_sparse_moe.switch_mlp.gate_proj": {
142
+ "group_size": 16,
143
+ "bits": 4,
144
+ "mode": "nvfp4",
145
+ "global_scale": true
146
+ },
147
+ "language_model.model.layers.9.block_sparse_moe.switch_mlp.up_proj": {
148
+ "group_size": 16,
149
+ "bits": 4,
150
+ "mode": "nvfp4",
151
+ "global_scale": true
152
+ },
153
+ "language_model.model.layers.10.block_sparse_moe.switch_mlp.down_proj": {
154
+ "group_size": 16,
155
+ "bits": 4,
156
+ "mode": "nvfp4",
157
+ "global_scale": true
158
+ },
159
+ "language_model.model.layers.10.block_sparse_moe.switch_mlp.gate_proj": {
160
+ "group_size": 16,
161
+ "bits": 4,
162
+ "mode": "nvfp4",
163
+ "global_scale": true
164
+ },
165
+ "language_model.model.layers.10.block_sparse_moe.switch_mlp.up_proj": {
166
+ "group_size": 16,
167
+ "bits": 4,
168
+ "mode": "nvfp4",
169
+ "global_scale": true
170
+ },
171
+ "language_model.model.layers.11.block_sparse_moe.switch_mlp.down_proj": {
172
+ "group_size": 16,
173
+ "bits": 4,
174
+ "mode": "nvfp4",
175
+ "global_scale": true
176
+ },
177
+ "language_model.model.layers.11.block_sparse_moe.switch_mlp.gate_proj": {
178
+ "group_size": 16,
179
+ "bits": 4,
180
+ "mode": "nvfp4",
181
+ "global_scale": true
182
+ },
183
+ "language_model.model.layers.11.block_sparse_moe.switch_mlp.up_proj": {
184
+ "group_size": 16,
185
+ "bits": 4,
186
+ "mode": "nvfp4",
187
+ "global_scale": true
188
+ },
189
+ "language_model.model.layers.12.block_sparse_moe.switch_mlp.down_proj": {
190
+ "group_size": 16,
191
+ "bits": 4,
192
+ "mode": "nvfp4",
193
+ "global_scale": true
194
+ },
195
+ "language_model.model.layers.12.block_sparse_moe.switch_mlp.gate_proj": {
196
+ "group_size": 16,
197
+ "bits": 4,
198
+ "mode": "nvfp4",
199
+ "global_scale": true
200
+ },
201
+ "language_model.model.layers.12.block_sparse_moe.switch_mlp.up_proj": {
202
+ "group_size": 16,
203
+ "bits": 4,
204
+ "mode": "nvfp4",
205
+ "global_scale": true
206
+ },
207
+ "language_model.model.layers.13.block_sparse_moe.switch_mlp.down_proj": {
208
+ "group_size": 16,
209
+ "bits": 4,
210
+ "mode": "nvfp4",
211
+ "global_scale": true
212
+ },
213
+ "language_model.model.layers.13.block_sparse_moe.switch_mlp.gate_proj": {
214
+ "group_size": 16,
215
+ "bits": 4,
216
+ "mode": "nvfp4",
217
+ "global_scale": true
218
+ },
219
+ "language_model.model.layers.13.block_sparse_moe.switch_mlp.up_proj": {
220
+ "group_size": 16,
221
+ "bits": 4,
222
+ "mode": "nvfp4",
223
+ "global_scale": true
224
+ },
225
+ "language_model.model.layers.14.block_sparse_moe.switch_mlp.down_proj": {
226
+ "group_size": 16,
227
+ "bits": 4,
228
+ "mode": "nvfp4",
229
+ "global_scale": true
230
+ },
231
+ "language_model.model.layers.14.block_sparse_moe.switch_mlp.gate_proj": {
232
+ "group_size": 16,
233
+ "bits": 4,
234
+ "mode": "nvfp4",
235
+ "global_scale": true
236
+ },
237
+ "language_model.model.layers.14.block_sparse_moe.switch_mlp.up_proj": {
238
+ "group_size": 16,
239
+ "bits": 4,
240
+ "mode": "nvfp4",
241
+ "global_scale": true
242
+ },
243
+ "language_model.model.layers.15.block_sparse_moe.switch_mlp.down_proj": {
244
+ "group_size": 16,
245
+ "bits": 4,
246
+ "mode": "nvfp4",
247
+ "global_scale": true
248
+ },
249
+ "language_model.model.layers.15.block_sparse_moe.switch_mlp.gate_proj": {
250
+ "group_size": 16,
251
+ "bits": 4,
252
+ "mode": "nvfp4",
253
+ "global_scale": true
254
+ },
255
+ "language_model.model.layers.15.block_sparse_moe.switch_mlp.up_proj": {
256
+ "group_size": 16,
257
+ "bits": 4,
258
+ "mode": "nvfp4",
259
+ "global_scale": true
260
+ },
261
+ "language_model.model.layers.16.block_sparse_moe.switch_mlp.down_proj": {
262
+ "group_size": 16,
263
+ "bits": 4,
264
+ "mode": "nvfp4",
265
+ "global_scale": true
266
+ },
267
+ "language_model.model.layers.16.block_sparse_moe.switch_mlp.gate_proj": {
268
+ "group_size": 16,
269
+ "bits": 4,
270
+ "mode": "nvfp4",
271
+ "global_scale": true
272
+ },
273
+ "language_model.model.layers.16.block_sparse_moe.switch_mlp.up_proj": {
274
+ "group_size": 16,
275
+ "bits": 4,
276
+ "mode": "nvfp4",
277
+ "global_scale": true
278
+ },
279
+ "language_model.model.layers.17.block_sparse_moe.switch_mlp.down_proj": {
280
+ "group_size": 16,
281
+ "bits": 4,
282
+ "mode": "nvfp4",
283
+ "global_scale": true
284
+ },
285
+ "language_model.model.layers.17.block_sparse_moe.switch_mlp.gate_proj": {
286
+ "group_size": 16,
287
+ "bits": 4,
288
+ "mode": "nvfp4",
289
+ "global_scale": true
290
+ },
291
+ "language_model.model.layers.17.block_sparse_moe.switch_mlp.up_proj": {
292
+ "group_size": 16,
293
+ "bits": 4,
294
+ "mode": "nvfp4",
295
+ "global_scale": true
296
+ },
297
+ "language_model.model.layers.18.block_sparse_moe.switch_mlp.down_proj": {
298
+ "group_size": 16,
299
+ "bits": 4,
300
+ "mode": "nvfp4",
301
+ "global_scale": true
302
+ },
303
+ "language_model.model.layers.18.block_sparse_moe.switch_mlp.gate_proj": {
304
+ "group_size": 16,
305
+ "bits": 4,
306
+ "mode": "nvfp4",
307
+ "global_scale": true
308
+ },
309
+ "language_model.model.layers.18.block_sparse_moe.switch_mlp.up_proj": {
310
+ "group_size": 16,
311
+ "bits": 4,
312
+ "mode": "nvfp4",
313
+ "global_scale": true
314
+ },
315
+ "language_model.model.layers.19.block_sparse_moe.switch_mlp.down_proj": {
316
+ "group_size": 16,
317
+ "bits": 4,
318
+ "mode": "nvfp4",
319
+ "global_scale": true
320
+ },
321
+ "language_model.model.layers.19.block_sparse_moe.switch_mlp.gate_proj": {
322
+ "group_size": 16,
323
+ "bits": 4,
324
+ "mode": "nvfp4",
325
+ "global_scale": true
326
+ },
327
+ "language_model.model.layers.19.block_sparse_moe.switch_mlp.up_proj": {
328
+ "group_size": 16,
329
+ "bits": 4,
330
+ "mode": "nvfp4",
331
+ "global_scale": true
332
+ },
333
+ "language_model.model.layers.20.block_sparse_moe.switch_mlp.down_proj": {
334
+ "group_size": 16,
335
+ "bits": 4,
336
+ "mode": "nvfp4",
337
+ "global_scale": true
338
+ },
339
+ "language_model.model.layers.20.block_sparse_moe.switch_mlp.gate_proj": {
340
+ "group_size": 16,
341
+ "bits": 4,
342
+ "mode": "nvfp4",
343
+ "global_scale": true
344
+ },
345
+ "language_model.model.layers.20.block_sparse_moe.switch_mlp.up_proj": {
346
+ "group_size": 16,
347
+ "bits": 4,
348
+ "mode": "nvfp4",
349
+ "global_scale": true
350
+ },
351
+ "language_model.model.layers.21.block_sparse_moe.switch_mlp.down_proj": {
352
+ "group_size": 16,
353
+ "bits": 4,
354
+ "mode": "nvfp4",
355
+ "global_scale": true
356
+ },
357
+ "language_model.model.layers.21.block_sparse_moe.switch_mlp.gate_proj": {
358
+ "group_size": 16,
359
+ "bits": 4,
360
+ "mode": "nvfp4",
361
+ "global_scale": true
362
+ },
363
+ "language_model.model.layers.21.block_sparse_moe.switch_mlp.up_proj": {
364
+ "group_size": 16,
365
+ "bits": 4,
366
+ "mode": "nvfp4",
367
+ "global_scale": true
368
+ },
369
+ "language_model.model.layers.22.block_sparse_moe.switch_mlp.down_proj": {
370
+ "group_size": 16,
371
+ "bits": 4,
372
+ "mode": "nvfp4",
373
+ "global_scale": true
374
+ },
375
+ "language_model.model.layers.22.block_sparse_moe.switch_mlp.gate_proj": {
376
+ "group_size": 16,
377
+ "bits": 4,
378
+ "mode": "nvfp4",
379
+ "global_scale": true
380
+ },
381
+ "language_model.model.layers.22.block_sparse_moe.switch_mlp.up_proj": {
382
+ "group_size": 16,
383
+ "bits": 4,
384
+ "mode": "nvfp4",
385
+ "global_scale": true
386
+ },
387
+ "language_model.model.layers.23.block_sparse_moe.switch_mlp.down_proj": {
388
+ "group_size": 16,
389
+ "bits": 4,
390
+ "mode": "nvfp4",
391
+ "global_scale": true
392
+ },
393
+ "language_model.model.layers.23.block_sparse_moe.switch_mlp.gate_proj": {
394
+ "group_size": 16,
395
+ "bits": 4,
396
+ "mode": "nvfp4",
397
+ "global_scale": true
398
+ },
399
+ "language_model.model.layers.23.block_sparse_moe.switch_mlp.up_proj": {
400
+ "group_size": 16,
401
+ "bits": 4,
402
+ "mode": "nvfp4",
403
+ "global_scale": true
404
+ },
405
+ "language_model.model.layers.24.block_sparse_moe.switch_mlp.down_proj": {
406
+ "group_size": 16,
407
+ "bits": 4,
408
+ "mode": "nvfp4",
409
+ "global_scale": true
410
+ },
411
+ "language_model.model.layers.24.block_sparse_moe.switch_mlp.gate_proj": {
412
+ "group_size": 16,
413
+ "bits": 4,
414
+ "mode": "nvfp4",
415
+ "global_scale": true
416
+ },
417
+ "language_model.model.layers.24.block_sparse_moe.switch_mlp.up_proj": {
418
+ "group_size": 16,
419
+ "bits": 4,
420
+ "mode": "nvfp4",
421
+ "global_scale": true
422
+ },
423
+ "language_model.model.layers.25.block_sparse_moe.switch_mlp.down_proj": {
424
+ "group_size": 16,
425
+ "bits": 4,
426
+ "mode": "nvfp4",
427
+ "global_scale": true
428
+ },
429
+ "language_model.model.layers.25.block_sparse_moe.switch_mlp.gate_proj": {
430
+ "group_size": 16,
431
+ "bits": 4,
432
+ "mode": "nvfp4",
433
+ "global_scale": true
434
+ },
435
+ "language_model.model.layers.25.block_sparse_moe.switch_mlp.up_proj": {
436
+ "group_size": 16,
437
+ "bits": 4,
438
+ "mode": "nvfp4",
439
+ "global_scale": true
440
+ },
441
+ "language_model.model.layers.26.block_sparse_moe.switch_mlp.down_proj": {
442
+ "group_size": 16,
443
+ "bits": 4,
444
+ "mode": "nvfp4",
445
+ "global_scale": true
446
+ },
447
+ "language_model.model.layers.26.block_sparse_moe.switch_mlp.gate_proj": {
448
+ "group_size": 16,
449
+ "bits": 4,
450
+ "mode": "nvfp4",
451
+ "global_scale": true
452
+ },
453
+ "language_model.model.layers.26.block_sparse_moe.switch_mlp.up_proj": {
454
+ "group_size": 16,
455
+ "bits": 4,
456
+ "mode": "nvfp4",
457
+ "global_scale": true
458
+ },
459
+ "language_model.model.layers.27.block_sparse_moe.switch_mlp.down_proj": {
460
+ "group_size": 16,
461
+ "bits": 4,
462
+ "mode": "nvfp4",
463
+ "global_scale": true
464
+ },
465
+ "language_model.model.layers.27.block_sparse_moe.switch_mlp.gate_proj": {
466
+ "group_size": 16,
467
+ "bits": 4,
468
+ "mode": "nvfp4",
469
+ "global_scale": true
470
+ },
471
+ "language_model.model.layers.27.block_sparse_moe.switch_mlp.up_proj": {
472
+ "group_size": 16,
473
+ "bits": 4,
474
+ "mode": "nvfp4",
475
+ "global_scale": true
476
+ },
477
+ "language_model.model.layers.28.block_sparse_moe.switch_mlp.down_proj": {
478
+ "group_size": 16,
479
+ "bits": 4,
480
+ "mode": "nvfp4",
481
+ "global_scale": true
482
+ },
483
+ "language_model.model.layers.28.block_sparse_moe.switch_mlp.gate_proj": {
484
+ "group_size": 16,
485
+ "bits": 4,
486
+ "mode": "nvfp4",
487
+ "global_scale": true
488
+ },
489
+ "language_model.model.layers.28.block_sparse_moe.switch_mlp.up_proj": {
490
+ "group_size": 16,
491
+ "bits": 4,
492
+ "mode": "nvfp4",
493
+ "global_scale": true
494
+ },
495
+ "language_model.model.layers.29.block_sparse_moe.switch_mlp.down_proj": {
496
+ "group_size": 16,
497
+ "bits": 4,
498
+ "mode": "nvfp4",
499
+ "global_scale": true
500
+ },
501
+ "language_model.model.layers.29.block_sparse_moe.switch_mlp.gate_proj": {
502
+ "group_size": 16,
503
+ "bits": 4,
504
+ "mode": "nvfp4",
505
+ "global_scale": true
506
+ },
507
+ "language_model.model.layers.29.block_sparse_moe.switch_mlp.up_proj": {
508
+ "group_size": 16,
509
+ "bits": 4,
510
+ "mode": "nvfp4",
511
+ "global_scale": true
512
+ },
513
+ "language_model.model.layers.30.block_sparse_moe.switch_mlp.down_proj": {
514
+ "group_size": 16,
515
+ "bits": 4,
516
+ "mode": "nvfp4",
517
+ "global_scale": true
518
+ },
519
+ "language_model.model.layers.30.block_sparse_moe.switch_mlp.gate_proj": {
520
+ "group_size": 16,
521
+ "bits": 4,
522
+ "mode": "nvfp4",
523
+ "global_scale": true
524
+ },
525
+ "language_model.model.layers.30.block_sparse_moe.switch_mlp.up_proj": {
526
+ "group_size": 16,
527
+ "bits": 4,
528
+ "mode": "nvfp4",
529
+ "global_scale": true
530
+ },
531
+ "language_model.model.layers.31.block_sparse_moe.switch_mlp.down_proj": {
532
+ "group_size": 16,
533
+ "bits": 4,
534
+ "mode": "nvfp4",
535
+ "global_scale": true
536
+ },
537
+ "language_model.model.layers.31.block_sparse_moe.switch_mlp.gate_proj": {
538
+ "group_size": 16,
539
+ "bits": 4,
540
+ "mode": "nvfp4",
541
+ "global_scale": true
542
+ },
543
+ "language_model.model.layers.31.block_sparse_moe.switch_mlp.up_proj": {
544
+ "group_size": 16,
545
+ "bits": 4,
546
+ "mode": "nvfp4",
547
+ "global_scale": true
548
+ },
549
+ "language_model.model.layers.32.block_sparse_moe.switch_mlp.down_proj": {
550
+ "group_size": 16,
551
+ "bits": 4,
552
+ "mode": "nvfp4",
553
+ "global_scale": true
554
+ },
555
+ "language_model.model.layers.32.block_sparse_moe.switch_mlp.gate_proj": {
556
+ "group_size": 16,
557
+ "bits": 4,
558
+ "mode": "nvfp4",
559
+ "global_scale": true
560
+ },
561
+ "language_model.model.layers.32.block_sparse_moe.switch_mlp.up_proj": {
562
+ "group_size": 16,
563
+ "bits": 4,
564
+ "mode": "nvfp4",
565
+ "global_scale": true
566
+ },
567
+ "language_model.model.layers.33.block_sparse_moe.switch_mlp.down_proj": {
568
+ "group_size": 16,
569
+ "bits": 4,
570
+ "mode": "nvfp4",
571
+ "global_scale": true
572
+ },
573
+ "language_model.model.layers.33.block_sparse_moe.switch_mlp.gate_proj": {
574
+ "group_size": 16,
575
+ "bits": 4,
576
+ "mode": "nvfp4",
577
+ "global_scale": true
578
+ },
579
+ "language_model.model.layers.33.block_sparse_moe.switch_mlp.up_proj": {
580
+ "group_size": 16,
581
+ "bits": 4,
582
+ "mode": "nvfp4",
583
+ "global_scale": true
584
+ },
585
+ "language_model.model.layers.34.block_sparse_moe.switch_mlp.down_proj": {
586
+ "group_size": 16,
587
+ "bits": 4,
588
+ "mode": "nvfp4",
589
+ "global_scale": true
590
+ },
591
+ "language_model.model.layers.34.block_sparse_moe.switch_mlp.gate_proj": {
592
+ "group_size": 16,
593
+ "bits": 4,
594
+ "mode": "nvfp4",
595
+ "global_scale": true
596
+ },
597
+ "language_model.model.layers.34.block_sparse_moe.switch_mlp.up_proj": {
598
+ "group_size": 16,
599
+ "bits": 4,
600
+ "mode": "nvfp4",
601
+ "global_scale": true
602
+ },
603
+ "language_model.model.layers.35.block_sparse_moe.switch_mlp.down_proj": {
604
+ "group_size": 16,
605
+ "bits": 4,
606
+ "mode": "nvfp4",
607
+ "global_scale": true
608
+ },
609
+ "language_model.model.layers.35.block_sparse_moe.switch_mlp.gate_proj": {
610
+ "group_size": 16,
611
+ "bits": 4,
612
+ "mode": "nvfp4",
613
+ "global_scale": true
614
+ },
615
+ "language_model.model.layers.35.block_sparse_moe.switch_mlp.up_proj": {
616
+ "group_size": 16,
617
+ "bits": 4,
618
+ "mode": "nvfp4",
619
+ "global_scale": true
620
+ },
621
+ "language_model.model.layers.36.block_sparse_moe.switch_mlp.down_proj": {
622
+ "group_size": 16,
623
+ "bits": 4,
624
+ "mode": "nvfp4",
625
+ "global_scale": true
626
+ },
627
+ "language_model.model.layers.36.block_sparse_moe.switch_mlp.gate_proj": {
628
+ "group_size": 16,
629
+ "bits": 4,
630
+ "mode": "nvfp4",
631
+ "global_scale": true
632
+ },
633
+ "language_model.model.layers.36.block_sparse_moe.switch_mlp.up_proj": {
634
+ "group_size": 16,
635
+ "bits": 4,
636
+ "mode": "nvfp4",
637
+ "global_scale": true
638
+ },
639
+ "language_model.model.layers.37.block_sparse_moe.switch_mlp.down_proj": {
640
+ "group_size": 16,
641
+ "bits": 4,
642
+ "mode": "nvfp4",
643
+ "global_scale": true
644
+ },
645
+ "language_model.model.layers.37.block_sparse_moe.switch_mlp.gate_proj": {
646
+ "group_size": 16,
647
+ "bits": 4,
648
+ "mode": "nvfp4",
649
+ "global_scale": true
650
+ },
651
+ "language_model.model.layers.37.block_sparse_moe.switch_mlp.up_proj": {
652
+ "group_size": 16,
653
+ "bits": 4,
654
+ "mode": "nvfp4",
655
+ "global_scale": true
656
+ },
657
+ "language_model.model.layers.38.block_sparse_moe.switch_mlp.down_proj": {
658
+ "group_size": 16,
659
+ "bits": 4,
660
+ "mode": "nvfp4",
661
+ "global_scale": true
662
+ },
663
+ "language_model.model.layers.38.block_sparse_moe.switch_mlp.gate_proj": {
664
+ "group_size": 16,
665
+ "bits": 4,
666
+ "mode": "nvfp4",
667
+ "global_scale": true
668
+ },
669
+ "language_model.model.layers.38.block_sparse_moe.switch_mlp.up_proj": {
670
+ "group_size": 16,
671
+ "bits": 4,
672
+ "mode": "nvfp4",
673
+ "global_scale": true
674
+ },
675
+ "language_model.model.layers.39.block_sparse_moe.switch_mlp.down_proj": {
676
+ "group_size": 16,
677
+ "bits": 4,
678
+ "mode": "nvfp4",
679
+ "global_scale": true
680
+ },
681
+ "language_model.model.layers.39.block_sparse_moe.switch_mlp.gate_proj": {
682
+ "group_size": 16,
683
+ "bits": 4,
684
+ "mode": "nvfp4",
685
+ "global_scale": true
686
+ },
687
+ "language_model.model.layers.39.block_sparse_moe.switch_mlp.up_proj": {
688
+ "group_size": 16,
689
+ "bits": 4,
690
+ "mode": "nvfp4",
691
+ "global_scale": true
692
+ },
693
+ "language_model.model.layers.40.block_sparse_moe.switch_mlp.down_proj": {
694
+ "group_size": 16,
695
+ "bits": 4,
696
+ "mode": "nvfp4",
697
+ "global_scale": true
698
+ },
699
+ "language_model.model.layers.40.block_sparse_moe.switch_mlp.gate_proj": {
700
+ "group_size": 16,
701
+ "bits": 4,
702
+ "mode": "nvfp4",
703
+ "global_scale": true
704
+ },
705
+ "language_model.model.layers.40.block_sparse_moe.switch_mlp.up_proj": {
706
+ "group_size": 16,
707
+ "bits": 4,
708
+ "mode": "nvfp4",
709
+ "global_scale": true
710
+ },
711
+ "language_model.model.layers.41.block_sparse_moe.switch_mlp.down_proj": {
712
+ "group_size": 16,
713
+ "bits": 4,
714
+ "mode": "nvfp4",
715
+ "global_scale": true
716
+ },
717
+ "language_model.model.layers.41.block_sparse_moe.switch_mlp.gate_proj": {
718
+ "group_size": 16,
719
+ "bits": 4,
720
+ "mode": "nvfp4",
721
+ "global_scale": true
722
+ },
723
+ "language_model.model.layers.41.block_sparse_moe.switch_mlp.up_proj": {
724
+ "group_size": 16,
725
+ "bits": 4,
726
+ "mode": "nvfp4",
727
+ "global_scale": true
728
+ },
729
+ "language_model.model.layers.42.block_sparse_moe.switch_mlp.down_proj": {
730
+ "group_size": 16,
731
+ "bits": 4,
732
+ "mode": "nvfp4",
733
+ "global_scale": true
734
+ },
735
+ "language_model.model.layers.42.block_sparse_moe.switch_mlp.gate_proj": {
736
+ "group_size": 16,
737
+ "bits": 4,
738
+ "mode": "nvfp4",
739
+ "global_scale": true
740
+ },
741
+ "language_model.model.layers.42.block_sparse_moe.switch_mlp.up_proj": {
742
+ "group_size": 16,
743
+ "bits": 4,
744
+ "mode": "nvfp4",
745
+ "global_scale": true
746
+ },
747
+ "language_model.model.layers.43.block_sparse_moe.switch_mlp.down_proj": {
748
+ "group_size": 16,
749
+ "bits": 4,
750
+ "mode": "nvfp4",
751
+ "global_scale": true
752
+ },
753
+ "language_model.model.layers.43.block_sparse_moe.switch_mlp.gate_proj": {
754
+ "group_size": 16,
755
+ "bits": 4,
756
+ "mode": "nvfp4",
757
+ "global_scale": true
758
+ },
759
+ "language_model.model.layers.43.block_sparse_moe.switch_mlp.up_proj": {
760
+ "group_size": 16,
761
+ "bits": 4,
762
+ "mode": "nvfp4",
763
+ "global_scale": true
764
+ },
765
+ "language_model.model.layers.44.block_sparse_moe.switch_mlp.down_proj": {
766
+ "group_size": 16,
767
+ "bits": 4,
768
+ "mode": "nvfp4",
769
+ "global_scale": true
770
+ },
771
+ "language_model.model.layers.44.block_sparse_moe.switch_mlp.gate_proj": {
772
+ "group_size": 16,
773
+ "bits": 4,
774
+ "mode": "nvfp4",
775
+ "global_scale": true
776
+ },
777
+ "language_model.model.layers.44.block_sparse_moe.switch_mlp.up_proj": {
778
+ "group_size": 16,
779
+ "bits": 4,
780
+ "mode": "nvfp4",
781
+ "global_scale": true
782
+ },
783
+ "language_model.model.layers.45.block_sparse_moe.switch_mlp.down_proj": {
784
+ "group_size": 16,
785
+ "bits": 4,
786
+ "mode": "nvfp4",
787
+ "global_scale": true
788
+ },
789
+ "language_model.model.layers.45.block_sparse_moe.switch_mlp.gate_proj": {
790
+ "group_size": 16,
791
+ "bits": 4,
792
+ "mode": "nvfp4",
793
+ "global_scale": true
794
+ },
795
+ "language_model.model.layers.45.block_sparse_moe.switch_mlp.up_proj": {
796
+ "group_size": 16,
797
+ "bits": 4,
798
+ "mode": "nvfp4",
799
+ "global_scale": true
800
+ },
801
+ "language_model.model.layers.46.block_sparse_moe.switch_mlp.down_proj": {
802
+ "group_size": 16,
803
+ "bits": 4,
804
+ "mode": "nvfp4",
805
+ "global_scale": true
806
+ },
807
+ "language_model.model.layers.46.block_sparse_moe.switch_mlp.gate_proj": {
808
+ "group_size": 16,
809
+ "bits": 4,
810
+ "mode": "nvfp4",
811
+ "global_scale": true
812
+ },
813
+ "language_model.model.layers.46.block_sparse_moe.switch_mlp.up_proj": {
814
+ "group_size": 16,
815
+ "bits": 4,
816
+ "mode": "nvfp4",
817
+ "global_scale": true
818
+ },
819
+ "language_model.model.layers.47.block_sparse_moe.switch_mlp.down_proj": {
820
+ "group_size": 16,
821
+ "bits": 4,
822
+ "mode": "nvfp4",
823
+ "global_scale": true
824
+ },
825
+ "language_model.model.layers.47.block_sparse_moe.switch_mlp.gate_proj": {
826
+ "group_size": 16,
827
+ "bits": 4,
828
+ "mode": "nvfp4",
829
+ "global_scale": true
830
+ },
831
+ "language_model.model.layers.47.block_sparse_moe.switch_mlp.up_proj": {
832
+ "group_size": 16,
833
+ "bits": 4,
834
+ "mode": "nvfp4",
835
+ "global_scale": true
836
+ },
837
+ "language_model.model.layers.48.block_sparse_moe.switch_mlp.down_proj": {
838
+ "group_size": 16,
839
+ "bits": 4,
840
+ "mode": "nvfp4",
841
+ "global_scale": true
842
+ },
843
+ "language_model.model.layers.48.block_sparse_moe.switch_mlp.gate_proj": {
844
+ "group_size": 16,
845
+ "bits": 4,
846
+ "mode": "nvfp4",
847
+ "global_scale": true
848
+ },
849
+ "language_model.model.layers.48.block_sparse_moe.switch_mlp.up_proj": {
850
+ "group_size": 16,
851
+ "bits": 4,
852
+ "mode": "nvfp4",
853
+ "global_scale": true
854
+ },
855
+ "language_model.model.layers.49.block_sparse_moe.switch_mlp.down_proj": {
856
+ "group_size": 16,
857
+ "bits": 4,
858
+ "mode": "nvfp4",
859
+ "global_scale": true
860
+ },
861
+ "language_model.model.layers.49.block_sparse_moe.switch_mlp.gate_proj": {
862
+ "group_size": 16,
863
+ "bits": 4,
864
+ "mode": "nvfp4",
865
+ "global_scale": true
866
+ },
867
+ "language_model.model.layers.49.block_sparse_moe.switch_mlp.up_proj": {
868
+ "group_size": 16,
869
+ "bits": 4,
870
+ "mode": "nvfp4",
871
+ "global_scale": true
872
+ },
873
+ "language_model.model.layers.50.block_sparse_moe.switch_mlp.down_proj": {
874
+ "group_size": 16,
875
+ "bits": 4,
876
+ "mode": "nvfp4",
877
+ "global_scale": true
878
+ },
879
+ "language_model.model.layers.50.block_sparse_moe.switch_mlp.gate_proj": {
880
+ "group_size": 16,
881
+ "bits": 4,
882
+ "mode": "nvfp4",
883
+ "global_scale": true
884
+ },
885
+ "language_model.model.layers.50.block_sparse_moe.switch_mlp.up_proj": {
886
+ "group_size": 16,
887
+ "bits": 4,
888
+ "mode": "nvfp4",
889
+ "global_scale": true
890
+ },
891
+ "language_model.model.layers.51.block_sparse_moe.switch_mlp.down_proj": {
892
+ "group_size": 16,
893
+ "bits": 4,
894
+ "mode": "nvfp4",
895
+ "global_scale": true
896
+ },
897
+ "language_model.model.layers.51.block_sparse_moe.switch_mlp.gate_proj": {
898
+ "group_size": 16,
899
+ "bits": 4,
900
+ "mode": "nvfp4",
901
+ "global_scale": true
902
+ },
903
+ "language_model.model.layers.51.block_sparse_moe.switch_mlp.up_proj": {
904
+ "group_size": 16,
905
+ "bits": 4,
906
+ "mode": "nvfp4",
907
+ "global_scale": true
908
+ },
909
+ "language_model.model.layers.52.block_sparse_moe.switch_mlp.down_proj": {
910
+ "group_size": 16,
911
+ "bits": 4,
912
+ "mode": "nvfp4",
913
+ "global_scale": true
914
+ },
915
+ "language_model.model.layers.52.block_sparse_moe.switch_mlp.gate_proj": {
916
+ "group_size": 16,
917
+ "bits": 4,
918
+ "mode": "nvfp4",
919
+ "global_scale": true
920
+ },
921
+ "language_model.model.layers.52.block_sparse_moe.switch_mlp.up_proj": {
922
+ "group_size": 16,
923
+ "bits": 4,
924
+ "mode": "nvfp4",
925
+ "global_scale": true
926
+ },
927
+ "language_model.model.layers.53.block_sparse_moe.switch_mlp.down_proj": {
928
+ "group_size": 16,
929
+ "bits": 4,
930
+ "mode": "nvfp4",
931
+ "global_scale": true
932
+ },
933
+ "language_model.model.layers.53.block_sparse_moe.switch_mlp.gate_proj": {
934
+ "group_size": 16,
935
+ "bits": 4,
936
+ "mode": "nvfp4",
937
+ "global_scale": true
938
+ },
939
+ "language_model.model.layers.53.block_sparse_moe.switch_mlp.up_proj": {
940
+ "group_size": 16,
941
+ "bits": 4,
942
+ "mode": "nvfp4",
943
+ "global_scale": true
944
+ },
945
+ "language_model.model.layers.54.block_sparse_moe.switch_mlp.down_proj": {
946
+ "group_size": 16,
947
+ "bits": 4,
948
+ "mode": "nvfp4",
949
+ "global_scale": true
950
+ },
951
+ "language_model.model.layers.54.block_sparse_moe.switch_mlp.gate_proj": {
952
+ "group_size": 16,
953
+ "bits": 4,
954
+ "mode": "nvfp4",
955
+ "global_scale": true
956
+ },
957
+ "language_model.model.layers.54.block_sparse_moe.switch_mlp.up_proj": {
958
+ "group_size": 16,
959
+ "bits": 4,
960
+ "mode": "nvfp4",
961
+ "global_scale": true
962
+ },
963
+ "language_model.model.layers.55.block_sparse_moe.switch_mlp.down_proj": {
964
+ "group_size": 16,
965
+ "bits": 4,
966
+ "mode": "nvfp4",
967
+ "global_scale": true
968
+ },
969
+ "language_model.model.layers.55.block_sparse_moe.switch_mlp.gate_proj": {
970
+ "group_size": 16,
971
+ "bits": 4,
972
+ "mode": "nvfp4",
973
+ "global_scale": true
974
+ },
975
+ "language_model.model.layers.55.block_sparse_moe.switch_mlp.up_proj": {
976
+ "group_size": 16,
977
+ "bits": 4,
978
+ "mode": "nvfp4",
979
+ "global_scale": true
980
+ },
981
+ "language_model.model.layers.56.block_sparse_moe.switch_mlp.down_proj": {
982
+ "group_size": 16,
983
+ "bits": 4,
984
+ "mode": "nvfp4",
985
+ "global_scale": true
986
+ },
987
+ "language_model.model.layers.56.block_sparse_moe.switch_mlp.gate_proj": {
988
+ "group_size": 16,
989
+ "bits": 4,
990
+ "mode": "nvfp4",
991
+ "global_scale": true
992
+ },
993
+ "language_model.model.layers.56.block_sparse_moe.switch_mlp.up_proj": {
994
+ "group_size": 16,
995
+ "bits": 4,
996
+ "mode": "nvfp4",
997
+ "global_scale": true
998
+ },
999
+ "language_model.model.layers.57.block_sparse_moe.switch_mlp.down_proj": {
1000
+ "group_size": 16,
1001
+ "bits": 4,
1002
+ "mode": "nvfp4",
1003
+ "global_scale": true
1004
+ },
1005
+ "language_model.model.layers.57.block_sparse_moe.switch_mlp.gate_proj": {
1006
+ "group_size": 16,
1007
+ "bits": 4,
1008
+ "mode": "nvfp4",
1009
+ "global_scale": true
1010
+ },
1011
+ "language_model.model.layers.57.block_sparse_moe.switch_mlp.up_proj": {
1012
+ "group_size": 16,
1013
+ "bits": 4,
1014
+ "mode": "nvfp4",
1015
+ "global_scale": true
1016
+ },
1017
+ "language_model.model.layers.58.block_sparse_moe.switch_mlp.down_proj": {
1018
+ "group_size": 16,
1019
+ "bits": 4,
1020
+ "mode": "nvfp4",
1021
+ "global_scale": true
1022
+ },
1023
+ "language_model.model.layers.58.block_sparse_moe.switch_mlp.gate_proj": {
1024
+ "group_size": 16,
1025
+ "bits": 4,
1026
+ "mode": "nvfp4",
1027
+ "global_scale": true
1028
+ },
1029
+ "language_model.model.layers.58.block_sparse_moe.switch_mlp.up_proj": {
1030
+ "group_size": 16,
1031
+ "bits": 4,
1032
+ "mode": "nvfp4",
1033
+ "global_scale": true
1034
+ },
1035
+ "language_model.model.layers.59.block_sparse_moe.switch_mlp.down_proj": {
1036
+ "group_size": 16,
1037
+ "bits": 4,
1038
+ "mode": "nvfp4",
1039
+ "global_scale": true
1040
+ },
1041
+ "language_model.model.layers.59.block_sparse_moe.switch_mlp.gate_proj": {
1042
+ "group_size": 16,
1043
+ "bits": 4,
1044
+ "mode": "nvfp4",
1045
+ "global_scale": true
1046
+ },
1047
+ "language_model.model.layers.59.block_sparse_moe.switch_mlp.up_proj": {
1048
+ "group_size": 16,
1049
+ "bits": 4,
1050
+ "mode": "nvfp4",
1051
+ "global_scale": true
1052
+ }
1053
  },
1054
  "quantization_config": {
1055
  "group_size": 32,
1056
  "bits": 8,
1057
+ "mode": "mxfp8",
1058
+ "language_model.model.layers.3.block_sparse_moe.switch_mlp.down_proj": {
1059
+ "group_size": 16,
1060
+ "bits": 4,
1061
+ "mode": "nvfp4",
1062
+ "global_scale": true
1063
+ },
1064
+ "language_model.model.layers.3.block_sparse_moe.switch_mlp.gate_proj": {
1065
+ "group_size": 16,
1066
+ "bits": 4,
1067
+ "mode": "nvfp4",
1068
+ "global_scale": true
1069
+ },
1070
+ "language_model.model.layers.3.block_sparse_moe.switch_mlp.up_proj": {
1071
+ "group_size": 16,
1072
+ "bits": 4,
1073
+ "mode": "nvfp4",
1074
+ "global_scale": true
1075
+ },
1076
+ "language_model.model.layers.4.block_sparse_moe.switch_mlp.down_proj": {
1077
+ "group_size": 16,
1078
+ "bits": 4,
1079
+ "mode": "nvfp4",
1080
+ "global_scale": true
1081
+ },
1082
+ "language_model.model.layers.4.block_sparse_moe.switch_mlp.gate_proj": {
1083
+ "group_size": 16,
1084
+ "bits": 4,
1085
+ "mode": "nvfp4",
1086
+ "global_scale": true
1087
+ },
1088
+ "language_model.model.layers.4.block_sparse_moe.switch_mlp.up_proj": {
1089
+ "group_size": 16,
1090
+ "bits": 4,
1091
+ "mode": "nvfp4",
1092
+ "global_scale": true
1093
+ },
1094
+ "language_model.model.layers.5.block_sparse_moe.switch_mlp.down_proj": {
1095
+ "group_size": 16,
1096
+ "bits": 4,
1097
+ "mode": "nvfp4",
1098
+ "global_scale": true
1099
+ },
1100
+ "language_model.model.layers.5.block_sparse_moe.switch_mlp.gate_proj": {
1101
+ "group_size": 16,
1102
+ "bits": 4,
1103
+ "mode": "nvfp4",
1104
+ "global_scale": true
1105
+ },
1106
+ "language_model.model.layers.5.block_sparse_moe.switch_mlp.up_proj": {
1107
+ "group_size": 16,
1108
+ "bits": 4,
1109
+ "mode": "nvfp4",
1110
+ "global_scale": true
1111
+ },
1112
+ "language_model.model.layers.6.block_sparse_moe.switch_mlp.down_proj": {
1113
+ "group_size": 16,
1114
+ "bits": 4,
1115
+ "mode": "nvfp4",
1116
+ "global_scale": true
1117
+ },
1118
+ "language_model.model.layers.6.block_sparse_moe.switch_mlp.gate_proj": {
1119
+ "group_size": 16,
1120
+ "bits": 4,
1121
+ "mode": "nvfp4",
1122
+ "global_scale": true
1123
+ },
1124
+ "language_model.model.layers.6.block_sparse_moe.switch_mlp.up_proj": {
1125
+ "group_size": 16,
1126
+ "bits": 4,
1127
+ "mode": "nvfp4",
1128
+ "global_scale": true
1129
+ },
1130
+ "language_model.model.layers.7.block_sparse_moe.switch_mlp.down_proj": {
1131
+ "group_size": 16,
1132
+ "bits": 4,
1133
+ "mode": "nvfp4",
1134
+ "global_scale": true
1135
+ },
1136
+ "language_model.model.layers.7.block_sparse_moe.switch_mlp.gate_proj": {
1137
+ "group_size": 16,
1138
+ "bits": 4,
1139
+ "mode": "nvfp4",
1140
+ "global_scale": true
1141
+ },
1142
+ "language_model.model.layers.7.block_sparse_moe.switch_mlp.up_proj": {
1143
+ "group_size": 16,
1144
+ "bits": 4,
1145
+ "mode": "nvfp4",
1146
+ "global_scale": true
1147
+ },
1148
+ "language_model.model.layers.8.block_sparse_moe.switch_mlp.down_proj": {
1149
+ "group_size": 16,
1150
+ "bits": 4,
1151
+ "mode": "nvfp4",
1152
+ "global_scale": true
1153
+ },
1154
+ "language_model.model.layers.8.block_sparse_moe.switch_mlp.gate_proj": {
1155
+ "group_size": 16,
1156
+ "bits": 4,
1157
+ "mode": "nvfp4",
1158
+ "global_scale": true
1159
+ },
1160
+ "language_model.model.layers.8.block_sparse_moe.switch_mlp.up_proj": {
1161
+ "group_size": 16,
1162
+ "bits": 4,
1163
+ "mode": "nvfp4",
1164
+ "global_scale": true
1165
+ },
1166
+ "language_model.model.layers.9.block_sparse_moe.switch_mlp.down_proj": {
1167
+ "group_size": 16,
1168
+ "bits": 4,
1169
+ "mode": "nvfp4",
1170
+ "global_scale": true
1171
+ },
1172
+ "language_model.model.layers.9.block_sparse_moe.switch_mlp.gate_proj": {
1173
+ "group_size": 16,
1174
+ "bits": 4,
1175
+ "mode": "nvfp4",
1176
+ "global_scale": true
1177
+ },
1178
+ "language_model.model.layers.9.block_sparse_moe.switch_mlp.up_proj": {
1179
+ "group_size": 16,
1180
+ "bits": 4,
1181
+ "mode": "nvfp4",
1182
+ "global_scale": true
1183
+ },
1184
+ "language_model.model.layers.10.block_sparse_moe.switch_mlp.down_proj": {
1185
+ "group_size": 16,
1186
+ "bits": 4,
1187
+ "mode": "nvfp4",
1188
+ "global_scale": true
1189
+ },
1190
+ "language_model.model.layers.10.block_sparse_moe.switch_mlp.gate_proj": {
1191
+ "group_size": 16,
1192
+ "bits": 4,
1193
+ "mode": "nvfp4",
1194
+ "global_scale": true
1195
+ },
1196
+ "language_model.model.layers.10.block_sparse_moe.switch_mlp.up_proj": {
1197
+ "group_size": 16,
1198
+ "bits": 4,
1199
+ "mode": "nvfp4",
1200
+ "global_scale": true
1201
+ },
1202
+ "language_model.model.layers.11.block_sparse_moe.switch_mlp.down_proj": {
1203
+ "group_size": 16,
1204
+ "bits": 4,
1205
+ "mode": "nvfp4",
1206
+ "global_scale": true
1207
+ },
1208
+ "language_model.model.layers.11.block_sparse_moe.switch_mlp.gate_proj": {
1209
+ "group_size": 16,
1210
+ "bits": 4,
1211
+ "mode": "nvfp4",
1212
+ "global_scale": true
1213
+ },
1214
+ "language_model.model.layers.11.block_sparse_moe.switch_mlp.up_proj": {
1215
+ "group_size": 16,
1216
+ "bits": 4,
1217
+ "mode": "nvfp4",
1218
+ "global_scale": true
1219
+ },
1220
+ "language_model.model.layers.12.block_sparse_moe.switch_mlp.down_proj": {
1221
+ "group_size": 16,
1222
+ "bits": 4,
1223
+ "mode": "nvfp4",
1224
+ "global_scale": true
1225
+ },
1226
+ "language_model.model.layers.12.block_sparse_moe.switch_mlp.gate_proj": {
1227
+ "group_size": 16,
1228
+ "bits": 4,
1229
+ "mode": "nvfp4",
1230
+ "global_scale": true
1231
+ },
1232
+ "language_model.model.layers.12.block_sparse_moe.switch_mlp.up_proj": {
1233
+ "group_size": 16,
1234
+ "bits": 4,
1235
+ "mode": "nvfp4",
1236
+ "global_scale": true
1237
+ },
1238
+ "language_model.model.layers.13.block_sparse_moe.switch_mlp.down_proj": {
1239
+ "group_size": 16,
1240
+ "bits": 4,
1241
+ "mode": "nvfp4",
1242
+ "global_scale": true
1243
+ },
1244
+ "language_model.model.layers.13.block_sparse_moe.switch_mlp.gate_proj": {
1245
+ "group_size": 16,
1246
+ "bits": 4,
1247
+ "mode": "nvfp4",
1248
+ "global_scale": true
1249
+ },
1250
+ "language_model.model.layers.13.block_sparse_moe.switch_mlp.up_proj": {
1251
+ "group_size": 16,
1252
+ "bits": 4,
1253
+ "mode": "nvfp4",
1254
+ "global_scale": true
1255
+ },
1256
+ "language_model.model.layers.14.block_sparse_moe.switch_mlp.down_proj": {
1257
+ "group_size": 16,
1258
+ "bits": 4,
1259
+ "mode": "nvfp4",
1260
+ "global_scale": true
1261
+ },
1262
+ "language_model.model.layers.14.block_sparse_moe.switch_mlp.gate_proj": {
1263
+ "group_size": 16,
1264
+ "bits": 4,
1265
+ "mode": "nvfp4",
1266
+ "global_scale": true
1267
+ },
1268
+ "language_model.model.layers.14.block_sparse_moe.switch_mlp.up_proj": {
1269
+ "group_size": 16,
1270
+ "bits": 4,
1271
+ "mode": "nvfp4",
1272
+ "global_scale": true
1273
+ },
1274
+ "language_model.model.layers.15.block_sparse_moe.switch_mlp.down_proj": {
1275
+ "group_size": 16,
1276
+ "bits": 4,
1277
+ "mode": "nvfp4",
1278
+ "global_scale": true
1279
+ },
1280
+ "language_model.model.layers.15.block_sparse_moe.switch_mlp.gate_proj": {
1281
+ "group_size": 16,
1282
+ "bits": 4,
1283
+ "mode": "nvfp4",
1284
+ "global_scale": true
1285
+ },
1286
+ "language_model.model.layers.15.block_sparse_moe.switch_mlp.up_proj": {
1287
+ "group_size": 16,
1288
+ "bits": 4,
1289
+ "mode": "nvfp4",
1290
+ "global_scale": true
1291
+ },
1292
+ "language_model.model.layers.16.block_sparse_moe.switch_mlp.down_proj": {
1293
+ "group_size": 16,
1294
+ "bits": 4,
1295
+ "mode": "nvfp4",
1296
+ "global_scale": true
1297
+ },
1298
+ "language_model.model.layers.16.block_sparse_moe.switch_mlp.gate_proj": {
1299
+ "group_size": 16,
1300
+ "bits": 4,
1301
+ "mode": "nvfp4",
1302
+ "global_scale": true
1303
+ },
1304
+ "language_model.model.layers.16.block_sparse_moe.switch_mlp.up_proj": {
1305
+ "group_size": 16,
1306
+ "bits": 4,
1307
+ "mode": "nvfp4",
1308
+ "global_scale": true
1309
+ },
1310
+ "language_model.model.layers.17.block_sparse_moe.switch_mlp.down_proj": {
1311
+ "group_size": 16,
1312
+ "bits": 4,
1313
+ "mode": "nvfp4",
1314
+ "global_scale": true
1315
+ },
1316
+ "language_model.model.layers.17.block_sparse_moe.switch_mlp.gate_proj": {
1317
+ "group_size": 16,
1318
+ "bits": 4,
1319
+ "mode": "nvfp4",
1320
+ "global_scale": true
1321
+ },
1322
+ "language_model.model.layers.17.block_sparse_moe.switch_mlp.up_proj": {
1323
+ "group_size": 16,
1324
+ "bits": 4,
1325
+ "mode": "nvfp4",
1326
+ "global_scale": true
1327
+ },
1328
+ "language_model.model.layers.18.block_sparse_moe.switch_mlp.down_proj": {
1329
+ "group_size": 16,
1330
+ "bits": 4,
1331
+ "mode": "nvfp4",
1332
+ "global_scale": true
1333
+ },
1334
+ "language_model.model.layers.18.block_sparse_moe.switch_mlp.gate_proj": {
1335
+ "group_size": 16,
1336
+ "bits": 4,
1337
+ "mode": "nvfp4",
1338
+ "global_scale": true
1339
+ },
1340
+ "language_model.model.layers.18.block_sparse_moe.switch_mlp.up_proj": {
1341
+ "group_size": 16,
1342
+ "bits": 4,
1343
+ "mode": "nvfp4",
1344
+ "global_scale": true
1345
+ },
1346
+ "language_model.model.layers.19.block_sparse_moe.switch_mlp.down_proj": {
1347
+ "group_size": 16,
1348
+ "bits": 4,
1349
+ "mode": "nvfp4",
1350
+ "global_scale": true
1351
+ },
1352
+ "language_model.model.layers.19.block_sparse_moe.switch_mlp.gate_proj": {
1353
+ "group_size": 16,
1354
+ "bits": 4,
1355
+ "mode": "nvfp4",
1356
+ "global_scale": true
1357
+ },
1358
+ "language_model.model.layers.19.block_sparse_moe.switch_mlp.up_proj": {
1359
+ "group_size": 16,
1360
+ "bits": 4,
1361
+ "mode": "nvfp4",
1362
+ "global_scale": true
1363
+ },
1364
+ "language_model.model.layers.20.block_sparse_moe.switch_mlp.down_proj": {
1365
+ "group_size": 16,
1366
+ "bits": 4,
1367
+ "mode": "nvfp4",
1368
+ "global_scale": true
1369
+ },
1370
+ "language_model.model.layers.20.block_sparse_moe.switch_mlp.gate_proj": {
1371
+ "group_size": 16,
1372
+ "bits": 4,
1373
+ "mode": "nvfp4",
1374
+ "global_scale": true
1375
+ },
1376
+ "language_model.model.layers.20.block_sparse_moe.switch_mlp.up_proj": {
1377
+ "group_size": 16,
1378
+ "bits": 4,
1379
+ "mode": "nvfp4",
1380
+ "global_scale": true
1381
+ },
1382
+ "language_model.model.layers.21.block_sparse_moe.switch_mlp.down_proj": {
1383
+ "group_size": 16,
1384
+ "bits": 4,
1385
+ "mode": "nvfp4",
1386
+ "global_scale": true
1387
+ },
1388
+ "language_model.model.layers.21.block_sparse_moe.switch_mlp.gate_proj": {
1389
+ "group_size": 16,
1390
+ "bits": 4,
1391
+ "mode": "nvfp4",
1392
+ "global_scale": true
1393
+ },
1394
+ "language_model.model.layers.21.block_sparse_moe.switch_mlp.up_proj": {
1395
+ "group_size": 16,
1396
+ "bits": 4,
1397
+ "mode": "nvfp4",
1398
+ "global_scale": true
1399
+ },
1400
+ "language_model.model.layers.22.block_sparse_moe.switch_mlp.down_proj": {
1401
+ "group_size": 16,
1402
+ "bits": 4,
1403
+ "mode": "nvfp4",
1404
+ "global_scale": true
1405
+ },
1406
+ "language_model.model.layers.22.block_sparse_moe.switch_mlp.gate_proj": {
1407
+ "group_size": 16,
1408
+ "bits": 4,
1409
+ "mode": "nvfp4",
1410
+ "global_scale": true
1411
+ },
1412
+ "language_model.model.layers.22.block_sparse_moe.switch_mlp.up_proj": {
1413
+ "group_size": 16,
1414
+ "bits": 4,
1415
+ "mode": "nvfp4",
1416
+ "global_scale": true
1417
+ },
1418
+ "language_model.model.layers.23.block_sparse_moe.switch_mlp.down_proj": {
1419
+ "group_size": 16,
1420
+ "bits": 4,
1421
+ "mode": "nvfp4",
1422
+ "global_scale": true
1423
+ },
1424
+ "language_model.model.layers.23.block_sparse_moe.switch_mlp.gate_proj": {
1425
+ "group_size": 16,
1426
+ "bits": 4,
1427
+ "mode": "nvfp4",
1428
+ "global_scale": true
1429
+ },
1430
+ "language_model.model.layers.23.block_sparse_moe.switch_mlp.up_proj": {
1431
+ "group_size": 16,
1432
+ "bits": 4,
1433
+ "mode": "nvfp4",
1434
+ "global_scale": true
1435
+ },
1436
+ "language_model.model.layers.24.block_sparse_moe.switch_mlp.down_proj": {
1437
+ "group_size": 16,
1438
+ "bits": 4,
1439
+ "mode": "nvfp4",
1440
+ "global_scale": true
1441
+ },
1442
+ "language_model.model.layers.24.block_sparse_moe.switch_mlp.gate_proj": {
1443
+ "group_size": 16,
1444
+ "bits": 4,
1445
+ "mode": "nvfp4",
1446
+ "global_scale": true
1447
+ },
1448
+ "language_model.model.layers.24.block_sparse_moe.switch_mlp.up_proj": {
1449
+ "group_size": 16,
1450
+ "bits": 4,
1451
+ "mode": "nvfp4",
1452
+ "global_scale": true
1453
+ },
1454
+ "language_model.model.layers.25.block_sparse_moe.switch_mlp.down_proj": {
1455
+ "group_size": 16,
1456
+ "bits": 4,
1457
+ "mode": "nvfp4",
1458
+ "global_scale": true
1459
+ },
1460
+ "language_model.model.layers.25.block_sparse_moe.switch_mlp.gate_proj": {
1461
+ "group_size": 16,
1462
+ "bits": 4,
1463
+ "mode": "nvfp4",
1464
+ "global_scale": true
1465
+ },
1466
+ "language_model.model.layers.25.block_sparse_moe.switch_mlp.up_proj": {
1467
+ "group_size": 16,
1468
+ "bits": 4,
1469
+ "mode": "nvfp4",
1470
+ "global_scale": true
1471
+ },
1472
+ "language_model.model.layers.26.block_sparse_moe.switch_mlp.down_proj": {
1473
+ "group_size": 16,
1474
+ "bits": 4,
1475
+ "mode": "nvfp4",
1476
+ "global_scale": true
1477
+ },
1478
+ "language_model.model.layers.26.block_sparse_moe.switch_mlp.gate_proj": {
1479
+ "group_size": 16,
1480
+ "bits": 4,
1481
+ "mode": "nvfp4",
1482
+ "global_scale": true
1483
+ },
1484
+ "language_model.model.layers.26.block_sparse_moe.switch_mlp.up_proj": {
1485
+ "group_size": 16,
1486
+ "bits": 4,
1487
+ "mode": "nvfp4",
1488
+ "global_scale": true
1489
+ },
1490
+ "language_model.model.layers.27.block_sparse_moe.switch_mlp.down_proj": {
1491
+ "group_size": 16,
1492
+ "bits": 4,
1493
+ "mode": "nvfp4",
1494
+ "global_scale": true
1495
+ },
1496
+ "language_model.model.layers.27.block_sparse_moe.switch_mlp.gate_proj": {
1497
+ "group_size": 16,
1498
+ "bits": 4,
1499
+ "mode": "nvfp4",
1500
+ "global_scale": true
1501
+ },
1502
+ "language_model.model.layers.27.block_sparse_moe.switch_mlp.up_proj": {
1503
+ "group_size": 16,
1504
+ "bits": 4,
1505
+ "mode": "nvfp4",
1506
+ "global_scale": true
1507
+ },
1508
+ "language_model.model.layers.28.block_sparse_moe.switch_mlp.down_proj": {
1509
+ "group_size": 16,
1510
+ "bits": 4,
1511
+ "mode": "nvfp4",
1512
+ "global_scale": true
1513
+ },
1514
+ "language_model.model.layers.28.block_sparse_moe.switch_mlp.gate_proj": {
1515
+ "group_size": 16,
1516
+ "bits": 4,
1517
+ "mode": "nvfp4",
1518
+ "global_scale": true
1519
+ },
1520
+ "language_model.model.layers.28.block_sparse_moe.switch_mlp.up_proj": {
1521
+ "group_size": 16,
1522
+ "bits": 4,
1523
+ "mode": "nvfp4",
1524
+ "global_scale": true
1525
+ },
1526
+ "language_model.model.layers.29.block_sparse_moe.switch_mlp.down_proj": {
1527
+ "group_size": 16,
1528
+ "bits": 4,
1529
+ "mode": "nvfp4",
1530
+ "global_scale": true
1531
+ },
1532
+ "language_model.model.layers.29.block_sparse_moe.switch_mlp.gate_proj": {
1533
+ "group_size": 16,
1534
+ "bits": 4,
1535
+ "mode": "nvfp4",
1536
+ "global_scale": true
1537
+ },
1538
+ "language_model.model.layers.29.block_sparse_moe.switch_mlp.up_proj": {
1539
+ "group_size": 16,
1540
+ "bits": 4,
1541
+ "mode": "nvfp4",
1542
+ "global_scale": true
1543
+ },
1544
+ "language_model.model.layers.30.block_sparse_moe.switch_mlp.down_proj": {
1545
+ "group_size": 16,
1546
+ "bits": 4,
1547
+ "mode": "nvfp4",
1548
+ "global_scale": true
1549
+ },
1550
+ "language_model.model.layers.30.block_sparse_moe.switch_mlp.gate_proj": {
1551
+ "group_size": 16,
1552
+ "bits": 4,
1553
+ "mode": "nvfp4",
1554
+ "global_scale": true
1555
+ },
1556
+ "language_model.model.layers.30.block_sparse_moe.switch_mlp.up_proj": {
1557
+ "group_size": 16,
1558
+ "bits": 4,
1559
+ "mode": "nvfp4",
1560
+ "global_scale": true
1561
+ },
1562
+ "language_model.model.layers.31.block_sparse_moe.switch_mlp.down_proj": {
1563
+ "group_size": 16,
1564
+ "bits": 4,
1565
+ "mode": "nvfp4",
1566
+ "global_scale": true
1567
+ },
1568
+ "language_model.model.layers.31.block_sparse_moe.switch_mlp.gate_proj": {
1569
+ "group_size": 16,
1570
+ "bits": 4,
1571
+ "mode": "nvfp4",
1572
+ "global_scale": true
1573
+ },
1574
+ "language_model.model.layers.31.block_sparse_moe.switch_mlp.up_proj": {
1575
+ "group_size": 16,
1576
+ "bits": 4,
1577
+ "mode": "nvfp4",
1578
+ "global_scale": true
1579
+ },
1580
+ "language_model.model.layers.32.block_sparse_moe.switch_mlp.down_proj": {
1581
+ "group_size": 16,
1582
+ "bits": 4,
1583
+ "mode": "nvfp4",
1584
+ "global_scale": true
1585
+ },
1586
+ "language_model.model.layers.32.block_sparse_moe.switch_mlp.gate_proj": {
1587
+ "group_size": 16,
1588
+ "bits": 4,
1589
+ "mode": "nvfp4",
1590
+ "global_scale": true
1591
+ },
1592
+ "language_model.model.layers.32.block_sparse_moe.switch_mlp.up_proj": {
1593
+ "group_size": 16,
1594
+ "bits": 4,
1595
+ "mode": "nvfp4",
1596
+ "global_scale": true
1597
+ },
1598
+ "language_model.model.layers.33.block_sparse_moe.switch_mlp.down_proj": {
1599
+ "group_size": 16,
1600
+ "bits": 4,
1601
+ "mode": "nvfp4",
1602
+ "global_scale": true
1603
+ },
1604
+ "language_model.model.layers.33.block_sparse_moe.switch_mlp.gate_proj": {
1605
+ "group_size": 16,
1606
+ "bits": 4,
1607
+ "mode": "nvfp4",
1608
+ "global_scale": true
1609
+ },
1610
+ "language_model.model.layers.33.block_sparse_moe.switch_mlp.up_proj": {
1611
+ "group_size": 16,
1612
+ "bits": 4,
1613
+ "mode": "nvfp4",
1614
+ "global_scale": true
1615
+ },
1616
+ "language_model.model.layers.34.block_sparse_moe.switch_mlp.down_proj": {
1617
+ "group_size": 16,
1618
+ "bits": 4,
1619
+ "mode": "nvfp4",
1620
+ "global_scale": true
1621
+ },
1622
+ "language_model.model.layers.34.block_sparse_moe.switch_mlp.gate_proj": {
1623
+ "group_size": 16,
1624
+ "bits": 4,
1625
+ "mode": "nvfp4",
1626
+ "global_scale": true
1627
+ },
1628
+ "language_model.model.layers.34.block_sparse_moe.switch_mlp.up_proj": {
1629
+ "group_size": 16,
1630
+ "bits": 4,
1631
+ "mode": "nvfp4",
1632
+ "global_scale": true
1633
+ },
1634
+ "language_model.model.layers.35.block_sparse_moe.switch_mlp.down_proj": {
1635
+ "group_size": 16,
1636
+ "bits": 4,
1637
+ "mode": "nvfp4",
1638
+ "global_scale": true
1639
+ },
1640
+ "language_model.model.layers.35.block_sparse_moe.switch_mlp.gate_proj": {
1641
+ "group_size": 16,
1642
+ "bits": 4,
1643
+ "mode": "nvfp4",
1644
+ "global_scale": true
1645
+ },
1646
+ "language_model.model.layers.35.block_sparse_moe.switch_mlp.up_proj": {
1647
+ "group_size": 16,
1648
+ "bits": 4,
1649
+ "mode": "nvfp4",
1650
+ "global_scale": true
1651
+ },
1652
+ "language_model.model.layers.36.block_sparse_moe.switch_mlp.down_proj": {
1653
+ "group_size": 16,
1654
+ "bits": 4,
1655
+ "mode": "nvfp4",
1656
+ "global_scale": true
1657
+ },
1658
+ "language_model.model.layers.36.block_sparse_moe.switch_mlp.gate_proj": {
1659
+ "group_size": 16,
1660
+ "bits": 4,
1661
+ "mode": "nvfp4",
1662
+ "global_scale": true
1663
+ },
1664
+ "language_model.model.layers.36.block_sparse_moe.switch_mlp.up_proj": {
1665
+ "group_size": 16,
1666
+ "bits": 4,
1667
+ "mode": "nvfp4",
1668
+ "global_scale": true
1669
+ },
1670
+ "language_model.model.layers.37.block_sparse_moe.switch_mlp.down_proj": {
1671
+ "group_size": 16,
1672
+ "bits": 4,
1673
+ "mode": "nvfp4",
1674
+ "global_scale": true
1675
+ },
1676
+ "language_model.model.layers.37.block_sparse_moe.switch_mlp.gate_proj": {
1677
+ "group_size": 16,
1678
+ "bits": 4,
1679
+ "mode": "nvfp4",
1680
+ "global_scale": true
1681
+ },
1682
+ "language_model.model.layers.37.block_sparse_moe.switch_mlp.up_proj": {
1683
+ "group_size": 16,
1684
+ "bits": 4,
1685
+ "mode": "nvfp4",
1686
+ "global_scale": true
1687
+ },
1688
+ "language_model.model.layers.38.block_sparse_moe.switch_mlp.down_proj": {
1689
+ "group_size": 16,
1690
+ "bits": 4,
1691
+ "mode": "nvfp4",
1692
+ "global_scale": true
1693
+ },
1694
+ "language_model.model.layers.38.block_sparse_moe.switch_mlp.gate_proj": {
1695
+ "group_size": 16,
1696
+ "bits": 4,
1697
+ "mode": "nvfp4",
1698
+ "global_scale": true
1699
+ },
1700
+ "language_model.model.layers.38.block_sparse_moe.switch_mlp.up_proj": {
1701
+ "group_size": 16,
1702
+ "bits": 4,
1703
+ "mode": "nvfp4",
1704
+ "global_scale": true
1705
+ },
1706
+ "language_model.model.layers.39.block_sparse_moe.switch_mlp.down_proj": {
1707
+ "group_size": 16,
1708
+ "bits": 4,
1709
+ "mode": "nvfp4",
1710
+ "global_scale": true
1711
+ },
1712
+ "language_model.model.layers.39.block_sparse_moe.switch_mlp.gate_proj": {
1713
+ "group_size": 16,
1714
+ "bits": 4,
1715
+ "mode": "nvfp4",
1716
+ "global_scale": true
1717
+ },
1718
+ "language_model.model.layers.39.block_sparse_moe.switch_mlp.up_proj": {
1719
+ "group_size": 16,
1720
+ "bits": 4,
1721
+ "mode": "nvfp4",
1722
+ "global_scale": true
1723
+ },
1724
+ "language_model.model.layers.40.block_sparse_moe.switch_mlp.down_proj": {
1725
+ "group_size": 16,
1726
+ "bits": 4,
1727
+ "mode": "nvfp4",
1728
+ "global_scale": true
1729
+ },
1730
+ "language_model.model.layers.40.block_sparse_moe.switch_mlp.gate_proj": {
1731
+ "group_size": 16,
1732
+ "bits": 4,
1733
+ "mode": "nvfp4",
1734
+ "global_scale": true
1735
+ },
1736
+ "language_model.model.layers.40.block_sparse_moe.switch_mlp.up_proj": {
1737
+ "group_size": 16,
1738
+ "bits": 4,
1739
+ "mode": "nvfp4",
1740
+ "global_scale": true
1741
+ },
1742
+ "language_model.model.layers.41.block_sparse_moe.switch_mlp.down_proj": {
1743
+ "group_size": 16,
1744
+ "bits": 4,
1745
+ "mode": "nvfp4",
1746
+ "global_scale": true
1747
+ },
1748
+ "language_model.model.layers.41.block_sparse_moe.switch_mlp.gate_proj": {
1749
+ "group_size": 16,
1750
+ "bits": 4,
1751
+ "mode": "nvfp4",
1752
+ "global_scale": true
1753
+ },
1754
+ "language_model.model.layers.41.block_sparse_moe.switch_mlp.up_proj": {
1755
+ "group_size": 16,
1756
+ "bits": 4,
1757
+ "mode": "nvfp4",
1758
+ "global_scale": true
1759
+ },
1760
+ "language_model.model.layers.42.block_sparse_moe.switch_mlp.down_proj": {
1761
+ "group_size": 16,
1762
+ "bits": 4,
1763
+ "mode": "nvfp4",
1764
+ "global_scale": true
1765
+ },
1766
+ "language_model.model.layers.42.block_sparse_moe.switch_mlp.gate_proj": {
1767
+ "group_size": 16,
1768
+ "bits": 4,
1769
+ "mode": "nvfp4",
1770
+ "global_scale": true
1771
+ },
1772
+ "language_model.model.layers.42.block_sparse_moe.switch_mlp.up_proj": {
1773
+ "group_size": 16,
1774
+ "bits": 4,
1775
+ "mode": "nvfp4",
1776
+ "global_scale": true
1777
+ },
1778
+ "language_model.model.layers.43.block_sparse_moe.switch_mlp.down_proj": {
1779
+ "group_size": 16,
1780
+ "bits": 4,
1781
+ "mode": "nvfp4",
1782
+ "global_scale": true
1783
+ },
1784
+ "language_model.model.layers.43.block_sparse_moe.switch_mlp.gate_proj": {
1785
+ "group_size": 16,
1786
+ "bits": 4,
1787
+ "mode": "nvfp4",
1788
+ "global_scale": true
1789
+ },
1790
+ "language_model.model.layers.43.block_sparse_moe.switch_mlp.up_proj": {
1791
+ "group_size": 16,
1792
+ "bits": 4,
1793
+ "mode": "nvfp4",
1794
+ "global_scale": true
1795
+ },
1796
+ "language_model.model.layers.44.block_sparse_moe.switch_mlp.down_proj": {
1797
+ "group_size": 16,
1798
+ "bits": 4,
1799
+ "mode": "nvfp4",
1800
+ "global_scale": true
1801
+ },
1802
+ "language_model.model.layers.44.block_sparse_moe.switch_mlp.gate_proj": {
1803
+ "group_size": 16,
1804
+ "bits": 4,
1805
+ "mode": "nvfp4",
1806
+ "global_scale": true
1807
+ },
1808
+ "language_model.model.layers.44.block_sparse_moe.switch_mlp.up_proj": {
1809
+ "group_size": 16,
1810
+ "bits": 4,
1811
+ "mode": "nvfp4",
1812
+ "global_scale": true
1813
+ },
1814
+ "language_model.model.layers.45.block_sparse_moe.switch_mlp.down_proj": {
1815
+ "group_size": 16,
1816
+ "bits": 4,
1817
+ "mode": "nvfp4",
1818
+ "global_scale": true
1819
+ },
1820
+ "language_model.model.layers.45.block_sparse_moe.switch_mlp.gate_proj": {
1821
+ "group_size": 16,
1822
+ "bits": 4,
1823
+ "mode": "nvfp4",
1824
+ "global_scale": true
1825
+ },
1826
+ "language_model.model.layers.45.block_sparse_moe.switch_mlp.up_proj": {
1827
+ "group_size": 16,
1828
+ "bits": 4,
1829
+ "mode": "nvfp4",
1830
+ "global_scale": true
1831
+ },
1832
+ "language_model.model.layers.46.block_sparse_moe.switch_mlp.down_proj": {
1833
+ "group_size": 16,
1834
+ "bits": 4,
1835
+ "mode": "nvfp4",
1836
+ "global_scale": true
1837
+ },
1838
+ "language_model.model.layers.46.block_sparse_moe.switch_mlp.gate_proj": {
1839
+ "group_size": 16,
1840
+ "bits": 4,
1841
+ "mode": "nvfp4",
1842
+ "global_scale": true
1843
+ },
1844
+ "language_model.model.layers.46.block_sparse_moe.switch_mlp.up_proj": {
1845
+ "group_size": 16,
1846
+ "bits": 4,
1847
+ "mode": "nvfp4",
1848
+ "global_scale": true
1849
+ },
1850
+ "language_model.model.layers.47.block_sparse_moe.switch_mlp.down_proj": {
1851
+ "group_size": 16,
1852
+ "bits": 4,
1853
+ "mode": "nvfp4",
1854
+ "global_scale": true
1855
+ },
1856
+ "language_model.model.layers.47.block_sparse_moe.switch_mlp.gate_proj": {
1857
+ "group_size": 16,
1858
+ "bits": 4,
1859
+ "mode": "nvfp4",
1860
+ "global_scale": true
1861
+ },
1862
+ "language_model.model.layers.47.block_sparse_moe.switch_mlp.up_proj": {
1863
+ "group_size": 16,
1864
+ "bits": 4,
1865
+ "mode": "nvfp4",
1866
+ "global_scale": true
1867
+ },
1868
+ "language_model.model.layers.48.block_sparse_moe.switch_mlp.down_proj": {
1869
+ "group_size": 16,
1870
+ "bits": 4,
1871
+ "mode": "nvfp4",
1872
+ "global_scale": true
1873
+ },
1874
+ "language_model.model.layers.48.block_sparse_moe.switch_mlp.gate_proj": {
1875
+ "group_size": 16,
1876
+ "bits": 4,
1877
+ "mode": "nvfp4",
1878
+ "global_scale": true
1879
+ },
1880
+ "language_model.model.layers.48.block_sparse_moe.switch_mlp.up_proj": {
1881
+ "group_size": 16,
1882
+ "bits": 4,
1883
+ "mode": "nvfp4",
1884
+ "global_scale": true
1885
+ },
1886
+ "language_model.model.layers.49.block_sparse_moe.switch_mlp.down_proj": {
1887
+ "group_size": 16,
1888
+ "bits": 4,
1889
+ "mode": "nvfp4",
1890
+ "global_scale": true
1891
+ },
1892
+ "language_model.model.layers.49.block_sparse_moe.switch_mlp.gate_proj": {
1893
+ "group_size": 16,
1894
+ "bits": 4,
1895
+ "mode": "nvfp4",
1896
+ "global_scale": true
1897
+ },
1898
+ "language_model.model.layers.49.block_sparse_moe.switch_mlp.up_proj": {
1899
+ "group_size": 16,
1900
+ "bits": 4,
1901
+ "mode": "nvfp4",
1902
+ "global_scale": true
1903
+ },
1904
+ "language_model.model.layers.50.block_sparse_moe.switch_mlp.down_proj": {
1905
+ "group_size": 16,
1906
+ "bits": 4,
1907
+ "mode": "nvfp4",
1908
+ "global_scale": true
1909
+ },
1910
+ "language_model.model.layers.50.block_sparse_moe.switch_mlp.gate_proj": {
1911
+ "group_size": 16,
1912
+ "bits": 4,
1913
+ "mode": "nvfp4",
1914
+ "global_scale": true
1915
+ },
1916
+ "language_model.model.layers.50.block_sparse_moe.switch_mlp.up_proj": {
1917
+ "group_size": 16,
1918
+ "bits": 4,
1919
+ "mode": "nvfp4",
1920
+ "global_scale": true
1921
+ },
1922
+ "language_model.model.layers.51.block_sparse_moe.switch_mlp.down_proj": {
1923
+ "group_size": 16,
1924
+ "bits": 4,
1925
+ "mode": "nvfp4",
1926
+ "global_scale": true
1927
+ },
1928
+ "language_model.model.layers.51.block_sparse_moe.switch_mlp.gate_proj": {
1929
+ "group_size": 16,
1930
+ "bits": 4,
1931
+ "mode": "nvfp4",
1932
+ "global_scale": true
1933
+ },
1934
+ "language_model.model.layers.51.block_sparse_moe.switch_mlp.up_proj": {
1935
+ "group_size": 16,
1936
+ "bits": 4,
1937
+ "mode": "nvfp4",
1938
+ "global_scale": true
1939
+ },
1940
+ "language_model.model.layers.52.block_sparse_moe.switch_mlp.down_proj": {
1941
+ "group_size": 16,
1942
+ "bits": 4,
1943
+ "mode": "nvfp4",
1944
+ "global_scale": true
1945
+ },
1946
+ "language_model.model.layers.52.block_sparse_moe.switch_mlp.gate_proj": {
1947
+ "group_size": 16,
1948
+ "bits": 4,
1949
+ "mode": "nvfp4",
1950
+ "global_scale": true
1951
+ },
1952
+ "language_model.model.layers.52.block_sparse_moe.switch_mlp.up_proj": {
1953
+ "group_size": 16,
1954
+ "bits": 4,
1955
+ "mode": "nvfp4",
1956
+ "global_scale": true
1957
+ },
1958
+ "language_model.model.layers.53.block_sparse_moe.switch_mlp.down_proj": {
1959
+ "group_size": 16,
1960
+ "bits": 4,
1961
+ "mode": "nvfp4",
1962
+ "global_scale": true
1963
+ },
1964
+ "language_model.model.layers.53.block_sparse_moe.switch_mlp.gate_proj": {
1965
+ "group_size": 16,
1966
+ "bits": 4,
1967
+ "mode": "nvfp4",
1968
+ "global_scale": true
1969
+ },
1970
+ "language_model.model.layers.53.block_sparse_moe.switch_mlp.up_proj": {
1971
+ "group_size": 16,
1972
+ "bits": 4,
1973
+ "mode": "nvfp4",
1974
+ "global_scale": true
1975
+ },
1976
+ "language_model.model.layers.54.block_sparse_moe.switch_mlp.down_proj": {
1977
+ "group_size": 16,
1978
+ "bits": 4,
1979
+ "mode": "nvfp4",
1980
+ "global_scale": true
1981
+ },
1982
+ "language_model.model.layers.54.block_sparse_moe.switch_mlp.gate_proj": {
1983
+ "group_size": 16,
1984
+ "bits": 4,
1985
+ "mode": "nvfp4",
1986
+ "global_scale": true
1987
+ },
1988
+ "language_model.model.layers.54.block_sparse_moe.switch_mlp.up_proj": {
1989
+ "group_size": 16,
1990
+ "bits": 4,
1991
+ "mode": "nvfp4",
1992
+ "global_scale": true
1993
+ },
1994
+ "language_model.model.layers.55.block_sparse_moe.switch_mlp.down_proj": {
1995
+ "group_size": 16,
1996
+ "bits": 4,
1997
+ "mode": "nvfp4",
1998
+ "global_scale": true
1999
+ },
2000
+ "language_model.model.layers.55.block_sparse_moe.switch_mlp.gate_proj": {
2001
+ "group_size": 16,
2002
+ "bits": 4,
2003
+ "mode": "nvfp4",
2004
+ "global_scale": true
2005
+ },
2006
+ "language_model.model.layers.55.block_sparse_moe.switch_mlp.up_proj": {
2007
+ "group_size": 16,
2008
+ "bits": 4,
2009
+ "mode": "nvfp4",
2010
+ "global_scale": true
2011
+ },
2012
+ "language_model.model.layers.56.block_sparse_moe.switch_mlp.down_proj": {
2013
+ "group_size": 16,
2014
+ "bits": 4,
2015
+ "mode": "nvfp4",
2016
+ "global_scale": true
2017
+ },
2018
+ "language_model.model.layers.56.block_sparse_moe.switch_mlp.gate_proj": {
2019
+ "group_size": 16,
2020
+ "bits": 4,
2021
+ "mode": "nvfp4",
2022
+ "global_scale": true
2023
+ },
2024
+ "language_model.model.layers.56.block_sparse_moe.switch_mlp.up_proj": {
2025
+ "group_size": 16,
2026
+ "bits": 4,
2027
+ "mode": "nvfp4",
2028
+ "global_scale": true
2029
+ },
2030
+ "language_model.model.layers.57.block_sparse_moe.switch_mlp.down_proj": {
2031
+ "group_size": 16,
2032
+ "bits": 4,
2033
+ "mode": "nvfp4",
2034
+ "global_scale": true
2035
+ },
2036
+ "language_model.model.layers.57.block_sparse_moe.switch_mlp.gate_proj": {
2037
+ "group_size": 16,
2038
+ "bits": 4,
2039
+ "mode": "nvfp4",
2040
+ "global_scale": true
2041
+ },
2042
+ "language_model.model.layers.57.block_sparse_moe.switch_mlp.up_proj": {
2043
+ "group_size": 16,
2044
+ "bits": 4,
2045
+ "mode": "nvfp4",
2046
+ "global_scale": true
2047
+ },
2048
+ "language_model.model.layers.58.block_sparse_moe.switch_mlp.down_proj": {
2049
+ "group_size": 16,
2050
+ "bits": 4,
2051
+ "mode": "nvfp4",
2052
+ "global_scale": true
2053
+ },
2054
+ "language_model.model.layers.58.block_sparse_moe.switch_mlp.gate_proj": {
2055
+ "group_size": 16,
2056
+ "bits": 4,
2057
+ "mode": "nvfp4",
2058
+ "global_scale": true
2059
+ },
2060
+ "language_model.model.layers.58.block_sparse_moe.switch_mlp.up_proj": {
2061
+ "group_size": 16,
2062
+ "bits": 4,
2063
+ "mode": "nvfp4",
2064
+ "global_scale": true
2065
+ },
2066
+ "language_model.model.layers.59.block_sparse_moe.switch_mlp.down_proj": {
2067
+ "group_size": 16,
2068
+ "bits": 4,
2069
+ "mode": "nvfp4",
2070
+ "global_scale": true
2071
+ },
2072
+ "language_model.model.layers.59.block_sparse_moe.switch_mlp.gate_proj": {
2073
+ "group_size": 16,
2074
+ "bits": 4,
2075
+ "mode": "nvfp4",
2076
+ "global_scale": true
2077
+ },
2078
+ "language_model.model.layers.59.block_sparse_moe.switch_mlp.up_proj": {
2079
+ "group_size": 16,
2080
+ "bits": 4,
2081
+ "mode": "nvfp4",
2082
+ "global_scale": true
2083
+ }
2084
  },
2085
  "text_config": {
2086
  "hidden_size": 6144,