esatapedico commited on
Commit
ae8c54c
·
verified ·
1 Parent(s): d6bd9f1

Upload overrides-very-low.txt with huggingface_hub

Browse files
Files changed (1) hide show
  1. overrides-very-low.txt +1107 -0
overrides-very-low.txt ADDED
@@ -0,0 +1,1107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ blk.0.attn_gate.weight=nvfp4
2
+ blk.0.attn_norm.weight=f32
3
+ blk.0.attn_qkv.weight=nvfp4
4
+ blk.0.ffn_down.scale=f32
5
+ blk.0.ffn_down.weight=nvfp4
6
+ blk.0.ffn_gate.scale=f32
7
+ blk.0.ffn_gate.weight=nvfp4
8
+ blk.0.ffn_up.scale=f32
9
+ blk.0.ffn_up.weight=nvfp4
10
+ blk.0.post_attention_norm.weight=f32
11
+ blk.0.ssm_a=f32
12
+ blk.0.ssm_alpha.weight=nvfp4
13
+ blk.0.ssm_beta.weight=nvfp4
14
+ blk.0.ssm_conv1d.weight=f32
15
+ blk.0.ssm_dt.bias=f32
16
+ blk.0.ssm_norm.weight=f32
17
+ blk.0.ssm_out.weight=nvfp4
18
+ blk.1.attn_gate.weight=nvfp4
19
+ blk.1.attn_norm.weight=f32
20
+ blk.1.attn_qkv.weight=nvfp4
21
+ blk.1.ffn_down.scale=f32
22
+ blk.1.ffn_down.weight=nvfp4
23
+ blk.1.ffn_gate.scale=f32
24
+ blk.1.ffn_gate.weight=nvfp4
25
+ blk.1.ffn_up.scale=f32
26
+ blk.1.ffn_up.weight=nvfp4
27
+ blk.1.post_attention_norm.weight=f32
28
+ blk.1.ssm_a=f32
29
+ blk.1.ssm_alpha.weight=nvfp4
30
+ blk.1.ssm_beta.weight=nvfp4
31
+ blk.1.ssm_conv1d.weight=f32
32
+ blk.1.ssm_dt.bias=f32
33
+ blk.1.ssm_norm.weight=f32
34
+ blk.1.ssm_out.weight=nvfp4
35
+ blk.10.attn_gate.weight=nvfp4
36
+ blk.10.attn_norm.weight=f32
37
+ blk.10.attn_qkv.weight=nvfp4
38
+ blk.10.ffn_down.scale=f32
39
+ blk.10.ffn_down.weight=nvfp4
40
+ blk.10.ffn_gate.scale=f32
41
+ blk.10.ffn_gate.weight=nvfp4
42
+ blk.10.ffn_up.scale=f32
43
+ blk.10.ffn_up.weight=nvfp4
44
+ blk.10.post_attention_norm.weight=f32
45
+ blk.10.ssm_a=f32
46
+ blk.10.ssm_alpha.weight=nvfp4
47
+ blk.10.ssm_beta.weight=nvfp4
48
+ blk.10.ssm_conv1d.weight=f32
49
+ blk.10.ssm_dt.bias=f32
50
+ blk.10.ssm_norm.weight=f32
51
+ blk.10.ssm_out.weight=nvfp4
52
+ blk.11.attn_k.scale=f32
53
+ blk.11.attn_k.weight=nvfp4
54
+ blk.11.attn_k_norm.weight=f32
55
+ blk.11.attn_norm.weight=f32
56
+ blk.11.attn_output.scale=f32
57
+ blk.11.attn_output.weight=nvfp4
58
+ blk.11.attn_q.scale=f32
59
+ blk.11.attn_q.weight=nvfp4
60
+ blk.11.attn_q_norm.weight=f32
61
+ blk.11.attn_v.scale=f32
62
+ blk.11.attn_v.weight=nvfp4
63
+ blk.11.ffn_down.scale=f32
64
+ blk.11.ffn_down.weight=nvfp4
65
+ blk.11.ffn_gate.scale=f32
66
+ blk.11.ffn_gate.weight=nvfp4
67
+ blk.11.ffn_up.scale=f32
68
+ blk.11.ffn_up.weight=nvfp4
69
+ blk.11.post_attention_norm.weight=f32
70
+ blk.12.attn_gate.weight=nvfp4
71
+ blk.12.attn_norm.weight=f32
72
+ blk.12.attn_qkv.weight=nvfp4
73
+ blk.12.ffn_down.scale=f32
74
+ blk.12.ffn_down.weight=nvfp4
75
+ blk.12.ffn_gate.scale=f32
76
+ blk.12.ffn_gate.weight=nvfp4
77
+ blk.12.ffn_up.scale=f32
78
+ blk.12.ffn_up.weight=nvfp4
79
+ blk.12.post_attention_norm.weight=f32
80
+ blk.12.ssm_a=f32
81
+ blk.12.ssm_alpha.weight=nvfp4
82
+ blk.12.ssm_beta.weight=nvfp4
83
+ blk.12.ssm_conv1d.weight=f32
84
+ blk.12.ssm_dt.bias=f32
85
+ blk.12.ssm_norm.weight=f32
86
+ blk.12.ssm_out.weight=nvfp4
87
+ blk.13.attn_gate.weight=nvfp4
88
+ blk.13.attn_norm.weight=f32
89
+ blk.13.attn_qkv.weight=nvfp4
90
+ blk.13.ffn_down.scale=f32
91
+ blk.13.ffn_down.weight=nvfp4
92
+ blk.13.ffn_gate.scale=f32
93
+ blk.13.ffn_gate.weight=nvfp4
94
+ blk.13.ffn_up.scale=f32
95
+ blk.13.ffn_up.weight=nvfp4
96
+ blk.13.post_attention_norm.weight=f32
97
+ blk.13.ssm_a=f32
98
+ blk.13.ssm_alpha.weight=nvfp4
99
+ blk.13.ssm_beta.weight=nvfp4
100
+ blk.13.ssm_conv1d.weight=f32
101
+ blk.13.ssm_dt.bias=f32
102
+ blk.13.ssm_norm.weight=f32
103
+ blk.13.ssm_out.weight=nvfp4
104
+ blk.14.attn_gate.weight=nvfp4
105
+ blk.14.attn_norm.weight=f32
106
+ blk.14.attn_qkv.weight=nvfp4
107
+ blk.14.ffn_down.scale=f32
108
+ blk.14.ffn_down.weight=nvfp4
109
+ blk.14.ffn_gate.scale=f32
110
+ blk.14.ffn_gate.weight=nvfp4
111
+ blk.14.ffn_up.scale=f32
112
+ blk.14.ffn_up.weight=nvfp4
113
+ blk.14.post_attention_norm.weight=f32
114
+ blk.14.ssm_a=f32
115
+ blk.14.ssm_alpha.weight=nvfp4
116
+ blk.14.ssm_beta.weight=nvfp4
117
+ blk.14.ssm_conv1d.weight=f32
118
+ blk.14.ssm_dt.bias=f32
119
+ blk.14.ssm_norm.weight=f32
120
+ blk.14.ssm_out.weight=nvfp4
121
+ blk.15.attn_k.scale=f32
122
+ blk.15.attn_k.weight=nvfp4
123
+ blk.15.attn_k_norm.weight=f32
124
+ blk.15.attn_norm.weight=f32
125
+ blk.15.attn_output.scale=f32
126
+ blk.15.attn_output.weight=nvfp4
127
+ blk.15.attn_q.scale=f32
128
+ blk.15.attn_q.weight=nvfp4
129
+ blk.15.attn_q_norm.weight=f32
130
+ blk.15.attn_v.scale=f32
131
+ blk.15.attn_v.weight=nvfp4
132
+ blk.15.ffn_down.scale=f32
133
+ blk.15.ffn_down.weight=nvfp4
134
+ blk.15.ffn_gate.scale=f32
135
+ blk.15.ffn_gate.weight=nvfp4
136
+ blk.15.ffn_up.scale=f32
137
+ blk.15.ffn_up.weight=nvfp4
138
+ blk.15.post_attention_norm.weight=f32
139
+ blk.16.attn_gate.weight=nvfp4
140
+ blk.16.attn_norm.weight=f32
141
+ blk.16.attn_qkv.weight=nvfp4
142
+ blk.16.ffn_down.scale=f32
143
+ blk.16.ffn_down.weight=nvfp4
144
+ blk.16.ffn_gate.scale=f32
145
+ blk.16.ffn_gate.weight=nvfp4
146
+ blk.16.ffn_up.scale=f32
147
+ blk.16.ffn_up.weight=nvfp4
148
+ blk.16.post_attention_norm.weight=f32
149
+ blk.16.ssm_a=f32
150
+ blk.16.ssm_alpha.weight=nvfp4
151
+ blk.16.ssm_beta.weight=nvfp4
152
+ blk.16.ssm_conv1d.weight=f32
153
+ blk.16.ssm_dt.bias=f32
154
+ blk.16.ssm_norm.weight=f32
155
+ blk.16.ssm_out.weight=nvfp4
156
+ blk.17.attn_gate.weight=nvfp4
157
+ blk.17.attn_norm.weight=f32
158
+ blk.17.attn_qkv.weight=nvfp4
159
+ blk.17.ffn_down.scale=f32
160
+ blk.17.ffn_down.weight=nvfp4
161
+ blk.17.ffn_gate.scale=f32
162
+ blk.17.ffn_gate.weight=nvfp4
163
+ blk.17.ffn_up.scale=f32
164
+ blk.17.ffn_up.weight=nvfp4
165
+ blk.17.post_attention_norm.weight=f32
166
+ blk.17.ssm_a=f32
167
+ blk.17.ssm_alpha.weight=nvfp4
168
+ blk.17.ssm_beta.weight=nvfp4
169
+ blk.17.ssm_conv1d.weight=f32
170
+ blk.17.ssm_dt.bias=f32
171
+ blk.17.ssm_norm.weight=f32
172
+ blk.17.ssm_out.weight=nvfp4
173
+ blk.18.attn_gate.weight=nvfp4
174
+ blk.18.attn_norm.weight=f32
175
+ blk.18.attn_qkv.weight=nvfp4
176
+ blk.18.ffn_down.scale=f32
177
+ blk.18.ffn_down.weight=nvfp4
178
+ blk.18.ffn_gate.scale=f32
179
+ blk.18.ffn_gate.weight=nvfp4
180
+ blk.18.ffn_up.scale=f32
181
+ blk.18.ffn_up.weight=nvfp4
182
+ blk.18.post_attention_norm.weight=f32
183
+ blk.18.ssm_a=f32
184
+ blk.18.ssm_alpha.weight=nvfp4
185
+ blk.18.ssm_beta.weight=nvfp4
186
+ blk.18.ssm_conv1d.weight=f32
187
+ blk.18.ssm_dt.bias=f32
188
+ blk.18.ssm_norm.weight=f32
189
+ blk.18.ssm_out.weight=nvfp4
190
+ blk.19.attn_k.scale=f32
191
+ blk.19.attn_k.weight=nvfp4
192
+ blk.19.attn_k_norm.weight=f32
193
+ blk.19.attn_norm.weight=f32
194
+ blk.19.attn_output.scale=f32
195
+ blk.19.attn_output.weight=nvfp4
196
+ blk.19.attn_q.scale=f32
197
+ blk.19.attn_q.weight=nvfp4
198
+ blk.19.attn_q_norm.weight=f32
199
+ blk.19.attn_v.scale=f32
200
+ blk.19.attn_v.weight=nvfp4
201
+ blk.19.ffn_down.scale=f32
202
+ blk.19.ffn_down.weight=nvfp4
203
+ blk.19.ffn_gate.scale=f32
204
+ blk.19.ffn_gate.weight=nvfp4
205
+ blk.19.ffn_up.scale=f32
206
+ blk.19.ffn_up.weight=nvfp4
207
+ blk.19.post_attention_norm.weight=f32
208
+ blk.2.attn_gate.weight=nvfp4
209
+ blk.2.attn_norm.weight=f32
210
+ blk.2.attn_qkv.weight=nvfp4
211
+ blk.2.ffn_down.scale=f32
212
+ blk.2.ffn_down.weight=nvfp4
213
+ blk.2.ffn_gate.scale=f32
214
+ blk.2.ffn_gate.weight=nvfp4
215
+ blk.2.ffn_up.scale=f32
216
+ blk.2.ffn_up.weight=nvfp4
217
+ blk.2.post_attention_norm.weight=f32
218
+ blk.2.ssm_a=f32
219
+ blk.2.ssm_alpha.weight=nvfp4
220
+ blk.2.ssm_beta.weight=nvfp4
221
+ blk.2.ssm_conv1d.weight=f32
222
+ blk.2.ssm_dt.bias=f32
223
+ blk.2.ssm_norm.weight=f32
224
+ blk.2.ssm_out.weight=nvfp4
225
+ blk.20.attn_gate.weight=nvfp4
226
+ blk.20.attn_norm.weight=f32
227
+ blk.20.attn_qkv.weight=nvfp4
228
+ blk.20.ffn_down.scale=f32
229
+ blk.20.ffn_down.weight=nvfp4
230
+ blk.20.ffn_gate.scale=f32
231
+ blk.20.ffn_gate.weight=nvfp4
232
+ blk.20.ffn_up.scale=f32
233
+ blk.20.ffn_up.weight=nvfp4
234
+ blk.20.post_attention_norm.weight=f32
235
+ blk.20.ssm_a=f32
236
+ blk.20.ssm_alpha.weight=nvfp4
237
+ blk.20.ssm_beta.weight=nvfp4
238
+ blk.20.ssm_conv1d.weight=f32
239
+ blk.20.ssm_dt.bias=f32
240
+ blk.20.ssm_norm.weight=f32
241
+ blk.20.ssm_out.weight=nvfp4
242
+ blk.21.attn_gate.weight=nvfp4
243
+ blk.21.attn_norm.weight=f32
244
+ blk.21.attn_qkv.weight=nvfp4
245
+ blk.21.ffn_down.scale=f32
246
+ blk.21.ffn_down.weight=nvfp4
247
+ blk.21.ffn_gate.scale=f32
248
+ blk.21.ffn_gate.weight=nvfp4
249
+ blk.21.ffn_up.scale=f32
250
+ blk.21.ffn_up.weight=nvfp4
251
+ blk.21.post_attention_norm.weight=f32
252
+ blk.21.ssm_a=f32
253
+ blk.21.ssm_alpha.weight=nvfp4
254
+ blk.21.ssm_beta.weight=nvfp4
255
+ blk.21.ssm_conv1d.weight=f32
256
+ blk.21.ssm_dt.bias=f32
257
+ blk.21.ssm_norm.weight=f32
258
+ blk.21.ssm_out.weight=nvfp4
259
+ blk.22.attn_gate.weight=nvfp4
260
+ blk.22.attn_norm.weight=f32
261
+ blk.22.attn_qkv.weight=nvfp4
262
+ blk.22.ffn_down.scale=f32
263
+ blk.22.ffn_down.weight=nvfp4
264
+ blk.22.ffn_gate.scale=f32
265
+ blk.22.ffn_gate.weight=nvfp4
266
+ blk.22.ffn_up.scale=f32
267
+ blk.22.ffn_up.weight=nvfp4
268
+ blk.22.post_attention_norm.weight=f32
269
+ blk.22.ssm_a=f32
270
+ blk.22.ssm_alpha.weight=nvfp4
271
+ blk.22.ssm_beta.weight=nvfp4
272
+ blk.22.ssm_conv1d.weight=f32
273
+ blk.22.ssm_dt.bias=f32
274
+ blk.22.ssm_norm.weight=f32
275
+ blk.22.ssm_out.weight=nvfp4
276
+ blk.23.attn_k.scale=f32
277
+ blk.23.attn_k.weight=nvfp4
278
+ blk.23.attn_k_norm.weight=f32
279
+ blk.23.attn_norm.weight=f32
280
+ blk.23.attn_output.scale=f32
281
+ blk.23.attn_output.weight=nvfp4
282
+ blk.23.attn_q.scale=f32
283
+ blk.23.attn_q.weight=nvfp4
284
+ blk.23.attn_q_norm.weight=f32
285
+ blk.23.attn_v.scale=f32
286
+ blk.23.attn_v.weight=nvfp4
287
+ blk.23.ffn_down.scale=f32
288
+ blk.23.ffn_down.weight=nvfp4
289
+ blk.23.ffn_gate.scale=f32
290
+ blk.23.ffn_gate.weight=nvfp4
291
+ blk.23.ffn_up.scale=f32
292
+ blk.23.ffn_up.weight=nvfp4
293
+ blk.23.post_attention_norm.weight=f32
294
+ blk.24.attn_gate.weight=nvfp4
295
+ blk.24.attn_norm.weight=f32
296
+ blk.24.attn_qkv.weight=nvfp4
297
+ blk.24.ffn_down.scale=f32
298
+ blk.24.ffn_down.weight=nvfp4
299
+ blk.24.ffn_gate.scale=f32
300
+ blk.24.ffn_gate.weight=nvfp4
301
+ blk.24.ffn_up.scale=f32
302
+ blk.24.ffn_up.weight=nvfp4
303
+ blk.24.post_attention_norm.weight=f32
304
+ blk.24.ssm_a=f32
305
+ blk.24.ssm_alpha.weight=nvfp4
306
+ blk.24.ssm_beta.weight=nvfp4
307
+ blk.24.ssm_conv1d.weight=f32
308
+ blk.24.ssm_dt.bias=f32
309
+ blk.24.ssm_norm.weight=f32
310
+ blk.24.ssm_out.weight=nvfp4
311
+ blk.25.attn_gate.weight=nvfp4
312
+ blk.25.attn_norm.weight=f32
313
+ blk.25.attn_qkv.weight=nvfp4
314
+ blk.25.ffn_down.scale=f32
315
+ blk.25.ffn_down.weight=nvfp4
316
+ blk.25.ffn_gate.scale=f32
317
+ blk.25.ffn_gate.weight=nvfp4
318
+ blk.25.ffn_up.scale=f32
319
+ blk.25.ffn_up.weight=nvfp4
320
+ blk.25.post_attention_norm.weight=f32
321
+ blk.25.ssm_a=f32
322
+ blk.25.ssm_alpha.weight=nvfp4
323
+ blk.25.ssm_beta.weight=nvfp4
324
+ blk.25.ssm_conv1d.weight=f32
325
+ blk.25.ssm_dt.bias=f32
326
+ blk.25.ssm_norm.weight=f32
327
+ blk.25.ssm_out.weight=nvfp4
328
+ blk.26.attn_gate.weight=nvfp4
329
+ blk.26.attn_norm.weight=f32
330
+ blk.26.attn_qkv.weight=nvfp4
331
+ blk.26.ffn_down.scale=f32
332
+ blk.26.ffn_down.weight=nvfp4
333
+ blk.26.ffn_gate.scale=f32
334
+ blk.26.ffn_gate.weight=nvfp4
335
+ blk.26.ffn_up.scale=f32
336
+ blk.26.ffn_up.weight=nvfp4
337
+ blk.26.post_attention_norm.weight=f32
338
+ blk.26.ssm_a=f32
339
+ blk.26.ssm_alpha.weight=nvfp4
340
+ blk.26.ssm_beta.weight=nvfp4
341
+ blk.26.ssm_conv1d.weight=f32
342
+ blk.26.ssm_dt.bias=f32
343
+ blk.26.ssm_norm.weight=f32
344
+ blk.26.ssm_out.weight=nvfp4
345
+ blk.27.attn_k.scale=f32
346
+ blk.27.attn_k.weight=nvfp4
347
+ blk.27.attn_k_norm.weight=f32
348
+ blk.27.attn_norm.weight=f32
349
+ blk.27.attn_output.scale=f32
350
+ blk.27.attn_output.weight=nvfp4
351
+ blk.27.attn_q.scale=f32
352
+ blk.27.attn_q.weight=nvfp4
353
+ blk.27.attn_q_norm.weight=f32
354
+ blk.27.attn_v.scale=f32
355
+ blk.27.attn_v.weight=nvfp4
356
+ blk.27.ffn_down.scale=f32
357
+ blk.27.ffn_down.weight=nvfp4
358
+ blk.27.ffn_gate.scale=f32
359
+ blk.27.ffn_gate.weight=nvfp4
360
+ blk.27.ffn_up.scale=f32
361
+ blk.27.ffn_up.weight=nvfp4
362
+ blk.27.post_attention_norm.weight=f32
363
+ blk.28.attn_gate.weight=nvfp4
364
+ blk.28.attn_norm.weight=f32
365
+ blk.28.attn_qkv.weight=nvfp4
366
+ blk.28.ffn_down.scale=f32
367
+ blk.28.ffn_down.weight=nvfp4
368
+ blk.28.ffn_gate.scale=f32
369
+ blk.28.ffn_gate.weight=nvfp4
370
+ blk.28.ffn_up.scale=f32
371
+ blk.28.ffn_up.weight=nvfp4
372
+ blk.28.post_attention_norm.weight=f32
373
+ blk.28.ssm_a=f32
374
+ blk.28.ssm_alpha.weight=nvfp4
375
+ blk.28.ssm_beta.weight=nvfp4
376
+ blk.28.ssm_conv1d.weight=f32
377
+ blk.28.ssm_dt.bias=f32
378
+ blk.28.ssm_norm.weight=f32
379
+ blk.28.ssm_out.weight=nvfp4
380
+ blk.29.attn_gate.weight=nvfp4
381
+ blk.29.attn_norm.weight=f32
382
+ blk.29.attn_qkv.weight=nvfp4
383
+ blk.29.ffn_down.scale=f32
384
+ blk.29.ffn_down.weight=nvfp4
385
+ blk.29.ffn_gate.scale=f32
386
+ blk.29.ffn_gate.weight=nvfp4
387
+ blk.29.ffn_up.scale=f32
388
+ blk.29.ffn_up.weight=nvfp4
389
+ blk.29.post_attention_norm.weight=f32
390
+ blk.29.ssm_a=f32
391
+ blk.29.ssm_alpha.weight=nvfp4
392
+ blk.29.ssm_beta.weight=nvfp4
393
+ blk.29.ssm_conv1d.weight=f32
394
+ blk.29.ssm_dt.bias=f32
395
+ blk.29.ssm_norm.weight=f32
396
+ blk.29.ssm_out.weight=nvfp4
397
+ blk.3.attn_k.scale=f32
398
+ blk.3.attn_k.weight=nvfp4
399
+ blk.3.attn_k_norm.weight=f32
400
+ blk.3.attn_norm.weight=f32
401
+ blk.3.attn_output.scale=f32
402
+ blk.3.attn_output.weight=nvfp4
403
+ blk.3.attn_q.scale=f32
404
+ blk.3.attn_q.weight=nvfp4
405
+ blk.3.attn_q_norm.weight=f32
406
+ blk.3.attn_v.scale=f32
407
+ blk.3.attn_v.weight=nvfp4
408
+ blk.3.ffn_down.scale=f32
409
+ blk.3.ffn_down.weight=nvfp4
410
+ blk.3.ffn_gate.scale=f32
411
+ blk.3.ffn_gate.weight=nvfp4
412
+ blk.3.ffn_up.scale=f32
413
+ blk.3.ffn_up.weight=nvfp4
414
+ blk.3.post_attention_norm.weight=f32
415
+ blk.30.attn_gate.weight=nvfp4
416
+ blk.30.attn_norm.weight=f32
417
+ blk.30.attn_qkv.weight=nvfp4
418
+ blk.30.ffn_down.scale=f32
419
+ blk.30.ffn_down.weight=nvfp4
420
+ blk.30.ffn_gate.scale=f32
421
+ blk.30.ffn_gate.weight=nvfp4
422
+ blk.30.ffn_up.scale=f32
423
+ blk.30.ffn_up.weight=nvfp4
424
+ blk.30.post_attention_norm.weight=f32
425
+ blk.30.ssm_a=f32
426
+ blk.30.ssm_alpha.weight=nvfp4
427
+ blk.30.ssm_beta.weight=nvfp4
428
+ blk.30.ssm_conv1d.weight=f32
429
+ blk.30.ssm_dt.bias=f32
430
+ blk.30.ssm_norm.weight=f32
431
+ blk.30.ssm_out.weight=nvfp4
432
+ blk.31.attn_k.scale=f32
433
+ blk.31.attn_k.weight=nvfp4
434
+ blk.31.attn_k_norm.weight=f32
435
+ blk.31.attn_norm.weight=f32
436
+ blk.31.attn_output.scale=f32
437
+ blk.31.attn_output.weight=nvfp4
438
+ blk.31.attn_q.scale=f32
439
+ blk.31.attn_q.weight=nvfp4
440
+ blk.31.attn_q_norm.weight=f32
441
+ blk.31.attn_v.scale=f32
442
+ blk.31.attn_v.weight=nvfp4
443
+ blk.31.ffn_down.scale=f32
444
+ blk.31.ffn_down.weight=nvfp4
445
+ blk.31.ffn_gate.scale=f32
446
+ blk.31.ffn_gate.weight=nvfp4
447
+ blk.31.ffn_up.scale=f32
448
+ blk.31.ffn_up.weight=nvfp4
449
+ blk.31.post_attention_norm.weight=f32
450
+ blk.32.attn_gate.weight=nvfp4
451
+ blk.32.attn_norm.weight=f32
452
+ blk.32.attn_qkv.weight=nvfp4
453
+ blk.32.ffn_down.scale=f32
454
+ blk.32.ffn_down.weight=nvfp4
455
+ blk.32.ffn_gate.scale=f32
456
+ blk.32.ffn_gate.weight=nvfp4
457
+ blk.32.ffn_up.scale=f32
458
+ blk.32.ffn_up.weight=nvfp4
459
+ blk.32.post_attention_norm.weight=f32
460
+ blk.32.ssm_a=f32
461
+ blk.32.ssm_alpha.weight=nvfp4
462
+ blk.32.ssm_beta.weight=nvfp4
463
+ blk.32.ssm_conv1d.weight=f32
464
+ blk.32.ssm_dt.bias=f32
465
+ blk.32.ssm_norm.weight=f32
466
+ blk.32.ssm_out.weight=nvfp4
467
+ blk.33.attn_gate.weight=nvfp4
468
+ blk.33.attn_norm.weight=f32
469
+ blk.33.attn_qkv.weight=nvfp4
470
+ blk.33.ffn_down.scale=f32
471
+ blk.33.ffn_down.weight=nvfp4
472
+ blk.33.ffn_gate.scale=f32
473
+ blk.33.ffn_gate.weight=nvfp4
474
+ blk.33.ffn_up.scale=f32
475
+ blk.33.ffn_up.weight=nvfp4
476
+ blk.33.post_attention_norm.weight=f32
477
+ blk.33.ssm_a=f32
478
+ blk.33.ssm_alpha.weight=nvfp4
479
+ blk.33.ssm_beta.weight=nvfp4
480
+ blk.33.ssm_conv1d.weight=f32
481
+ blk.33.ssm_dt.bias=f32
482
+ blk.33.ssm_norm.weight=f32
483
+ blk.33.ssm_out.weight=nvfp4
484
+ blk.34.attn_gate.weight=nvfp4
485
+ blk.34.attn_norm.weight=f32
486
+ blk.34.attn_qkv.weight=nvfp4
487
+ blk.34.ffn_down.scale=f32
488
+ blk.34.ffn_down.weight=nvfp4
489
+ blk.34.ffn_gate.scale=f32
490
+ blk.34.ffn_gate.weight=nvfp4
491
+ blk.34.ffn_up.scale=f32
492
+ blk.34.ffn_up.weight=nvfp4
493
+ blk.34.post_attention_norm.weight=f32
494
+ blk.34.ssm_a=f32
495
+ blk.34.ssm_alpha.weight=nvfp4
496
+ blk.34.ssm_beta.weight=nvfp4
497
+ blk.34.ssm_conv1d.weight=f32
498
+ blk.34.ssm_dt.bias=f32
499
+ blk.34.ssm_norm.weight=f32
500
+ blk.34.ssm_out.weight=nvfp4
501
+ blk.35.attn_k.scale=f32
502
+ blk.35.attn_k.weight=nvfp4
503
+ blk.35.attn_k_norm.weight=f32
504
+ blk.35.attn_norm.weight=f32
505
+ blk.35.attn_output.scale=f32
506
+ blk.35.attn_output.weight=nvfp4
507
+ blk.35.attn_q.scale=f32
508
+ blk.35.attn_q.weight=nvfp4
509
+ blk.35.attn_q_norm.weight=f32
510
+ blk.35.attn_v.scale=f32
511
+ blk.35.attn_v.weight=nvfp4
512
+ blk.35.ffn_down.scale=f32
513
+ blk.35.ffn_down.weight=nvfp4
514
+ blk.35.ffn_gate.scale=f32
515
+ blk.35.ffn_gate.weight=nvfp4
516
+ blk.35.ffn_up.scale=f32
517
+ blk.35.ffn_up.weight=nvfp4
518
+ blk.35.post_attention_norm.weight=f32
519
+ blk.36.attn_gate.weight=nvfp4
520
+ blk.36.attn_norm.weight=f32
521
+ blk.36.attn_qkv.weight=nvfp4
522
+ blk.36.ffn_down.scale=f32
523
+ blk.36.ffn_down.weight=nvfp4
524
+ blk.36.ffn_gate.scale=f32
525
+ blk.36.ffn_gate.weight=nvfp4
526
+ blk.36.ffn_up.scale=f32
527
+ blk.36.ffn_up.weight=nvfp4
528
+ blk.36.post_attention_norm.weight=f32
529
+ blk.36.ssm_a=f32
530
+ blk.36.ssm_alpha.weight=nvfp4
531
+ blk.36.ssm_beta.weight=nvfp4
532
+ blk.36.ssm_conv1d.weight=f32
533
+ blk.36.ssm_dt.bias=f32
534
+ blk.36.ssm_norm.weight=f32
535
+ blk.36.ssm_out.weight=nvfp4
536
+ blk.37.attn_gate.weight=nvfp4
537
+ blk.37.attn_norm.weight=f32
538
+ blk.37.attn_qkv.weight=nvfp4
539
+ blk.37.ffn_down.scale=f32
540
+ blk.37.ffn_down.weight=nvfp4
541
+ blk.37.ffn_gate.scale=f32
542
+ blk.37.ffn_gate.weight=nvfp4
543
+ blk.37.ffn_up.scale=f32
544
+ blk.37.ffn_up.weight=nvfp4
545
+ blk.37.post_attention_norm.weight=f32
546
+ blk.37.ssm_a=f32
547
+ blk.37.ssm_alpha.weight=nvfp4
548
+ blk.37.ssm_beta.weight=nvfp4
549
+ blk.37.ssm_conv1d.weight=f32
550
+ blk.37.ssm_dt.bias=f32
551
+ blk.37.ssm_norm.weight=f32
552
+ blk.37.ssm_out.weight=nvfp4
553
+ blk.38.attn_gate.weight=nvfp4
554
+ blk.38.attn_norm.weight=f32
555
+ blk.38.attn_qkv.weight=nvfp4
556
+ blk.38.ffn_down.scale=f32
557
+ blk.38.ffn_down.weight=nvfp4
558
+ blk.38.ffn_gate.scale=f32
559
+ blk.38.ffn_gate.weight=nvfp4
560
+ blk.38.ffn_up.scale=f32
561
+ blk.38.ffn_up.weight=nvfp4
562
+ blk.38.post_attention_norm.weight=f32
563
+ blk.38.ssm_a=f32
564
+ blk.38.ssm_alpha.weight=nvfp4
565
+ blk.38.ssm_beta.weight=nvfp4
566
+ blk.38.ssm_conv1d.weight=f32
567
+ blk.38.ssm_dt.bias=f32
568
+ blk.38.ssm_norm.weight=f32
569
+ blk.38.ssm_out.weight=nvfp4
570
+ blk.39.attn_k.scale=f32
571
+ blk.39.attn_k.weight=nvfp4
572
+ blk.39.attn_k_norm.weight=f32
573
+ blk.39.attn_norm.weight=f32
574
+ blk.39.attn_output.scale=f32
575
+ blk.39.attn_output.weight=nvfp4
576
+ blk.39.attn_q.scale=f32
577
+ blk.39.attn_q.weight=nvfp4
578
+ blk.39.attn_q_norm.weight=f32
579
+ blk.39.attn_v.scale=f32
580
+ blk.39.attn_v.weight=nvfp4
581
+ blk.39.ffn_down.scale=f32
582
+ blk.39.ffn_down.weight=nvfp4
583
+ blk.39.ffn_gate.scale=f32
584
+ blk.39.ffn_gate.weight=nvfp4
585
+ blk.39.ffn_up.scale=f32
586
+ blk.39.ffn_up.weight=nvfp4
587
+ blk.39.post_attention_norm.weight=f32
588
+ blk.4.attn_gate.weight=nvfp4
589
+ blk.4.attn_norm.weight=f32
590
+ blk.4.attn_qkv.weight=nvfp4
591
+ blk.4.ffn_down.scale=f32
592
+ blk.4.ffn_down.weight=nvfp4
593
+ blk.4.ffn_gate.scale=f32
594
+ blk.4.ffn_gate.weight=nvfp4
595
+ blk.4.ffn_up.scale=f32
596
+ blk.4.ffn_up.weight=nvfp4
597
+ blk.4.post_attention_norm.weight=f32
598
+ blk.4.ssm_a=f32
599
+ blk.4.ssm_alpha.weight=nvfp4
600
+ blk.4.ssm_beta.weight=nvfp4
601
+ blk.4.ssm_conv1d.weight=f32
602
+ blk.4.ssm_dt.bias=f32
603
+ blk.4.ssm_norm.weight=f32
604
+ blk.4.ssm_out.weight=nvfp4
605
+ blk.40.attn_gate.weight=nvfp4
606
+ blk.40.attn_norm.weight=f32
607
+ blk.40.attn_qkv.weight=nvfp4
608
+ blk.40.ffn_down.scale=f32
609
+ blk.40.ffn_down.weight=nvfp4
610
+ blk.40.ffn_gate.scale=f32
611
+ blk.40.ffn_gate.weight=nvfp4
612
+ blk.40.ffn_up.scale=f32
613
+ blk.40.ffn_up.weight=nvfp4
614
+ blk.40.post_attention_norm.weight=f32
615
+ blk.40.ssm_a=f32
616
+ blk.40.ssm_alpha.weight=nvfp4
617
+ blk.40.ssm_beta.weight=nvfp4
618
+ blk.40.ssm_conv1d.weight=f32
619
+ blk.40.ssm_dt.bias=f32
620
+ blk.40.ssm_norm.weight=f32
621
+ blk.40.ssm_out.weight=nvfp4
622
+ blk.41.attn_gate.weight=nvfp4
623
+ blk.41.attn_norm.weight=f32
624
+ blk.41.attn_qkv.weight=nvfp4
625
+ blk.41.ffn_down.scale=f32
626
+ blk.41.ffn_down.weight=nvfp4
627
+ blk.41.ffn_gate.scale=f32
628
+ blk.41.ffn_gate.weight=nvfp4
629
+ blk.41.ffn_up.scale=f32
630
+ blk.41.ffn_up.weight=nvfp4
631
+ blk.41.post_attention_norm.weight=f32
632
+ blk.41.ssm_a=f32
633
+ blk.41.ssm_alpha.weight=nvfp4
634
+ blk.41.ssm_beta.weight=nvfp4
635
+ blk.41.ssm_conv1d.weight=f32
636
+ blk.41.ssm_dt.bias=f32
637
+ blk.41.ssm_norm.weight=f32
638
+ blk.41.ssm_out.weight=nvfp4
639
+ blk.42.attn_gate.weight=nvfp4
640
+ blk.42.attn_norm.weight=f32
641
+ blk.42.attn_qkv.weight=nvfp4
642
+ blk.42.ffn_down.scale=f32
643
+ blk.42.ffn_down.weight=nvfp4
644
+ blk.42.ffn_gate.scale=f32
645
+ blk.42.ffn_gate.weight=nvfp4
646
+ blk.42.ffn_up.scale=f32
647
+ blk.42.ffn_up.weight=nvfp4
648
+ blk.42.post_attention_norm.weight=f32
649
+ blk.42.ssm_a=f32
650
+ blk.42.ssm_alpha.weight=nvfp4
651
+ blk.42.ssm_beta.weight=nvfp4
652
+ blk.42.ssm_conv1d.weight=f32
653
+ blk.42.ssm_dt.bias=f32
654
+ blk.42.ssm_norm.weight=f32
655
+ blk.42.ssm_out.weight=nvfp4
656
+ blk.43.attn_k.scale=f32
657
+ blk.43.attn_k.weight=nvfp4
658
+ blk.43.attn_k_norm.weight=f32
659
+ blk.43.attn_norm.weight=f32
660
+ blk.43.attn_output.scale=f32
661
+ blk.43.attn_output.weight=nvfp4
662
+ blk.43.attn_q.scale=f32
663
+ blk.43.attn_q.weight=nvfp4
664
+ blk.43.attn_q_norm.weight=f32
665
+ blk.43.attn_v.scale=f32
666
+ blk.43.attn_v.weight=nvfp4
667
+ blk.43.ffn_down.scale=f32
668
+ blk.43.ffn_down.weight=nvfp4
669
+ blk.43.ffn_gate.scale=f32
670
+ blk.43.ffn_gate.weight=nvfp4
671
+ blk.43.ffn_up.scale=f32
672
+ blk.43.ffn_up.weight=nvfp4
673
+ blk.43.post_attention_norm.weight=f32
674
+ blk.44.attn_gate.weight=nvfp4
675
+ blk.44.attn_norm.weight=f32
676
+ blk.44.attn_qkv.weight=nvfp4
677
+ blk.44.ffn_down.scale=f32
678
+ blk.44.ffn_down.weight=nvfp4
679
+ blk.44.ffn_gate.scale=f32
680
+ blk.44.ffn_gate.weight=nvfp4
681
+ blk.44.ffn_up.scale=f32
682
+ blk.44.ffn_up.weight=nvfp4
683
+ blk.44.post_attention_norm.weight=f32
684
+ blk.44.ssm_a=f32
685
+ blk.44.ssm_alpha.weight=nvfp4
686
+ blk.44.ssm_beta.weight=nvfp4
687
+ blk.44.ssm_conv1d.weight=f32
688
+ blk.44.ssm_dt.bias=f32
689
+ blk.44.ssm_norm.weight=f32
690
+ blk.44.ssm_out.weight=nvfp4
691
+ blk.45.attn_gate.weight=nvfp4
692
+ blk.45.attn_norm.weight=f32
693
+ blk.45.attn_qkv.weight=nvfp4
694
+ blk.45.ffn_down.scale=f32
695
+ blk.45.ffn_down.weight=nvfp4
696
+ blk.45.ffn_gate.scale=f32
697
+ blk.45.ffn_gate.weight=nvfp4
698
+ blk.45.ffn_up.scale=f32
699
+ blk.45.ffn_up.weight=nvfp4
700
+ blk.45.post_attention_norm.weight=f32
701
+ blk.45.ssm_a=f32
702
+ blk.45.ssm_alpha.weight=nvfp4
703
+ blk.45.ssm_beta.weight=nvfp4
704
+ blk.45.ssm_conv1d.weight=f32
705
+ blk.45.ssm_dt.bias=f32
706
+ blk.45.ssm_norm.weight=f32
707
+ blk.45.ssm_out.weight=nvfp4
708
+ blk.46.attn_gate.weight=nvfp4
709
+ blk.46.attn_norm.weight=f32
710
+ blk.46.attn_qkv.weight=nvfp4
711
+ blk.46.ffn_down.scale=f32
712
+ blk.46.ffn_down.weight=nvfp4
713
+ blk.46.ffn_gate.scale=f32
714
+ blk.46.ffn_gate.weight=nvfp4
715
+ blk.46.ffn_up.scale=f32
716
+ blk.46.ffn_up.weight=nvfp4
717
+ blk.46.post_attention_norm.weight=f32
718
+ blk.46.ssm_a=f32
719
+ blk.46.ssm_alpha.weight=nvfp4
720
+ blk.46.ssm_beta.weight=nvfp4
721
+ blk.46.ssm_conv1d.weight=f32
722
+ blk.46.ssm_dt.bias=f32
723
+ blk.46.ssm_norm.weight=f32
724
+ blk.46.ssm_out.weight=nvfp4
725
+ blk.47.attn_k.scale=f32
726
+ blk.47.attn_k.weight=nvfp4
727
+ blk.47.attn_k_norm.weight=f32
728
+ blk.47.attn_norm.weight=f32
729
+ blk.47.attn_output.scale=f32
730
+ blk.47.attn_output.weight=nvfp4
731
+ blk.47.attn_q.scale=f32
732
+ blk.47.attn_q.weight=nvfp4
733
+ blk.47.attn_q_norm.weight=f32
734
+ blk.47.attn_v.scale=f32
735
+ blk.47.attn_v.weight=nvfp4
736
+ blk.47.ffn_down.scale=f32
737
+ blk.47.ffn_down.weight=nvfp4
738
+ blk.47.ffn_gate.scale=f32
739
+ blk.47.ffn_gate.weight=nvfp4
740
+ blk.47.ffn_up.scale=f32
741
+ blk.47.ffn_up.weight=nvfp4
742
+ blk.47.post_attention_norm.weight=f32
743
+ blk.48.attn_gate.weight=nvfp4
744
+ blk.48.attn_norm.weight=f32
745
+ blk.48.attn_qkv.weight=nvfp4
746
+ blk.48.ffn_down.scale=f32
747
+ blk.48.ffn_down.weight=nvfp4
748
+ blk.48.ffn_gate.scale=f32
749
+ blk.48.ffn_gate.weight=nvfp4
750
+ blk.48.ffn_up.scale=f32
751
+ blk.48.ffn_up.weight=nvfp4
752
+ blk.48.post_attention_norm.weight=f32
753
+ blk.48.ssm_a=f32
754
+ blk.48.ssm_alpha.weight=nvfp4
755
+ blk.48.ssm_beta.weight=nvfp4
756
+ blk.48.ssm_conv1d.weight=f32
757
+ blk.48.ssm_dt.bias=f32
758
+ blk.48.ssm_norm.weight=f32
759
+ blk.48.ssm_out.weight=nvfp4
760
+ blk.49.attn_gate.weight=nvfp4
761
+ blk.49.attn_norm.weight=f32
762
+ blk.49.attn_qkv.weight=nvfp4
763
+ blk.49.ffn_down.scale=f32
764
+ blk.49.ffn_down.weight=nvfp4
765
+ blk.49.ffn_gate.scale=f32
766
+ blk.49.ffn_gate.weight=nvfp4
767
+ blk.49.ffn_up.scale=f32
768
+ blk.49.ffn_up.weight=nvfp4
769
+ blk.49.post_attention_norm.weight=f32
770
+ blk.49.ssm_a=f32
771
+ blk.49.ssm_alpha.weight=nvfp4
772
+ blk.49.ssm_beta.weight=nvfp4
773
+ blk.49.ssm_conv1d.weight=f32
774
+ blk.49.ssm_dt.bias=f32
775
+ blk.49.ssm_norm.weight=f32
776
+ blk.49.ssm_out.weight=nvfp4
777
+ blk.5.attn_gate.weight=nvfp4
778
+ blk.5.attn_norm.weight=f32
779
+ blk.5.attn_qkv.weight=nvfp4
780
+ blk.5.ffn_down.scale=f32
781
+ blk.5.ffn_down.weight=nvfp4
782
+ blk.5.ffn_gate.scale=f32
783
+ blk.5.ffn_gate.weight=nvfp4
784
+ blk.5.ffn_up.scale=f32
785
+ blk.5.ffn_up.weight=nvfp4
786
+ blk.5.post_attention_norm.weight=f32
787
+ blk.5.ssm_a=f32
788
+ blk.5.ssm_alpha.weight=nvfp4
789
+ blk.5.ssm_beta.weight=nvfp4
790
+ blk.5.ssm_conv1d.weight=f32
791
+ blk.5.ssm_dt.bias=f32
792
+ blk.5.ssm_norm.weight=f32
793
+ blk.5.ssm_out.weight=nvfp4
794
+ blk.50.attn_gate.weight=nvfp4
795
+ blk.50.attn_norm.weight=f32
796
+ blk.50.attn_qkv.weight=nvfp4
797
+ blk.50.ffn_down.scale=f32
798
+ blk.50.ffn_down.weight=nvfp4
799
+ blk.50.ffn_gate.scale=f32
800
+ blk.50.ffn_gate.weight=nvfp4
801
+ blk.50.ffn_up.scale=f32
802
+ blk.50.ffn_up.weight=nvfp4
803
+ blk.50.post_attention_norm.weight=f32
804
+ blk.50.ssm_a=f32
805
+ blk.50.ssm_alpha.weight=nvfp4
806
+ blk.50.ssm_beta.weight=nvfp4
807
+ blk.50.ssm_conv1d.weight=f32
808
+ blk.50.ssm_dt.bias=f32
809
+ blk.50.ssm_norm.weight=f32
810
+ blk.50.ssm_out.weight=nvfp4
811
+ blk.51.attn_k.scale=f32
812
+ blk.51.attn_k.weight=nvfp4
813
+ blk.51.attn_k_norm.weight=f32
814
+ blk.51.attn_norm.weight=f32
815
+ blk.51.attn_output.scale=f32
816
+ blk.51.attn_output.weight=nvfp4
817
+ blk.51.attn_q.scale=f32
818
+ blk.51.attn_q.weight=nvfp4
819
+ blk.51.attn_q_norm.weight=f32
820
+ blk.51.attn_v.scale=f32
821
+ blk.51.attn_v.weight=nvfp4
822
+ blk.51.ffn_down.scale=f32
823
+ blk.51.ffn_down.weight=nvfp4
824
+ blk.51.ffn_gate.scale=f32
825
+ blk.51.ffn_gate.weight=nvfp4
826
+ blk.51.ffn_up.scale=f32
827
+ blk.51.ffn_up.weight=nvfp4
828
+ blk.51.post_attention_norm.weight=f32
829
+ blk.52.attn_gate.weight=nvfp4
830
+ blk.52.attn_norm.weight=f32
831
+ blk.52.attn_qkv.weight=nvfp4
832
+ blk.52.ffn_down.scale=f32
833
+ blk.52.ffn_down.weight=nvfp4
834
+ blk.52.ffn_gate.scale=f32
835
+ blk.52.ffn_gate.weight=nvfp4
836
+ blk.52.ffn_up.scale=f32
837
+ blk.52.ffn_up.weight=nvfp4
838
+ blk.52.post_attention_norm.weight=f32
839
+ blk.52.ssm_a=f32
840
+ blk.52.ssm_alpha.weight=nvfp4
841
+ blk.52.ssm_beta.weight=nvfp4
842
+ blk.52.ssm_conv1d.weight=f32
843
+ blk.52.ssm_dt.bias=f32
844
+ blk.52.ssm_norm.weight=f32
845
+ blk.52.ssm_out.weight=nvfp4
846
+ blk.53.attn_gate.weight=nvfp4
847
+ blk.53.attn_norm.weight=f32
848
+ blk.53.attn_qkv.weight=nvfp4
849
+ blk.53.ffn_down.scale=f32
850
+ blk.53.ffn_down.weight=nvfp4
851
+ blk.53.ffn_gate.scale=f32
852
+ blk.53.ffn_gate.weight=nvfp4
853
+ blk.53.ffn_up.scale=f32
854
+ blk.53.ffn_up.weight=nvfp4
855
+ blk.53.post_attention_norm.weight=f32
856
+ blk.53.ssm_a=f32
857
+ blk.53.ssm_alpha.weight=nvfp4
858
+ blk.53.ssm_beta.weight=nvfp4
859
+ blk.53.ssm_conv1d.weight=f32
860
+ blk.53.ssm_dt.bias=f32
861
+ blk.53.ssm_norm.weight=f32
862
+ blk.53.ssm_out.weight=nvfp4
863
+ blk.54.attn_gate.weight=nvfp4
864
+ blk.54.attn_norm.weight=f32
865
+ blk.54.attn_qkv.weight=nvfp4
866
+ blk.54.ffn_down.scale=f32
867
+ blk.54.ffn_down.weight=nvfp4
868
+ blk.54.ffn_gate.scale=f32
869
+ blk.54.ffn_gate.weight=nvfp4
870
+ blk.54.ffn_up.scale=f32
871
+ blk.54.ffn_up.weight=nvfp4
872
+ blk.54.post_attention_norm.weight=f32
873
+ blk.54.ssm_a=f32
874
+ blk.54.ssm_alpha.weight=nvfp4
875
+ blk.54.ssm_beta.weight=nvfp4
876
+ blk.54.ssm_conv1d.weight=f32
877
+ blk.54.ssm_dt.bias=f32
878
+ blk.54.ssm_norm.weight=f32
879
+ blk.54.ssm_out.weight=nvfp4
880
+ blk.55.attn_k.scale=f32
881
+ blk.55.attn_k.weight=nvfp4
882
+ blk.55.attn_k_norm.weight=f32
883
+ blk.55.attn_norm.weight=f32
884
+ blk.55.attn_output.scale=f32
885
+ blk.55.attn_output.weight=nvfp4
886
+ blk.55.attn_q.scale=f32
887
+ blk.55.attn_q.weight=nvfp4
888
+ blk.55.attn_q_norm.weight=f32
889
+ blk.55.attn_v.scale=f32
890
+ blk.55.attn_v.weight=nvfp4
891
+ blk.55.ffn_down.scale=f32
892
+ blk.55.ffn_down.weight=nvfp4
893
+ blk.55.ffn_gate.scale=f32
894
+ blk.55.ffn_gate.weight=nvfp4
895
+ blk.55.ffn_up.scale=f32
896
+ blk.55.ffn_up.weight=nvfp4
897
+ blk.55.post_attention_norm.weight=f32
898
+ blk.56.attn_gate.weight=nvfp4
899
+ blk.56.attn_norm.weight=f32
900
+ blk.56.attn_qkv.weight=nvfp4
901
+ blk.56.ffn_down.scale=f32
902
+ blk.56.ffn_down.weight=nvfp4
903
+ blk.56.ffn_gate.scale=f32
904
+ blk.56.ffn_gate.weight=nvfp4
905
+ blk.56.ffn_up.scale=f32
906
+ blk.56.ffn_up.weight=nvfp4
907
+ blk.56.post_attention_norm.weight=f32
908
+ blk.56.ssm_a=f32
909
+ blk.56.ssm_alpha.weight=nvfp4
910
+ blk.56.ssm_beta.weight=nvfp4
911
+ blk.56.ssm_conv1d.weight=f32
912
+ blk.56.ssm_dt.bias=f32
913
+ blk.56.ssm_norm.weight=f32
914
+ blk.56.ssm_out.weight=nvfp4
915
+ blk.57.attn_gate.weight=nvfp4
916
+ blk.57.attn_norm.weight=f32
917
+ blk.57.attn_qkv.weight=nvfp4
918
+ blk.57.ffn_down.scale=f32
919
+ blk.57.ffn_down.weight=nvfp4
920
+ blk.57.ffn_gate.scale=f32
921
+ blk.57.ffn_gate.weight=nvfp4
922
+ blk.57.ffn_up.scale=f32
923
+ blk.57.ffn_up.weight=nvfp4
924
+ blk.57.post_attention_norm.weight=f32
925
+ blk.57.ssm_a=f32
926
+ blk.57.ssm_alpha.weight=nvfp4
927
+ blk.57.ssm_beta.weight=nvfp4
928
+ blk.57.ssm_conv1d.weight=f32
929
+ blk.57.ssm_dt.bias=f32
930
+ blk.57.ssm_norm.weight=f32
931
+ blk.57.ssm_out.weight=nvfp4
932
+ blk.58.attn_gate.weight=nvfp4
933
+ blk.58.attn_norm.weight=f32
934
+ blk.58.attn_qkv.weight=nvfp4
935
+ blk.58.ffn_down.scale=f32
936
+ blk.58.ffn_down.weight=nvfp4
937
+ blk.58.ffn_gate.scale=f32
938
+ blk.58.ffn_gate.weight=nvfp4
939
+ blk.58.ffn_up.scale=f32
940
+ blk.58.ffn_up.weight=nvfp4
941
+ blk.58.post_attention_norm.weight=f32
942
+ blk.58.ssm_a=f32
943
+ blk.58.ssm_alpha.weight=nvfp4
944
+ blk.58.ssm_beta.weight=nvfp4
945
+ blk.58.ssm_conv1d.weight=f32
946
+ blk.58.ssm_dt.bias=f32
947
+ blk.58.ssm_norm.weight=f32
948
+ blk.58.ssm_out.weight=nvfp4
949
+ blk.59.attn_k.scale=f32
950
+ blk.59.attn_k.weight=nvfp4
951
+ blk.59.attn_k_norm.weight=f32
952
+ blk.59.attn_norm.weight=f32
953
+ blk.59.attn_output.scale=f32
954
+ blk.59.attn_output.weight=nvfp4
955
+ blk.59.attn_q.scale=f32
956
+ blk.59.attn_q.weight=nvfp4
957
+ blk.59.attn_q_norm.weight=f32
958
+ blk.59.attn_v.scale=f32
959
+ blk.59.attn_v.weight=nvfp4
960
+ blk.59.ffn_down.scale=f32
961
+ blk.59.ffn_down.weight=nvfp4
962
+ blk.59.ffn_gate.scale=f32
963
+ blk.59.ffn_gate.weight=nvfp4
964
+ blk.59.ffn_up.scale=f32
965
+ blk.59.ffn_up.weight=nvfp4
966
+ blk.59.post_attention_norm.weight=f32
967
+ blk.6.attn_gate.weight=nvfp4
968
+ blk.6.attn_norm.weight=f32
969
+ blk.6.attn_qkv.weight=nvfp4
970
+ blk.6.ffn_down.scale=f32
971
+ blk.6.ffn_down.weight=nvfp4
972
+ blk.6.ffn_gate.scale=f32
973
+ blk.6.ffn_gate.weight=nvfp4
974
+ blk.6.ffn_up.scale=f32
975
+ blk.6.ffn_up.weight=nvfp4
976
+ blk.6.post_attention_norm.weight=f32
977
+ blk.6.ssm_a=f32
978
+ blk.6.ssm_alpha.weight=nvfp4
979
+ blk.6.ssm_beta.weight=nvfp4
980
+ blk.6.ssm_conv1d.weight=f32
981
+ blk.6.ssm_dt.bias=f32
982
+ blk.6.ssm_norm.weight=f32
983
+ blk.6.ssm_out.weight=nvfp4
984
+ blk.60.attn_gate.weight=nvfp4
985
+ blk.60.attn_norm.weight=f32
986
+ blk.60.attn_qkv.weight=nvfp4
987
+ blk.60.ffn_down.scale=f32
988
+ blk.60.ffn_down.weight=nvfp4
989
+ blk.60.ffn_gate.scale=f32
990
+ blk.60.ffn_gate.weight=nvfp4
991
+ blk.60.ffn_up.scale=f32
992
+ blk.60.ffn_up.weight=nvfp4
993
+ blk.60.post_attention_norm.weight=f32
994
+ blk.60.ssm_a=f32
995
+ blk.60.ssm_alpha.weight=nvfp4
996
+ blk.60.ssm_beta.weight=nvfp4
997
+ blk.60.ssm_conv1d.weight=f32
998
+ blk.60.ssm_dt.bias=f32
999
+ blk.60.ssm_norm.weight=f32
1000
+ blk.60.ssm_out.weight=nvfp4
1001
+ blk.61.attn_gate.weight=nvfp4
1002
+ blk.61.attn_norm.weight=f32
1003
+ blk.61.attn_qkv.weight=nvfp4
1004
+ blk.61.ffn_down.scale=f32
1005
+ blk.61.ffn_down.weight=nvfp4
1006
+ blk.61.ffn_gate.scale=f32
1007
+ blk.61.ffn_gate.weight=nvfp4
1008
+ blk.61.ffn_up.scale=f32
1009
+ blk.61.ffn_up.weight=nvfp4
1010
+ blk.61.post_attention_norm.weight=f32
1011
+ blk.61.ssm_a=f32
1012
+ blk.61.ssm_alpha.weight=nvfp4
1013
+ blk.61.ssm_beta.weight=nvfp4
1014
+ blk.61.ssm_conv1d.weight=f32
1015
+ blk.61.ssm_dt.bias=f32
1016
+ blk.61.ssm_norm.weight=f32
1017
+ blk.61.ssm_out.weight=nvfp4
1018
+ blk.62.attn_gate.weight=nvfp4
1019
+ blk.62.attn_norm.weight=f32
1020
+ blk.62.attn_qkv.weight=nvfp4
1021
+ blk.62.ffn_down.scale=f32
1022
+ blk.62.ffn_down.weight=nvfp4
1023
+ blk.62.ffn_gate.scale=f32
1024
+ blk.62.ffn_gate.weight=nvfp4
1025
+ blk.62.ffn_up.scale=f32
1026
+ blk.62.ffn_up.weight=nvfp4
1027
+ blk.62.post_attention_norm.weight=f32
1028
+ blk.62.ssm_a=f32
1029
+ blk.62.ssm_alpha.weight=nvfp4
1030
+ blk.62.ssm_beta.weight=nvfp4
1031
+ blk.62.ssm_conv1d.weight=f32
1032
+ blk.62.ssm_dt.bias=f32
1033
+ blk.62.ssm_norm.weight=f32
1034
+ blk.62.ssm_out.weight=nvfp4
1035
+ blk.63.attn_k.scale=f32
1036
+ blk.63.attn_k.weight=nvfp4
1037
+ blk.63.attn_k_norm.weight=f32
1038
+ blk.63.attn_norm.weight=f32
1039
+ blk.63.attn_output.scale=f32
1040
+ blk.63.attn_output.weight=nvfp4
1041
+ blk.63.attn_q.scale=f32
1042
+ blk.63.attn_q.weight=nvfp4
1043
+ blk.63.attn_q_norm.weight=f32
1044
+ blk.63.attn_v.scale=f32
1045
+ blk.63.attn_v.weight=nvfp4
1046
+ blk.63.ffn_down.scale=f32
1047
+ blk.63.ffn_down.weight=nvfp4
1048
+ blk.63.ffn_gate.scale=f32
1049
+ blk.63.ffn_gate.weight=nvfp4
1050
+ blk.63.ffn_up.scale=f32
1051
+ blk.63.ffn_up.weight=nvfp4
1052
+ blk.63.post_attention_norm.weight=f32
1053
+ blk.7.attn_k.scale=f32
1054
+ blk.7.attn_k.weight=nvfp4
1055
+ blk.7.attn_k_norm.weight=f32
1056
+ blk.7.attn_norm.weight=f32
1057
+ blk.7.attn_output.scale=f32
1058
+ blk.7.attn_output.weight=nvfp4
1059
+ blk.7.attn_q.scale=f32
1060
+ blk.7.attn_q.weight=nvfp4
1061
+ blk.7.attn_q_norm.weight=f32
1062
+ blk.7.attn_v.scale=f32
1063
+ blk.7.attn_v.weight=nvfp4
1064
+ blk.7.ffn_down.scale=f32
1065
+ blk.7.ffn_down.weight=nvfp4
1066
+ blk.7.ffn_gate.scale=f32
1067
+ blk.7.ffn_gate.weight=nvfp4
1068
+ blk.7.ffn_up.scale=f32
1069
+ blk.7.ffn_up.weight=nvfp4
1070
+ blk.7.post_attention_norm.weight=f32
1071
+ blk.8.attn_gate.weight=nvfp4
1072
+ blk.8.attn_norm.weight=f32
1073
+ blk.8.attn_qkv.weight=nvfp4
1074
+ blk.8.ffn_down.scale=f32
1075
+ blk.8.ffn_down.weight=nvfp4
1076
+ blk.8.ffn_gate.scale=f32
1077
+ blk.8.ffn_gate.weight=nvfp4
1078
+ blk.8.ffn_up.scale=f32
1079
+ blk.8.ffn_up.weight=nvfp4
1080
+ blk.8.post_attention_norm.weight=f32
1081
+ blk.8.ssm_a=f32
1082
+ blk.8.ssm_alpha.weight=nvfp4
1083
+ blk.8.ssm_beta.weight=nvfp4
1084
+ blk.8.ssm_conv1d.weight=f32
1085
+ blk.8.ssm_dt.bias=f32
1086
+ blk.8.ssm_norm.weight=f32
1087
+ blk.8.ssm_out.weight=nvfp4
1088
+ blk.9.attn_gate.weight=nvfp4
1089
+ blk.9.attn_norm.weight=f32
1090
+ blk.9.attn_qkv.weight=nvfp4
1091
+ blk.9.ffn_down.scale=f32
1092
+ blk.9.ffn_down.weight=nvfp4
1093
+ blk.9.ffn_gate.scale=f32
1094
+ blk.9.ffn_gate.weight=nvfp4
1095
+ blk.9.ffn_up.scale=f32
1096
+ blk.9.ffn_up.weight=nvfp4
1097
+ blk.9.post_attention_norm.weight=f32
1098
+ blk.9.ssm_a=f32
1099
+ blk.9.ssm_alpha.weight=nvfp4
1100
+ blk.9.ssm_beta.weight=nvfp4
1101
+ blk.9.ssm_conv1d.weight=f32
1102
+ blk.9.ssm_dt.bias=f32
1103
+ blk.9.ssm_norm.weight=f32
1104
+ blk.9.ssm_out.weight=nvfp4
1105
+ output_norm.weight=f32
1106
+ output.weight=q3_k
1107
+ token_embd.weight=q2_k