Hugging Face's logo

stepfun-ai
/
Step-3.7-Flash-NVFP4

Image-Text-to-Text
Transformers
Safetensors
English
step3p7
vision-language
multimodal
Mixture of Experts
conversational
custom_code
8-bit precision
modelopt
Model card Files
xet
 Community
7
huangyu-nv commited on
Commit
4275532
·
verified ·
1 Parent(s): 32fee87

Add MTP draft layers to NVFP4 checkpoint

Browse files

Adds BF16 MTP draft layers for vLLM speculative decoding and updates safetensors index plus quantization metadata. Original NVFP4 model shards are unchanged.

Files changed (4) hide show
  1. config.json +24 -5
  2. hf_quant_config.json +150 -144
  3. model-mtp-bf16.safetensors +3 -0
  4. model.safetensors.index.json +53 -2
config.json CHANGED
@@ -86,7 +86,10 @@
86
  "sliding_attention",
87
  "sliding_attention",
88
  "sliding_attention",
89
- "full_attention"
 
 
 
90
  ],
91
  "max_position_embeddings": 262144,
92
  "max_seq_len": 262144,
@@ -151,7 +154,10 @@
151
  1.0,
152
  1.0,
153
  1.0,
154
- 0.5
 
 
 
155
  ],
156
  "rms_norm_eps": 1e-05,
157
  "rope_scaling": {
@@ -259,7 +265,10 @@
259
  0.0,
260
  0.0,
261
  7,
262
- 7
 
 
 
263
  ],
264
  "swiglu_limits_shared": [
265
  0.0,
@@ -306,7 +315,10 @@
306
  0.0,
307
  0.0,
308
  16,
309
- 16
 
 
 
310
  ],
311
  "torch_dtype": "bfloat16",
312
  "use_cache": false,
@@ -314,6 +326,7 @@
314
  "use_mfa": false,
315
  "use_moe": true,
316
  "use_moe_router_bias": true,
 
317
  "use_rope_layers": [],
318
  "vocab_size": 128896,
319
  "yarn_only_types": [
@@ -506,7 +519,13 @@
506
  "model.language_model.layers.9.self_attn*",
507
  "model.language_model.layers.9.share_expert*",
508
  "model.vision_model*",
509
- "model.vit_large_projector"
 
 
 
 
 
 
510
  ],
511
  "quant_algo": "NVFP4",
512
  "kv_cache_scheme": {
 
86
  "sliding_attention",
87
  "sliding_attention",
88
  "sliding_attention",
89
+ "full_attention",
90
+ "sliding_attention",
91
+ "sliding_attention",
92
+ "sliding_attention"
93
  ],
94
  "max_position_embeddings": 262144,
95
  "max_seq_len": 262144,
 
154
  1.0,
155
  1.0,
156
  1.0,
157
+ 0.5,
158
+ 1.0,
159
+ 1.0,
160
+ 1.0
161
  ],
162
  "rms_norm_eps": 1e-05,
163
  "rope_scaling": {
 
265
  0.0,
266
  0.0,
267
  7,
268
+ 7,
269
+ 0.0,
270
+ 0.0,
271
+ 0.0
272
  ],
273
  "swiglu_limits_shared": [
274
  0.0,
 
315
  0.0,
316
  0.0,
317
  16,
318
+ 16,
319
+ 0.0,
320
+ 0.0,
321
+ 0.0
322
  ],
323
  "torch_dtype": "bfloat16",
324
  "use_cache": false,
 
326
  "use_mfa": false,
327
  "use_moe": true,
328
  "use_moe_router_bias": true,
329
+ "use_qk_norm": false,
330
  "use_rope_layers": [],
331
  "vocab_size": 128896,
332
  "yarn_only_types": [
 
519
  "model.language_model.layers.9.self_attn*",
520
  "model.language_model.layers.9.share_expert*",
521
  "model.vision_model*",
522
+ "model.vit_large_projector",
523
+ "model.language_model.layers.45.*",
524
+ "model.language_model.layers.46.*",
525
+ "model.language_model.layers.47.*",
526
+ "model.layers.45.*",
527
+ "model.layers.46.*",
528
+ "model.layers.47.*"
529
  ],
530
  "quant_algo": "NVFP4",
531
  "kv_cache_scheme": {
hf_quant_config.json CHANGED
@@ -1,145 +1,151 @@
1
  {
2
- "producer": {
3
- "name": "modelopt",
4
- "version": "0.45.0.dev37+g3ad4f4f09.d20260524"
5
- },
6
- "quantization": {
7
- "quant_algo": "NVFP4",
8
- "kv_cache_quant_algo": "FP8",
9
- "group_size": 16,
10
- "exclude_modules": [
11
- "lm_head",
12
- "model.language_model.layers.0*",
13
- "model.language_model.layers.1.*",
14
- "model.language_model.layers.10.moe.gate",
15
- "model.language_model.layers.10.self_attn*",
16
- "model.language_model.layers.10.share_expert*",
17
- "model.language_model.layers.11.moe.gate",
18
- "model.language_model.layers.11.self_attn*",
19
- "model.language_model.layers.11.share_expert*",
20
- "model.language_model.layers.12.moe.gate",
21
- "model.language_model.layers.12.self_attn*",
22
- "model.language_model.layers.12.share_expert*",
23
- "model.language_model.layers.13.moe.gate",
24
- "model.language_model.layers.13.self_attn*",
25
- "model.language_model.layers.13.share_expert*",
26
- "model.language_model.layers.14.moe.gate",
27
- "model.language_model.layers.14.self_attn*",
28
- "model.language_model.layers.14.share_expert*",
29
- "model.language_model.layers.15.moe.gate",
30
- "model.language_model.layers.15.self_attn*",
31
- "model.language_model.layers.15.share_expert*",
32
- "model.language_model.layers.16.moe.gate",
33
- "model.language_model.layers.16.self_attn*",
34
- "model.language_model.layers.16.share_expert*",
35
- "model.language_model.layers.17.moe.gate",
36
- "model.language_model.layers.17.self_attn*",
37
- "model.language_model.layers.17.share_expert*",
38
- "model.language_model.layers.18.moe.gate",
39
- "model.language_model.layers.18.self_attn*",
40
- "model.language_model.layers.18.share_expert*",
41
- "model.language_model.layers.19.moe.gate",
42
- "model.language_model.layers.19.self_attn*",
43
- "model.language_model.layers.19.share_expert*",
44
- "model.language_model.layers.2.*",
45
- "model.language_model.layers.20.moe.gate",
46
- "model.language_model.layers.20.self_attn*",
47
- "model.language_model.layers.20.share_expert*",
48
- "model.language_model.layers.21.moe.gate",
49
- "model.language_model.layers.21.self_attn*",
50
- "model.language_model.layers.21.share_expert*",
51
- "model.language_model.layers.22.moe.gate",
52
- "model.language_model.layers.22.self_attn*",
53
- "model.language_model.layers.22.share_expert*",
54
- "model.language_model.layers.23.moe.gate",
55
- "model.language_model.layers.23.self_attn*",
56
- "model.language_model.layers.23.share_expert*",
57
- "model.language_model.layers.24.moe.gate",
58
- "model.language_model.layers.24.self_attn*",
59
- "model.language_model.layers.24.share_expert*",
60
- "model.language_model.layers.25.moe.gate",
61
- "model.language_model.layers.25.self_attn*",
62
- "model.language_model.layers.25.share_expert*",
63
- "model.language_model.layers.26.moe.gate",
64
- "model.language_model.layers.26.self_attn*",
65
- "model.language_model.layers.26.share_expert*",
66
- "model.language_model.layers.27.moe.gate",
67
- "model.language_model.layers.27.self_attn*",
68
- "model.language_model.layers.27.share_expert*",
69
- "model.language_model.layers.28.moe.gate",
70
- "model.language_model.layers.28.self_attn*",
71
- "model.language_model.layers.28.share_expert*",
72
- "model.language_model.layers.29.moe.gate",
73
- "model.language_model.layers.29.self_attn*",
74
- "model.language_model.layers.29.share_expert*",
75
- "model.language_model.layers.3.moe.gate",
76
- "model.language_model.layers.3.self_attn*",
77
- "model.language_model.layers.3.share_expert*",
78
- "model.language_model.layers.30.moe.gate",
79
- "model.language_model.layers.30.self_attn*",
80
- "model.language_model.layers.30.share_expert*",
81
- "model.language_model.layers.31.moe.gate",
82
- "model.language_model.layers.31.self_attn*",
83
- "model.language_model.layers.31.share_expert*",
84
- "model.language_model.layers.32.moe.gate",
85
- "model.language_model.layers.32.self_attn*",
86
- "model.language_model.layers.32.share_expert*",
87
- "model.language_model.layers.33.moe.gate",
88
- "model.language_model.layers.33.self_attn*",
89
- "model.language_model.layers.33.share_expert*",
90
- "model.language_model.layers.34.moe.gate",
91
- "model.language_model.layers.34.self_attn*",
92
- "model.language_model.layers.34.share_expert*",
93
- "model.language_model.layers.35.moe.gate",
94
- "model.language_model.layers.35.self_attn*",
95
- "model.language_model.layers.35.share_expert*",
96
- "model.language_model.layers.36.moe.gate",
97
- "model.language_model.layers.36.self_attn*",
98
- "model.language_model.layers.36.share_expert*",
99
- "model.language_model.layers.37.moe.gate",
100
- "model.language_model.layers.37.self_attn*",
101
- "model.language_model.layers.37.share_expert*",
102
- "model.language_model.layers.38.moe.gate",
103
- "model.language_model.layers.38.self_attn*",
104
- "model.language_model.layers.38.share_expert*",
105
- "model.language_model.layers.39.moe.gate",
106
- "model.language_model.layers.39.self_attn*",
107
- "model.language_model.layers.39.share_expert*",
108
- "model.language_model.layers.4.moe.gate",
109
- "model.language_model.layers.4.self_attn*",
110
- "model.language_model.layers.4.share_expert*",
111
- "model.language_model.layers.40.moe.gate",
112
- "model.language_model.layers.40.self_attn*",
113
- "model.language_model.layers.40.share_expert*",
114
- "model.language_model.layers.41.moe.gate",
115
- "model.language_model.layers.41.self_attn*",
116
- "model.language_model.layers.41.share_expert*",
117
- "model.language_model.layers.42.moe.gate",
118
- "model.language_model.layers.42.self_attn*",
119
- "model.language_model.layers.42.share_expert*",
120
- "model.language_model.layers.43.moe.gate",
121
- "model.language_model.layers.43.self_attn*",
122
- "model.language_model.layers.43.share_expert*",
123
- "model.language_model.layers.44.moe.gate",
124
- "model.language_model.layers.44.self_attn*",
125
- "model.language_model.layers.44.share_expert*",
126
- "model.language_model.layers.5.moe.gate",
127
- "model.language_model.layers.5.self_attn*",
128
- "model.language_model.layers.5.share_expert*",
129
- "model.language_model.layers.6.moe.gate",
130
- "model.language_model.layers.6.self_attn*",
131
- "model.language_model.layers.6.share_expert*",
132
- "model.language_model.layers.7.moe.gate",
133
- "model.language_model.layers.7.self_attn*",
134
- "model.language_model.layers.7.share_expert*",
135
- "model.language_model.layers.8.moe.gate",
136
- "model.language_model.layers.8.self_attn*",
137
- "model.language_model.layers.8.share_expert*",
138
- "model.language_model.layers.9.moe.gate",
139
- "model.language_model.layers.9.self_attn*",
140
- "model.language_model.layers.9.share_expert*",
141
- "model.vision_model*",
142
- "model.vit_large_projector"
143
- ]
144
- }
145
- }
 
 
 
 
 
 
 
1
  {
2
+ "producer": {
3
+ "name": "modelopt",
4
+ "version": "0.45.0.dev37+g3ad4f4f09.d20260524"
5
+ },
6
+ "quantization": {
7
+ "quant_algo": "NVFP4",
8
+ "kv_cache_quant_algo": "FP8",
9
+ "group_size": 16,
10
+ "exclude_modules": [
11
+ "lm_head",
12
+ "model.language_model.layers.0*",
13
+ "model.language_model.layers.1.*",
14
+ "model.language_model.layers.10.moe.gate",
15
+ "model.language_model.layers.10.self_attn*",
16
+ "model.language_model.layers.10.share_expert*",
17
+ "model.language_model.layers.11.moe.gate",
18
+ "model.language_model.layers.11.self_attn*",
19
+ "model.language_model.layers.11.share_expert*",
20
+ "model.language_model.layers.12.moe.gate",
21
+ "model.language_model.layers.12.self_attn*",
22
+ "model.language_model.layers.12.share_expert*",
23
+ "model.language_model.layers.13.moe.gate",
24
+ "model.language_model.layers.13.self_attn*",
25
+ "model.language_model.layers.13.share_expert*",
26
+ "model.language_model.layers.14.moe.gate",
27
+ "model.language_model.layers.14.self_attn*",
28
+ "model.language_model.layers.14.share_expert*",
29
+ "model.language_model.layers.15.moe.gate",
30
+ "model.language_model.layers.15.self_attn*",
31
+ "model.language_model.layers.15.share_expert*",
32
+ "model.language_model.layers.16.moe.gate",
33
+ "model.language_model.layers.16.self_attn*",
34
+ "model.language_model.layers.16.share_expert*",
35
+ "model.language_model.layers.17.moe.gate",
36
+ "model.language_model.layers.17.self_attn*",
37
+ "model.language_model.layers.17.share_expert*",
38
+ "model.language_model.layers.18.moe.gate",
39
+ "model.language_model.layers.18.self_attn*",
40
+ "model.language_model.layers.18.share_expert*",
41
+ "model.language_model.layers.19.moe.gate",
42
+ "model.language_model.layers.19.self_attn*",
43
+ "model.language_model.layers.19.share_expert*",
44
+ "model.language_model.layers.2.*",
45
+ "model.language_model.layers.20.moe.gate",
46
+ "model.language_model.layers.20.self_attn*",
47
+ "model.language_model.layers.20.share_expert*",
48
+ "model.language_model.layers.21.moe.gate",
49
+ "model.language_model.layers.21.self_attn*",
50
+ "model.language_model.layers.21.share_expert*",
51
+ "model.language_model.layers.22.moe.gate",
52
+ "model.language_model.layers.22.self_attn*",
53
+ "model.language_model.layers.22.share_expert*",
54
+ "model.language_model.layers.23.moe.gate",
55
+ "model.language_model.layers.23.self_attn*",
56
+ "model.language_model.layers.23.share_expert*",
57
+ "model.language_model.layers.24.moe.gate",
58
+ "model.language_model.layers.24.self_attn*",
59
+ "model.language_model.layers.24.share_expert*",
60
+ "model.language_model.layers.25.moe.gate",
61
+ "model.language_model.layers.25.self_attn*",
62
+ "model.language_model.layers.25.share_expert*",
63
+ "model.language_model.layers.26.moe.gate",
64
+ "model.language_model.layers.26.self_attn*",
65
+ "model.language_model.layers.26.share_expert*",
66
+ "model.language_model.layers.27.moe.gate",
67
+ "model.language_model.layers.27.self_attn*",
68
+ "model.language_model.layers.27.share_expert*",
69
+ "model.language_model.layers.28.moe.gate",
70
+ "model.language_model.layers.28.self_attn*",
71
+ "model.language_model.layers.28.share_expert*",
72
+ "model.language_model.layers.29.moe.gate",
73
+ "model.language_model.layers.29.self_attn*",
74
+ "model.language_model.layers.29.share_expert*",
75
+ "model.language_model.layers.3.moe.gate",
76
+ "model.language_model.layers.3.self_attn*",
77
+ "model.language_model.layers.3.share_expert*",
78
+ "model.language_model.layers.30.moe.gate",
79
+ "model.language_model.layers.30.self_attn*",
80
+ "model.language_model.layers.30.share_expert*",
81
+ "model.language_model.layers.31.moe.gate",
82
+ "model.language_model.layers.31.self_attn*",
83
+ "model.language_model.layers.31.share_expert*",
84
+ "model.language_model.layers.32.moe.gate",
85
+ "model.language_model.layers.32.self_attn*",
86
+ "model.language_model.layers.32.share_expert*",
87
+ "model.language_model.layers.33.moe.gate",
88
+ "model.language_model.layers.33.self_attn*",
89
+ "model.language_model.layers.33.share_expert*",
90
+ "model.language_model.layers.34.moe.gate",
91
+ "model.language_model.layers.34.self_attn*",
92
+ "model.language_model.layers.34.share_expert*",
93
+ "model.language_model.layers.35.moe.gate",
94
+ "model.language_model.layers.35.self_attn*",
95
+ "model.language_model.layers.35.share_expert*",
96
+ "model.language_model.layers.36.moe.gate",
97
+ "model.language_model.layers.36.self_attn*",
98
+ "model.language_model.layers.36.share_expert*",
99
+ "model.language_model.layers.37.moe.gate",
100
+ "model.language_model.layers.37.self_attn*",
101
+ "model.language_model.layers.37.share_expert*",
102
+ "model.language_model.layers.38.moe.gate",
103
+ "model.language_model.layers.38.self_attn*",
104
+ "model.language_model.layers.38.share_expert*",
105
+ "model.language_model.layers.39.moe.gate",
106
+ "model.language_model.layers.39.self_attn*",
107
+ "model.language_model.layers.39.share_expert*",
108
+ "model.language_model.layers.4.moe.gate",
109
+ "model.language_model.layers.4.self_attn*",
110
+ "model.language_model.layers.4.share_expert*",
111
+ "model.language_model.layers.40.moe.gate",
112
+ "model.language_model.layers.40.self_attn*",
113
+ "model.language_model.layers.40.share_expert*",
114
+ "model.language_model.layers.41.moe.gate",
115
+ "model.language_model.layers.41.self_attn*",
116
+ "model.language_model.layers.41.share_expert*",
117
+ "model.language_model.layers.42.moe.gate",
118
+ "model.language_model.layers.42.self_attn*",
119
+ "model.language_model.layers.42.share_expert*",
120
+ "model.language_model.layers.43.moe.gate",
121
+ "model.language_model.layers.43.self_attn*",
122
+ "model.language_model.layers.43.share_expert*",
123
+ "model.language_model.layers.44.moe.gate",
124
+ "model.language_model.layers.44.self_attn*",
125
+ "model.language_model.layers.44.share_expert*",
126
+ "model.language_model.layers.5.moe.gate",
127
+ "model.language_model.layers.5.self_attn*",
128
+ "model.language_model.layers.5.share_expert*",
129
+ "model.language_model.layers.6.moe.gate",
130
+ "model.language_model.layers.6.self_attn*",
131
+ "model.language_model.layers.6.share_expert*",
132
+ "model.language_model.layers.7.moe.gate",
133
+ "model.language_model.layers.7.self_attn*",
134
+ "model.language_model.layers.7.share_expert*",
135
+ "model.language_model.layers.8.moe.gate",
136
+ "model.language_model.layers.8.self_attn*",
137
+ "model.language_model.layers.8.share_expert*",
138
+ "model.language_model.layers.9.moe.gate",
139
+ "model.language_model.layers.9.self_attn*",
140
+ "model.language_model.layers.9.share_expert*",
141
+ "model.vision_model*",
142
+ "model.vit_large_projector",
143
+ "model.language_model.layers.45.*",
144
+ "model.language_model.layers.46.*",
145
+ "model.language_model.layers.47.*",
146
+ "model.layers.45.*",
147
+ "model.layers.46.*",
148
+ "model.layers.47.*"
149
+ ]
150
+ }
151
+ }
model-mtp-bf16.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7af1f8ae0a316de2197c76c0764c8512ff0fb9263a55bb4dfb3f35f25214b7e7
3
+ size 4856347904
model.safetensors.index.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
  "metadata": {
3
  "total_parameters": 103810330432,
4
- "total_size": 124385012840
5
  },
6
  "weight_map": {
7
  "lm_head.weight": "model-00013-of-00013.safetensors",
@@ -1891,6 +1891,57 @@
1891
  "model.vision_model.vit_downsampler1.weight": "model-00001-of-00013.safetensors",
1892
  "model.vision_model.vit_downsampler2.bias": "model-00001-of-00013.safetensors",
1893
  "model.vision_model.vit_downsampler2.weight": "model-00001-of-00013.safetensors",
1894
- "model.vit_large_projector.weight": "model-00013-of-00013.safetensors"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1895
  }
1896
  }
 
1
  {
2
  "metadata": {
3
  "total_parameters": 103810330432,
4
+ "total_size": 129241604232
5
  },
6
  "weight_map": {
7
  "lm_head.weight": "model-00013-of-00013.safetensors",
 
1891
  "model.vision_model.vit_downsampler1.weight": "model-00001-of-00013.safetensors",
1892
  "model.vision_model.vit_downsampler2.bias": "model-00001-of-00013.safetensors",
1893
  "model.vision_model.vit_downsampler2.weight": "model-00001-of-00013.safetensors",
1894
+ "model.vit_large_projector.weight": "model-00013-of-00013.safetensors",
1895
+ "model.layers.45.eh_proj.weight": "model-mtp-bf16.safetensors",
1896
+ "model.layers.45.enorm.weight": "model-mtp-bf16.safetensors",
1897
+ "model.layers.45.hnorm.weight": "model-mtp-bf16.safetensors",
1898
+ "model.layers.45.input_layernorm.weight": "model-mtp-bf16.safetensors",
1899
+ "model.layers.45.mlp.down_proj.weight": "model-mtp-bf16.safetensors",
1900
+ "model.layers.45.mlp.gate_proj.weight": "model-mtp-bf16.safetensors",
1901
+ "model.layers.45.mlp.up_proj.weight": "model-mtp-bf16.safetensors",
1902
+ "model.layers.45.post_attention_layernorm.weight": "model-mtp-bf16.safetensors",
1903
+ "model.layers.45.self_attn.g_proj.weight": "model-mtp-bf16.safetensors",
1904
+ "model.layers.45.self_attn.k_norm.weight": "model-mtp-bf16.safetensors",
1905
+ "model.layers.45.self_attn.k_proj.weight": "model-mtp-bf16.safetensors",
1906
+ "model.layers.45.self_attn.o_proj.weight": "model-mtp-bf16.safetensors",
1907
+ "model.layers.45.self_attn.q_norm.weight": "model-mtp-bf16.safetensors",
1908
+ "model.layers.45.self_attn.q_proj.weight": "model-mtp-bf16.safetensors",
1909
+ "model.layers.45.self_attn.v_proj.weight": "model-mtp-bf16.safetensors",
1910
+ "model.layers.45.transformer.shared_head.norm.weight": "model-mtp-bf16.safetensors",
1911
+ "model.layers.45.transformer.shared_head.output.weight": "model-mtp-bf16.safetensors",
1912
+ "model.layers.46.eh_proj.weight": "model-mtp-bf16.safetensors",
1913
+ "model.layers.46.enorm.weight": "model-mtp-bf16.safetensors",
1914
+ "model.layers.46.hnorm.weight": "model-mtp-bf16.safetensors",
1915
+ "model.layers.46.input_layernorm.weight": "model-mtp-bf16.safetensors",
1916
+ "model.layers.46.mlp.down_proj.weight": "model-mtp-bf16.safetensors",
1917
+ "model.layers.46.mlp.gate_proj.weight": "model-mtp-bf16.safetensors",
1918
+ "model.layers.46.mlp.up_proj.weight": "model-mtp-bf16.safetensors",
1919
+ "model.layers.46.post_attention_layernorm.weight": "model-mtp-bf16.safetensors",
1920
+ "model.layers.46.self_attn.g_proj.weight": "model-mtp-bf16.safetensors",
1921
+ "model.layers.46.self_attn.k_norm.weight": "model-mtp-bf16.safetensors",
1922
+ "model.layers.46.self_attn.k_proj.weight": "model-mtp-bf16.safetensors",
1923
+ "model.layers.46.self_attn.o_proj.weight": "model-mtp-bf16.safetensors",
1924
+ "model.layers.46.self_attn.q_norm.weight": "model-mtp-bf16.safetensors",
1925
+ "model.layers.46.self_attn.q_proj.weight": "model-mtp-bf16.safetensors",
1926
+ "model.layers.46.self_attn.v_proj.weight": "model-mtp-bf16.safetensors",
1927
+ "model.layers.46.transformer.shared_head.norm.weight": "model-mtp-bf16.safetensors",
1928
+ "model.layers.46.transformer.shared_head.output.weight": "model-mtp-bf16.safetensors",
1929
+ "model.layers.47.eh_proj.weight": "model-mtp-bf16.safetensors",
1930
+ "model.layers.47.enorm.weight": "model-mtp-bf16.safetensors",
1931
+ "model.layers.47.hnorm.weight": "model-mtp-bf16.safetensors",
1932
+ "model.layers.47.input_layernorm.weight": "model-mtp-bf16.safetensors",
1933
+ "model.layers.47.mlp.down_proj.weight": "model-mtp-bf16.safetensors",
1934
+ "model.layers.47.mlp.gate_proj.weight": "model-mtp-bf16.safetensors",
1935
+ "model.layers.47.mlp.up_proj.weight": "model-mtp-bf16.safetensors",
1936
+ "model.layers.47.post_attention_layernorm.weight": "model-mtp-bf16.safetensors",
1937
+ "model.layers.47.self_attn.g_proj.weight": "model-mtp-bf16.safetensors",
1938
+ "model.layers.47.self_attn.k_norm.weight": "model-mtp-bf16.safetensors",
1939
+ "model.layers.47.self_attn.k_proj.weight": "model-mtp-bf16.safetensors",
1940
+ "model.layers.47.self_attn.o_proj.weight": "model-mtp-bf16.safetensors",
1941
+ "model.layers.47.self_attn.q_norm.weight": "model-mtp-bf16.safetensors",
1942
+ "model.layers.47.self_attn.q_proj.weight": "model-mtp-bf16.safetensors",
1943
+ "model.layers.47.self_attn.v_proj.weight": "model-mtp-bf16.safetensors",
1944
+ "model.layers.47.transformer.shared_head.norm.weight": "model-mtp-bf16.safetensors",
1945
+ "model.layers.47.transformer.shared_head.output.weight": "model-mtp-bf16.safetensors"
1946
  }
1947
  }