Hugging Face's logo

RedHatAI
/
DeepSeek-V4-Flash-NVFP4-FP8

Text Generation
Transformers
Safetensors
deepseek_v4
compressed-tensors
nvfp4
vllm
8-bit precision
Model card Files
xet
 Community
2
kylesayrs commited on
Commit
bf7ad50
·
verified ·
1 Parent(s): 5be6632

Upload folder using huggingface_hub

Browse files
Files changed (11) hide show
  1. .gitattributes +1 -0
  2. config.json +238 -0
  3. generation_config.json +9 -0
  4. model-00001-of-00004.safetensors +3 -0
  5. model-00002-of-00004.safetensors +3 -0
  6. model-00003-of-00004.safetensors +3 -0
  7. model-00004-of-00004.safetensors +3 -0
  8. model.safetensors.index.json +3 -0
  9. recipe.yaml +67 -0
  10. tokenizer.json +0 -0
  11. tokenizer_config.json +14 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ model.safetensors.index.json filter=lfs diff=lfs merge=lfs -text
config.json ADDED
@@ -0,0 +1,238 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "DeepseekV4ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 0,
8
+ "compress_ratios": [
9
+ 0,
10
+ 0,
11
+ 4,
12
+ 128,
13
+ 4,
14
+ 128,
15
+ 4,
16
+ 128,
17
+ 4,
18
+ 128,
19
+ 4,
20
+ 128,
21
+ 4,
22
+ 128,
23
+ 4,
24
+ 128,
25
+ 4,
26
+ 128,
27
+ 4,
28
+ 128,
29
+ 4,
30
+ 128,
31
+ 4,
32
+ 128,
33
+ 4,
34
+ 128,
35
+ 4,
36
+ 128,
37
+ 4,
38
+ 128,
39
+ 4,
40
+ 128,
41
+ 4,
42
+ 128,
43
+ 4,
44
+ 128,
45
+ 4,
46
+ 128,
47
+ 4,
48
+ 128,
49
+ 4,
50
+ 128,
51
+ 4
52
+ ],
53
+ "compress_rope_parameters": {
54
+ "beta_fast": 32,
55
+ "beta_slow": 1,
56
+ "factor": 16,
57
+ "original_max_position_embeddings": 65536,
58
+ "partial_rotary_factor": 0.125,
59
+ "rope_theta": 160000,
60
+ "rope_type": "yarn",
61
+ "type": "yarn"
62
+ },
63
+ "compress_rope_theta": 160000,
64
+ "dtype": "bfloat16",
65
+ "eos_token_id": 1,
66
+ "expert_dtype": "fp4",
67
+ "first_k_dense_replace": null,
68
+ "hc_eps": 1e-06,
69
+ "hc_mult": 4,
70
+ "hc_sinkhorn_iters": 20,
71
+ "head_dim": 512,
72
+ "hidden_act": "silu",
73
+ "hidden_size": 4096,
74
+ "index_head_dim": 128,
75
+ "index_n_heads": 64,
76
+ "index_topk": 512,
77
+ "initializer_range": 0.02,
78
+ "intermediate_size": 18432,
79
+ "kv_lora_rank": null,
80
+ "max_position_embeddings": 1048576,
81
+ "model_type": "deepseek_v4",
82
+ "moe_intermediate_size": 2048,
83
+ "n_group": null,
84
+ "n_routed_experts": 256,
85
+ "n_shared_experts": 1,
86
+ "norm_topk_prob": true,
87
+ "num_attention_heads": 64,
88
+ "num_experts_per_tok": 6,
89
+ "num_hash_layers": 3,
90
+ "num_hidden_layers": 43,
91
+ "num_key_value_heads": 1,
92
+ "num_nextn_predict_layers": 1,
93
+ "o_groups": 8,
94
+ "o_lora_rank": 1024,
95
+ "output_router_logits": false,
96
+ "pad_token_id": null,
97
+ "partial_rotary_factor": 0.125,
98
+ "pretraining_tp": 1,
99
+ "q_lora_rank": 1024,
100
+ "qk_nope_head_dim": 448,
101
+ "qk_rope_head_dim": 64,
102
+ "quantization_config": {
103
+ "config_groups": {
104
+ "group_0": {
105
+ "format": "float-quantized",
106
+ "input_activations": {
107
+ "actorder": null,
108
+ "block_structure": null,
109
+ "dynamic": true,
110
+ "group_size": 128,
111
+ "num_bits": 8,
112
+ "observer": null,
113
+ "observer_kwargs": {},
114
+ "scale_dtype": null,
115
+ "strategy": "group",
116
+ "symmetric": true,
117
+ "type": "float",
118
+ "zp_dtype": null
119
+ },
120
+ "output_activations": null,
121
+ "targets": [
122
+ "re:model.*attn.*(wkv|wo_a|wo_b|wq_a|wq_b)$",
123
+ "re:model.*attn\\.compressor.*(wgate|wkv)$"
124
+ ],
125
+ "weights": {
126
+ "actorder": null,
127
+ "block_structure": [
128
+ 128,
129
+ 128
130
+ ],
131
+ "dynamic": false,
132
+ "group_size": null,
133
+ "num_bits": 8,
134
+ "observer": "memoryless_minmax",
135
+ "observer_kwargs": {},
136
+ "scale_dtype": null,
137
+ "strategy": "block",
138
+ "symmetric": true,
139
+ "type": "float",
140
+ "zp_dtype": null
141
+ }
142
+ },
143
+ "group_1": {
144
+ "format": "nvfp4-pack-quantized",
145
+ "input_activations": {
146
+ "actorder": null,
147
+ "block_structure": null,
148
+ "dynamic": "local",
149
+ "group_size": 16,
150
+ "num_bits": 4,
151
+ "observer": "static_minmax",
152
+ "observer_kwargs": {},
153
+ "scale_dtype": "torch.float8_e4m3fn",
154
+ "strategy": "tensor_group",
155
+ "symmetric": true,
156
+ "type": "float",
157
+ "zp_dtype": null
158
+ },
159
+ "output_activations": null,
160
+ "targets": [
161
+ "re:model.*mlp.*(gate|up|down)_proj$"
162
+ ],
163
+ "weights": {
164
+ "actorder": null,
165
+ "block_structure": null,
166
+ "dynamic": false,
167
+ "group_size": 16,
168
+ "num_bits": 4,
169
+ "observer": "memoryless_minmax",
170
+ "observer_kwargs": {},
171
+ "scale_dtype": "torch.float8_e4m3fn",
172
+ "strategy": "tensor_group",
173
+ "symmetric": true,
174
+ "type": "float",
175
+ "zp_dtype": null
176
+ }
177
+ }
178
+ },
179
+ "format": "mixed-precision",
180
+ "global_compression_ratio": null,
181
+ "ignore": [
182
+ "model.layers.2.self_attn.compressor.indexer.weights_proj",
183
+ "model.layers.4.self_attn.compressor.indexer.weights_proj",
184
+ "model.layers.6.self_attn.compressor.indexer.weights_proj",
185
+ "model.layers.8.self_attn.compressor.indexer.weights_proj",
186
+ "model.layers.10.self_attn.compressor.indexer.weights_proj",
187
+ "model.layers.12.self_attn.compressor.indexer.weights_proj",
188
+ "model.layers.14.self_attn.compressor.indexer.weights_proj",
189
+ "model.layers.16.self_attn.compressor.indexer.weights_proj",
190
+ "model.layers.18.self_attn.compressor.indexer.weights_proj",
191
+ "model.layers.20.self_attn.compressor.indexer.weights_proj",
192
+ "model.layers.22.self_attn.compressor.indexer.weights_proj",
193
+ "model.layers.24.self_attn.compressor.indexer.weights_proj",
194
+ "model.layers.26.self_attn.compressor.indexer.weights_proj",
195
+ "model.layers.28.self_attn.compressor.indexer.weights_proj",
196
+ "model.layers.30.self_attn.compressor.indexer.weights_proj",
197
+ "model.layers.32.self_attn.compressor.indexer.weights_proj",
198
+ "model.layers.34.self_attn.compressor.indexer.weights_proj",
199
+ "model.layers.36.self_attn.compressor.indexer.weights_proj",
200
+ "model.layers.38.self_attn.compressor.indexer.weights_proj",
201
+ "model.layers.40.self_attn.compressor.indexer.weights_proj",
202
+ "model.layers.42.self_attn.compressor.indexer.weights_proj",
203
+ "lm_head"
204
+ ],
205
+ "kv_cache_scheme": null,
206
+ "quant_method": "compressed-tensors",
207
+ "quantization_status": "compressed",
208
+ "sparsity_config": {},
209
+ "transform_config": {},
210
+ "version": "0.15.1.dev15+g11daf97.d20260427"
211
+ },
212
+ "rms_norm_eps": 1e-06,
213
+ "rope_interleave": true,
214
+ "rope_parameters": {
215
+ "beta_fast": 32,
216
+ "beta_slow": 1,
217
+ "factor": 16,
218
+ "original_max_position_embeddings": 65536,
219
+ "partial_rotary_factor": 0.125,
220
+ "rope_theta": 10000,
221
+ "rope_type": "yarn",
222
+ "type": "yarn"
223
+ },
224
+ "rope_theta": 10000,
225
+ "routed_scaling_factor": 1.5,
226
+ "router_aux_loss_coef": 0.001,
227
+ "router_jitter_noise": 0.0,
228
+ "scoring_func": "sqrtsoftplus",
229
+ "sliding_window": 128,
230
+ "swiglu_limit": 10.0,
231
+ "tie_word_embeddings": false,
232
+ "topk_group": null,
233
+ "topk_method": "noaux_tc",
234
+ "transformers_version": "5.7.0.dev0",
235
+ "use_cache": true,
236
+ "v_head_dim": null,
237
+ "vocab_size": 129280
238
+ }
generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 0,
4
+ "do_sample": true,
5
+ "eos_token_id": 1,
6
+ "temperature": 1.0,
7
+ "top_p": 1.0,
8
+ "transformers_version": "5.7.0.dev0"
9
+ }
model-00001-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7a0db4ccc0dbf59d4efa7827a03c1419e936a4acaa209a159ea21cca2de8056d
3
+ size 50004001220
model-00002-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:218785a9faf4aa9caffec98b8397179cb4d23e52c26a123155c4880342f320d1
3
+ size 50004594224
model-00003-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4c23c539c92ca38c459729c727ea0341ae965efd2ff2639da62411d9f4569d74
3
+ size 50001290592
model-00004-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:136c70a9a390971d7422cbc38ac6944386a8b1892d5c80162846dec7649e9038
3
+ size 13900266712
model.safetensors.index.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2976aeaf60de0cb11ba7bf7330839e58c915a43fcc3f2873f47d88a51e01dda0
3
+ size 13404816
recipe.yaml ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ default_stage:
2
+ default_modifiers:
3
+ QuantizationModifier:
4
+ config_groups:
5
+ attention:
6
+ targets: ['re:model.*attn.*(wkv|wo_a|wo_b|wq_a|wq_b)$', 're:model.*attn\.compressor.*(wgate|wkv)$']
7
+ weights:
8
+ num_bits: 8
9
+ type: float
10
+ symmetric: true
11
+ group_size: null
12
+ strategy: block
13
+ block_structure: [128, 128]
14
+ dynamic: false
15
+ actorder: null
16
+ scale_dtype: null
17
+ zp_dtype: null
18
+ observer: memoryless_minmax
19
+ observer_kwargs: {}
20
+ input_activations:
21
+ num_bits: 8
22
+ type: float
23
+ symmetric: true
24
+ group_size: 128
25
+ strategy: group
26
+ block_structure: null
27
+ dynamic: true
28
+ actorder: null
29
+ scale_dtype: null
30
+ zp_dtype: null
31
+ observer: null
32
+ observer_kwargs: {}
33
+ output_activations: null
34
+ format: null
35
+ experts:
36
+ targets: ['re:model.*mlp.*(gate|up|down)_proj$']
37
+ weights:
38
+ num_bits: 4
39
+ type: float
40
+ symmetric: true
41
+ group_size: 16
42
+ strategy: tensor_group
43
+ block_structure: null
44
+ dynamic: false
45
+ actorder: null
46
+ scale_dtype: torch.float8_e4m3fn
47
+ zp_dtype: null
48
+ observer: memoryless_minmax
49
+ observer_kwargs: {}
50
+ input_activations:
51
+ num_bits: 4
52
+ type: float
53
+ symmetric: true
54
+ group_size: 16
55
+ strategy: tensor_group
56
+ block_structure: null
57
+ dynamic: local
58
+ actorder: null
59
+ scale_dtype: torch.float8_e4m3fn
60
+ zp_dtype: null
61
+ observer: static_minmax
62
+ observer_kwargs: {}
63
+ output_activations: null
64
+ format: null
65
+ targets: [Linear]
66
+ ignore: [lm_head]
67
+ bypass_divisibility_checks: false
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "bos_token": "<｜begin▁of▁sentence｜>",
4
+ "clean_up_tokenization_spaces": false,
5
+ "eos_token": "<｜end▁of▁sentence｜>",
6
+ "is_local": true,
7
+ "legacy": true,
8
+ "local_files_only": false,
9
+ "model_max_length": 1048576,
10
+ "pad_token": "<｜end▁of▁sentence｜>",
11
+ "sp_model_kwargs": {},
12
+ "tokenizer_class": "TokenizersBackend",
13
+ "unk_token": null
14
+ }