Hugging Face's logo

lyf
/
Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated-NVFP4

Image-Text-to-Text
Transformers
Safetensors
qwen3_5
quantized
nvfp4
fp4
4-bit precision
compressed-tensors
llm-compressor
vllm
multimodal
reasoning
conversational
8-bit precision
Model card Files
xet
 Community
lyf commited on
Commit
9c0286f
·
verified ·
1 Parent(s): e80df14

Add files using upload-large-folder tool

Browse files
Files changed (16) hide show
  1. .gitattributes +1 -0
  2. README.md +99 -0
  3. chat_template.jinja +88 -0
  4. config.json +404 -0
  5. generation_config.json +7 -0
  6. model-00001-of-00005.safetensors +3 -0
  7. model-00002-of-00005.safetensors +3 -0
  8. model-00003-of-00005.safetensors +3 -0
  9. model-00004-of-00005.safetensors +3 -0
  10. model-00005-of-00005.safetensors +3 -0
  11. model-multimodal-extra.safetensors +3 -0
  12. model.safetensors.index.json +0 -0
  13. processor_config.json +63 -0
  14. recipe.yaml +7 -0
  15. tokenizer.json +3 -0
  16. tokenizer_config.json +33 -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
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: transformers
4
+ pipeline_tag: image-text-to-text
5
+ base_model: huihui-ai/Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated
6
+ base_model_relation: quantized
7
+ tags:
8
+ - transformers
9
+ - safetensors
10
+ - qwen3_5
11
+ - quantized
12
+ - nvfp4
13
+ - fp4
14
+ - 4-bit
15
+ - compressed-tensors
16
+ - llm-compressor
17
+ - vllm
18
+ - image-text-to-text
19
+ - multimodal
20
+ - reasoning
21
+ datasets:
22
+ - neuralmagic/calibration
23
+ ---
24
+
25
+ # Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated-NVFP4
26
+
27
+ This repository contains an NVFP4-compressed version of [huihui-ai/Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated](https://huggingface.co/huihui-ai/Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated).
28
+
29
+ The language model weights are compressed to NVFP4 for efficient inference on recent NVIDIA GPUs, while the multimodal weights are kept in BF16 and repacked into a separate `model-multimodal-extra.safetensors` file so that `Qwen3_5ForConditionalGeneration` behavior is preserved.
30
+
31
+ ## What Was Quantized
32
+
33
+ - Source model: `huihui-ai/Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated`
34
+ - Quantization method: `llmcompressor` one-shot NVFP4
35
+ - Calibration dataset: `neuralmagic/calibration` (`LLM` split)
36
+ - Calibration samples: `512`
37
+ - Calibration sequence length: `2048`
38
+ - Quantized targets: `Linear`
39
+ - Excluded from quantization:
40
+ - `lm_head`
41
+ - all `model.visual.*` linear layers
42
+ - `linear_attn.in_proj_a`
43
+ - `linear_attn.in_proj_b`
44
+
45
+ ## Repository Layout
46
+
47
+ - `model-00001-of-00005.safetensors` to `model-00005-of-00005.safetensors`
48
+ - NVFP4 main language-model shards
49
+ - `model-multimodal-extra.safetensors`
50
+ - BF16 multimodal tensors preserved from the source checkpoint
51
+ - `model.safetensors.index.json`
52
+ - combined index for the main NVFP4 shards plus multimodal extra tensors
53
+ - `processor_config.json`
54
+ - multimodal processor config copied from the source model
55
+ - `recipe.yaml`
56
+ - the quantization recipe used for this build
57
+
58
+ ### Stored Tensor Metadata
59
+
60
+ - `total_parameters`: `16713682960`
61
+ - `total_size`: `19743450720`
62
+ - `hybrid_extra_tensor_count`: `333`
63
+ - `hybrid_extra_tensor_bytes`: `921460192`
64
+
65
+ ## Serving Notes
66
+
67
+ Tested locally with:
68
+
69
+ - `vllm/vllm-openai:cu130-nightly`
70
+ - vLLM `0.17.2rc1.dev153+g39474513f`
71
+ - NVIDIA RTX 5090
72
+ - `VLLM_NVFP4_GEMM_BACKEND=marlin`
73
+
74
+ Observed behavior with reasoning enabled:
75
+
76
+ - `POST /v1/chat/completions`
77
+ - returns `message.reasoning`
78
+ - can also return a normal `message.content` if `max_tokens` is large enough
79
+ - `POST /v1/responses`
80
+ - returns reasoning blocks under `output[].type = "reasoning"`
81
+ - returns final text under `output[].type = "message"` and `content[].type = "output_text"`
82
+
83
+ For robust client integration, prefer reading the structured `responses` output instead of assuming the top-level `text` field is populated.
84
+
85
+ ## Example vLLM Command
86
+
87
+ ```bash
88
+ vllm serve /path/to/Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated-NVFP4 \
89
+ --reasoning-parser qwen3 \
90
+ --enable-auto-tool-choice \
91
+ --tool-call-parser qwen3_coder \
92
+ --gpu-memory-utilization 0.95 \
93
+ --kv-cache-dtype fp8
94
+ ```
95
+
96
+ ## Notes
97
+
98
+ - The source model's safety, licensing, and usage constraints still apply.
99
+ - This repo keeps multimodal capability by preserving the original visual tower in BF16 instead of re-quantizing it.
chat_template.jinja ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if tools %}
2
+ {{- '<|im_start|>system\n' }}
3
+ {%- if messages[0].role == 'system' %}
4
+ {{- messages[0].content + '\n\n' }}
5
+ {%- endif %}
6
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
7
+ {%- for tool in tools %}
8
+ {{- "\n" }}
9
+ {{- tool | tojson }}
10
+ {%- endfor %}
11
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
12
+ {%- else %}
13
+ {%- if messages[0].role == 'system' %}
14
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
15
+ {%- endif %}
16
+ {%- endif %}
17
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
18
+ {%- for message in messages[::-1] %}
19
+ {%- set index = (messages|length - 1) - loop.index0 %}
20
+ {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
21
+ {%- set ns.multi_step_tool = false %}
22
+ {%- set ns.last_query_index = index %}
23
+ {%- endif %}
24
+ {%- endfor %}
25
+ {%- for message in messages %}
26
+ {%- if message.content is string %}
27
+ {%- set content = message.content %}
28
+ {%- else %}
29
+ {%- set content = '' %}
30
+ {%- endif %}
31
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
32
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
33
+ {%- elif message.role == "assistant" %}
34
+ {%- set reasoning_content = '' %}
35
+ {%- if message.reasoning_content is string %}
36
+ {%- set reasoning_content = message.reasoning_content %}
37
+ {%- else %}
38
+ {%- if '</think>' in content %}
39
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
40
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if loop.index0 > ns.last_query_index %}
44
+ {%- if loop.last or (not loop.last and reasoning_content) %}
45
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
46
+ {%- else %}
47
+ {{- '<|im_start|>' + message.role + '\n' + content }}
48
+ {%- endif %}
49
+ {%- else %}
50
+ {{- '<|im_start|>' + message.role + '\n' + content }}
51
+ {%- endif %}
52
+ {%- if message.tool_calls %}
53
+ {%- for tool_call in message.tool_calls %}
54
+ {%- if (loop.first and content) or (not loop.first) %}
55
+ {{- '\n' }}
56
+ {%- endif %}
57
+ {%- if tool_call.function %}
58
+ {%- set tool_call = tool_call.function %}
59
+ {%- endif %}
60
+ {{- '<tool_call>\n{"name": "' }}
61
+ {{- tool_call.name }}
62
+ {{- '", "arguments": ' }}
63
+ {%- if tool_call.arguments is string %}
64
+ {{- tool_call.arguments }}
65
+ {%- else %}
66
+ {{- tool_call.arguments | tojson }}
67
+ {%- endif %}
68
+ {{- '}\n</tool_call>' }}
69
+ {%- endfor %}
70
+ {%- endif %}
71
+ {{- '<|im_end|>\n' }}
72
+ {%- elif message.role == "tool" %}
73
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
74
+ {{- '<|im_start|>user' }}
75
+ {%- endif %}
76
+ {{- '\n<tool_response>\n' }}
77
+ {{- content }}
78
+ {{- '\n</tool_response>' }}
79
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
80
+ {{- '<|im_end|>\n' }}
81
+ {%- endif %}
82
+ {%- endif %}
83
+ {%- endfor %}
84
+ {%- if add_generation_prompt %}
85
+ {{- '<|im_start|>assistant
86
+ <think>
87
+ ' }}
88
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,404 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3_5ForConditionalGeneration"
4
+ ],
5
+ "bos_token_id": null,
6
+ "torch_dtype": "bfloat16",
7
+ "eos_token_id": 248046,
8
+ "image_token_id": 248056,
9
+ "model_name": "qwen/Qwen3.5-27B",
10
+ "model_type": "qwen3_5",
11
+ "pad_token_id": 248044,
12
+ "text_config": {
13
+ "attention_bias": false,
14
+ "attention_dropout": 0.0,
15
+ "attn_output_gate": true,
16
+ "bos_token_id": null,
17
+ "torch_dtype": "bfloat16",
18
+ "eos_token_id": 248044,
19
+ "full_attention_interval": 4,
20
+ "head_dim": 256,
21
+ "hidden_act": "silu",
22
+ "hidden_size": 5120,
23
+ "initializer_range": 0.02,
24
+ "intermediate_size": 17408,
25
+ "layer_types": [
26
+ "linear_attention",
27
+ "linear_attention",
28
+ "linear_attention",
29
+ "full_attention",
30
+ "linear_attention",
31
+ "linear_attention",
32
+ "linear_attention",
33
+ "full_attention",
34
+ "linear_attention",
35
+ "linear_attention",
36
+ "linear_attention",
37
+ "full_attention",
38
+ "linear_attention",
39
+ "linear_attention",
40
+ "linear_attention",
41
+ "full_attention",
42
+ "linear_attention",
43
+ "linear_attention",
44
+ "linear_attention",
45
+ "full_attention",
46
+ "linear_attention",
47
+ "linear_attention",
48
+ "linear_attention",
49
+ "full_attention",
50
+ "linear_attention",
51
+ "linear_attention",
52
+ "linear_attention",
53
+ "full_attention",
54
+ "linear_attention",
55
+ "linear_attention",
56
+ "linear_attention",
57
+ "full_attention",
58
+ "linear_attention",
59
+ "linear_attention",
60
+ "linear_attention",
61
+ "full_attention",
62
+ "linear_attention",
63
+ "linear_attention",
64
+ "linear_attention",
65
+ "full_attention",
66
+ "linear_attention",
67
+ "linear_attention",
68
+ "linear_attention",
69
+ "full_attention",
70
+ "linear_attention",
71
+ "linear_attention",
72
+ "linear_attention",
73
+ "full_attention",
74
+ "linear_attention",
75
+ "linear_attention",
76
+ "linear_attention",
77
+ "full_attention",
78
+ "linear_attention",
79
+ "linear_attention",
80
+ "linear_attention",
81
+ "full_attention",
82
+ "linear_attention",
83
+ "linear_attention",
84
+ "linear_attention",
85
+ "full_attention",
86
+ "linear_attention",
87
+ "linear_attention",
88
+ "linear_attention",
89
+ "full_attention"
90
+ ],
91
+ "linear_conv_kernel_dim": 4,
92
+ "linear_key_head_dim": 128,
93
+ "linear_num_key_heads": 16,
94
+ "linear_num_value_heads": 48,
95
+ "linear_value_head_dim": 128,
96
+ "mamba_ssm_dtype": "float32",
97
+ "max_position_embeddings": 262144,
98
+ "mlp_only_layers": [],
99
+ "model_type": "qwen3_5_text",
100
+ "mtp_num_hidden_layers": 1,
101
+ "mtp_use_dedicated_embeddings": false,
102
+ "num_attention_heads": 24,
103
+ "num_hidden_layers": 64,
104
+ "num_key_value_heads": 4,
105
+ "pad_token_id": null,
106
+ "partial_rotary_factor": 0.25,
107
+ "rms_norm_eps": 1e-06,
108
+ "rope_parameters": {
109
+ "mrope_interleaved": true,
110
+ "mrope_section": [
111
+ 11,
112
+ 11,
113
+ 10
114
+ ],
115
+ "partial_rotary_factor": 0.25,
116
+ "rope_theta": 10000000,
117
+ "rope_type": "default"
118
+ },
119
+ "tie_word_embeddings": false,
120
+ "use_cache": true,
121
+ "vocab_size": 248320
122
+ },
123
+ "tie_word_embeddings": false,
124
+ "unsloth_version": "2026.3.3",
125
+ "use_cache": false,
126
+ "video_token_id": 248057,
127
+ "vision_config": {
128
+ "deepstack_visual_indexes": [],
129
+ "depth": 27,
130
+ "torch_dtype": "bfloat16",
131
+ "hidden_act": "gelu_pytorch_tanh",
132
+ "hidden_size": 1152,
133
+ "in_channels": 3,
134
+ "initializer_range": 0.02,
135
+ "intermediate_size": 4304,
136
+ "model_type": "qwen3_5",
137
+ "num_heads": 16,
138
+ "num_position_embeddings": 2304,
139
+ "out_hidden_size": 5120,
140
+ "patch_size": 16,
141
+ "spatial_merge_size": 2,
142
+ "temporal_patch_size": 2
143
+ },
144
+ "vision_end_token_id": 248054,
145
+ "vision_start_token_id": 248053,
146
+ "quantization_config": {
147
+ "config_groups": {
148
+ "group_0": {
149
+ "format": "nvfp4-pack-quantized",
150
+ "input_activations": {
151
+ "actorder": null,
152
+ "block_structure": null,
153
+ "dynamic": "local",
154
+ "group_size": 16,
155
+ "num_bits": 4,
156
+ "observer": "static_minmax",
157
+ "observer_kwargs": {},
158
+ "scale_dtype": "torch.float8_e4m3fn",
159
+ "strategy": "tensor_group",
160
+ "symmetric": true,
161
+ "type": "float",
162
+ "zp_dtype": null
163
+ },
164
+ "output_activations": null,
165
+ "targets": [
166
+ "Linear"
167
+ ],
168
+ "weights": {
169
+ "actorder": null,
170
+ "block_structure": null,
171
+ "dynamic": false,
172
+ "group_size": 16,
173
+ "num_bits": 4,
174
+ "observer": "memoryless_minmax",
175
+ "observer_kwargs": {},
176
+ "scale_dtype": "torch.float8_e4m3fn",
177
+ "strategy": "tensor_group",
178
+ "symmetric": true,
179
+ "type": "float",
180
+ "zp_dtype": null
181
+ }
182
+ }
183
+ },
184
+ "format": "nvfp4-pack-quantized",
185
+ "global_compression_ratio": null,
186
+ "ignore": [
187
+ "model.language_model.layers.0.linear_attn.in_proj_b",
188
+ "model.language_model.layers.0.linear_attn.in_proj_a",
189
+ "model.language_model.layers.1.linear_attn.in_proj_b",
190
+ "model.language_model.layers.1.linear_attn.in_proj_a",
191
+ "model.language_model.layers.2.linear_attn.in_proj_b",
192
+ "model.language_model.layers.2.linear_attn.in_proj_a",
193
+ "model.language_model.layers.4.linear_attn.in_proj_b",
194
+ "model.language_model.layers.4.linear_attn.in_proj_a",
195
+ "model.language_model.layers.5.linear_attn.in_proj_b",
196
+ "model.language_model.layers.5.linear_attn.in_proj_a",
197
+ "model.language_model.layers.6.linear_attn.in_proj_b",
198
+ "model.language_model.layers.6.linear_attn.in_proj_a",
199
+ "model.language_model.layers.8.linear_attn.in_proj_b",
200
+ "model.language_model.layers.8.linear_attn.in_proj_a",
201
+ "model.language_model.layers.9.linear_attn.in_proj_b",
202
+ "model.language_model.layers.9.linear_attn.in_proj_a",
203
+ "model.language_model.layers.10.linear_attn.in_proj_b",
204
+ "model.language_model.layers.10.linear_attn.in_proj_a",
205
+ "model.language_model.layers.12.linear_attn.in_proj_b",
206
+ "model.language_model.layers.12.linear_attn.in_proj_a",
207
+ "model.language_model.layers.13.linear_attn.in_proj_b",
208
+ "model.language_model.layers.13.linear_attn.in_proj_a",
209
+ "model.language_model.layers.14.linear_attn.in_proj_b",
210
+ "model.language_model.layers.14.linear_attn.in_proj_a",
211
+ "model.language_model.layers.16.linear_attn.in_proj_b",
212
+ "model.language_model.layers.16.linear_attn.in_proj_a",
213
+ "model.language_model.layers.17.linear_attn.in_proj_b",
214
+ "model.language_model.layers.17.linear_attn.in_proj_a",
215
+ "model.language_model.layers.18.linear_attn.in_proj_b",
216
+ "model.language_model.layers.18.linear_attn.in_proj_a",
217
+ "model.language_model.layers.20.linear_attn.in_proj_b",
218
+ "model.language_model.layers.20.linear_attn.in_proj_a",
219
+ "model.language_model.layers.21.linear_attn.in_proj_b",
220
+ "model.language_model.layers.21.linear_attn.in_proj_a",
221
+ "model.language_model.layers.22.linear_attn.in_proj_b",
222
+ "model.language_model.layers.22.linear_attn.in_proj_a",
223
+ "model.language_model.layers.24.linear_attn.in_proj_b",
224
+ "model.language_model.layers.24.linear_attn.in_proj_a",
225
+ "model.language_model.layers.25.linear_attn.in_proj_b",
226
+ "model.language_model.layers.25.linear_attn.in_proj_a",
227
+ "model.language_model.layers.26.linear_attn.in_proj_b",
228
+ "model.language_model.layers.26.linear_attn.in_proj_a",
229
+ "model.language_model.layers.28.linear_attn.in_proj_b",
230
+ "model.language_model.layers.28.linear_attn.in_proj_a",
231
+ "model.language_model.layers.29.linear_attn.in_proj_b",
232
+ "model.language_model.layers.29.linear_attn.in_proj_a",
233
+ "model.language_model.layers.30.linear_attn.in_proj_b",
234
+ "model.language_model.layers.30.linear_attn.in_proj_a",
235
+ "model.language_model.layers.32.linear_attn.in_proj_b",
236
+ "model.language_model.layers.32.linear_attn.in_proj_a",
237
+ "model.language_model.layers.33.linear_attn.in_proj_b",
238
+ "model.language_model.layers.33.linear_attn.in_proj_a",
239
+ "model.language_model.layers.34.linear_attn.in_proj_b",
240
+ "model.language_model.layers.34.linear_attn.in_proj_a",
241
+ "model.language_model.layers.36.linear_attn.in_proj_b",
242
+ "model.language_model.layers.36.linear_attn.in_proj_a",
243
+ "model.language_model.layers.37.linear_attn.in_proj_b",
244
+ "model.language_model.layers.37.linear_attn.in_proj_a",
245
+ "model.language_model.layers.38.linear_attn.in_proj_b",
246
+ "model.language_model.layers.38.linear_attn.in_proj_a",
247
+ "model.language_model.layers.40.linear_attn.in_proj_b",
248
+ "model.language_model.layers.40.linear_attn.in_proj_a",
249
+ "model.language_model.layers.41.linear_attn.in_proj_b",
250
+ "model.language_model.layers.41.linear_attn.in_proj_a",
251
+ "model.language_model.layers.42.linear_attn.in_proj_b",
252
+ "model.language_model.layers.42.linear_attn.in_proj_a",
253
+ "model.language_model.layers.44.linear_attn.in_proj_b",
254
+ "model.language_model.layers.44.linear_attn.in_proj_a",
255
+ "model.language_model.layers.45.linear_attn.in_proj_b",
256
+ "model.language_model.layers.45.linear_attn.in_proj_a",
257
+ "model.language_model.layers.46.linear_attn.in_proj_b",
258
+ "model.language_model.layers.46.linear_attn.in_proj_a",
259
+ "model.language_model.layers.48.linear_attn.in_proj_b",
260
+ "model.language_model.layers.48.linear_attn.in_proj_a",
261
+ "model.language_model.layers.49.linear_attn.in_proj_b",
262
+ "model.language_model.layers.49.linear_attn.in_proj_a",
263
+ "model.language_model.layers.50.linear_attn.in_proj_b",
264
+ "model.language_model.layers.50.linear_attn.in_proj_a",
265
+ "model.language_model.layers.52.linear_attn.in_proj_b",
266
+ "model.language_model.layers.52.linear_attn.in_proj_a",
267
+ "model.language_model.layers.53.linear_attn.in_proj_b",
268
+ "model.language_model.layers.53.linear_attn.in_proj_a",
269
+ "model.language_model.layers.54.linear_attn.in_proj_b",
270
+ "model.language_model.layers.54.linear_attn.in_proj_a",
271
+ "model.language_model.layers.56.linear_attn.in_proj_b",
272
+ "model.language_model.layers.56.linear_attn.in_proj_a",
273
+ "model.language_model.layers.57.linear_attn.in_proj_b",
274
+ "model.language_model.layers.57.linear_attn.in_proj_a",
275
+ "model.language_model.layers.58.linear_attn.in_proj_b",
276
+ "model.language_model.layers.58.linear_attn.in_proj_a",
277
+ "model.language_model.layers.60.linear_attn.in_proj_b",
278
+ "model.language_model.layers.60.linear_attn.in_proj_a",
279
+ "model.language_model.layers.61.linear_attn.in_proj_b",
280
+ "model.language_model.layers.61.linear_attn.in_proj_a",
281
+ "model.language_model.layers.62.linear_attn.in_proj_b",
282
+ "model.language_model.layers.62.linear_attn.in_proj_a",
283
+ "lm_head",
284
+ "model.visual.blocks.0.attn.qkv",
285
+ "model.visual.blocks.0.attn.proj",
286
+ "model.visual.blocks.0.mlp.linear_fc1",
287
+ "model.visual.blocks.0.mlp.linear_fc2",
288
+ "model.visual.blocks.1.attn.qkv",
289
+ "model.visual.blocks.1.attn.proj",
290
+ "model.visual.blocks.1.mlp.linear_fc1",
291
+ "model.visual.blocks.1.mlp.linear_fc2",
292
+ "model.visual.blocks.2.attn.qkv",
293
+ "model.visual.blocks.2.attn.proj",
294
+ "model.visual.blocks.2.mlp.linear_fc1",
295
+ "model.visual.blocks.2.mlp.linear_fc2",
296
+ "model.visual.blocks.3.attn.qkv",
297
+ "model.visual.blocks.3.attn.proj",
298
+ "model.visual.blocks.3.mlp.linear_fc1",
299
+ "model.visual.blocks.3.mlp.linear_fc2",
300
+ "model.visual.blocks.4.attn.qkv",
301
+ "model.visual.blocks.4.attn.proj",
302
+ "model.visual.blocks.4.mlp.linear_fc1",
303
+ "model.visual.blocks.4.mlp.linear_fc2",
304
+ "model.visual.blocks.5.attn.qkv",
305
+ "model.visual.blocks.5.attn.proj",
306
+ "model.visual.blocks.5.mlp.linear_fc1",
307
+ "model.visual.blocks.5.mlp.linear_fc2",
308
+ "model.visual.blocks.6.attn.qkv",
309
+ "model.visual.blocks.6.attn.proj",
310
+ "model.visual.blocks.6.mlp.linear_fc1",
311
+ "model.visual.blocks.6.mlp.linear_fc2",
312
+ "model.visual.blocks.7.attn.qkv",
313
+ "model.visual.blocks.7.attn.proj",
314
+ "model.visual.blocks.7.mlp.linear_fc1",
315
+ "model.visual.blocks.7.mlp.linear_fc2",
316
+ "model.visual.blocks.8.attn.qkv",
317
+ "model.visual.blocks.8.attn.proj",
318
+ "model.visual.blocks.8.mlp.linear_fc1",
319
+ "model.visual.blocks.8.mlp.linear_fc2",
320
+ "model.visual.blocks.9.attn.qkv",
321
+ "model.visual.blocks.9.attn.proj",
322
+ "model.visual.blocks.9.mlp.linear_fc1",
323
+ "model.visual.blocks.9.mlp.linear_fc2",
324
+ "model.visual.blocks.10.attn.qkv",
325
+ "model.visual.blocks.10.attn.proj",
326
+ "model.visual.blocks.10.mlp.linear_fc1",
327
+ "model.visual.blocks.10.mlp.linear_fc2",
328
+ "model.visual.blocks.11.attn.qkv",
329
+ "model.visual.blocks.11.attn.proj",
330
+ "model.visual.blocks.11.mlp.linear_fc1",
331
+ "model.visual.blocks.11.mlp.linear_fc2",
332
+ "model.visual.blocks.12.attn.qkv",
333
+ "model.visual.blocks.12.attn.proj",
334
+ "model.visual.blocks.12.mlp.linear_fc1",
335
+ "model.visual.blocks.12.mlp.linear_fc2",
336
+ "model.visual.blocks.13.attn.qkv",
337
+ "model.visual.blocks.13.attn.proj",
338
+ "model.visual.blocks.13.mlp.linear_fc1",
339
+ "model.visual.blocks.13.mlp.linear_fc2",
340
+ "model.visual.blocks.14.attn.qkv",
341
+ "model.visual.blocks.14.attn.proj",
342
+ "model.visual.blocks.14.mlp.linear_fc1",
343
+ "model.visual.blocks.14.mlp.linear_fc2",
344
+ "model.visual.blocks.15.attn.qkv",
345
+ "model.visual.blocks.15.attn.proj",
346
+ "model.visual.blocks.15.mlp.linear_fc1",
347
+ "model.visual.blocks.15.mlp.linear_fc2",
348
+ "model.visual.blocks.16.attn.qkv",
349
+ "model.visual.blocks.16.attn.proj",
350
+ "model.visual.blocks.16.mlp.linear_fc1",
351
+ "model.visual.blocks.16.mlp.linear_fc2",
352
+ "model.visual.blocks.17.attn.qkv",
353
+ "model.visual.blocks.17.attn.proj",
354
+ "model.visual.blocks.17.mlp.linear_fc1",
355
+ "model.visual.blocks.17.mlp.linear_fc2",
356
+ "model.visual.blocks.18.attn.qkv",
357
+ "model.visual.blocks.18.attn.proj",
358
+ "model.visual.blocks.18.mlp.linear_fc1",
359
+ "model.visual.blocks.18.mlp.linear_fc2",
360
+ "model.visual.blocks.19.attn.qkv",
361
+ "model.visual.blocks.19.attn.proj",
362
+ "model.visual.blocks.19.mlp.linear_fc1",
363
+ "model.visual.blocks.19.mlp.linear_fc2",
364
+ "model.visual.blocks.20.attn.qkv",
365
+ "model.visual.blocks.20.attn.proj",
366
+ "model.visual.blocks.20.mlp.linear_fc1",
367
+ "model.visual.blocks.20.mlp.linear_fc2",
368
+ "model.visual.blocks.21.attn.qkv",
369
+ "model.visual.blocks.21.attn.proj",
370
+ "model.visual.blocks.21.mlp.linear_fc1",
371
+ "model.visual.blocks.21.mlp.linear_fc2",
372
+ "model.visual.blocks.22.attn.qkv",
373
+ "model.visual.blocks.22.attn.proj",
374
+ "model.visual.blocks.22.mlp.linear_fc1",
375
+ "model.visual.blocks.22.mlp.linear_fc2",
376
+ "model.visual.blocks.23.attn.qkv",
377
+ "model.visual.blocks.23.attn.proj",
378
+ "model.visual.blocks.23.mlp.linear_fc1",
379
+ "model.visual.blocks.23.mlp.linear_fc2",
380
+ "model.visual.blocks.24.attn.qkv",
381
+ "model.visual.blocks.24.attn.proj",
382
+ "model.visual.blocks.24.mlp.linear_fc1",
383
+ "model.visual.blocks.24.mlp.linear_fc2",
384
+ "model.visual.blocks.25.attn.qkv",
385
+ "model.visual.blocks.25.attn.proj",
386
+ "model.visual.blocks.25.mlp.linear_fc1",
387
+ "model.visual.blocks.25.mlp.linear_fc2",
388
+ "model.visual.blocks.26.attn.qkv",
389
+ "model.visual.blocks.26.attn.proj",
390
+ "model.visual.blocks.26.mlp.linear_fc1",
391
+ "model.visual.blocks.26.mlp.linear_fc2",
392
+ "model.visual.merger.linear_fc1",
393
+ "model.visual.merger.linear_fc2"
394
+ ],
395
+ "kv_cache_scheme": null,
396
+ "quant_method": "compressed-tensors",
397
+ "quantization_status": "compressed",
398
+ "sparsity_config": {},
399
+ "transform_config": {},
400
+ "version": "0.14.0.1"
401
+ },
402
+ "dtype": "bfloat16",
403
+ "transformers_version": "5.3.0.dev0"
404
+ }
generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": 248046,
4
+ "pad_token_id": 248044,
5
+ "transformers_version": "5.3.0.dev0",
6
+ "use_cache": false
7
+ }
model-00001-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86df42517669a86ad5aac1c49776c5a97b36ab72ccb796da209ec07bd17c9013
3
+ size 2542796896
model-00002-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ab9ff5b349c8286e731ea99b48d0e6f2d3bc228dd118f022801cc15c9d94c81
3
+ size 4290446896
model-00003-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be30df9d46198c51b3805bf01f9c270e308d75ba2464c07e80fdc69a18721284
3
+ size 4292703536
model-00004-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e90c45fac02c4b74fcec64a2161e5bbb252b2da31a79efa64a298db607c1fed5
3
+ size 4292703512
model-00005-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d4f86272118e49cac1425e1434635c38f5b528985573c85330e7cc58705c4dd3
3
+ size 3403604840
model-multimodal-extra.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8e8dbe7fe0403612bde012dae91ca85d3c1fc4dc31c32e9aee1a6a8c7b88161b
3
+ size 921497200
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
processor_config.json ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_processor": {
3
+ "data_format": "channels_first",
4
+ "do_convert_rgb": true,
5
+ "do_normalize": true,
6
+ "do_rescale": true,
7
+ "do_resize": true,
8
+ "image_mean": [
9
+ 0.5,
10
+ 0.5,
11
+ 0.5
12
+ ],
13
+ "image_processor_type": "Qwen2VLImageProcessorFast",
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "merge_size": 2,
20
+ "patch_size": 16,
21
+ "resample": 3,
22
+ "rescale_factor": 0.00392156862745098,
23
+ "size": {
24
+ "longest_edge": 16777216,
25
+ "shortest_edge": 65536
26
+ },
27
+ "temporal_patch_size": 2
28
+ },
29
+ "processor_class": "Qwen3VLProcessor",
30
+ "video_processor": {
31
+ "data_format": "channels_first",
32
+ "default_to_square": true,
33
+ "do_convert_rgb": true,
34
+ "do_normalize": true,
35
+ "do_rescale": true,
36
+ "do_resize": true,
37
+ "do_sample_frames": true,
38
+ "fps": 2,
39
+ "image_mean": [
40
+ 0.5,
41
+ 0.5,
42
+ 0.5
43
+ ],
44
+ "image_std": [
45
+ 0.5,
46
+ 0.5,
47
+ 0.5
48
+ ],
49
+ "max_frames": 768,
50
+ "merge_size": 2,
51
+ "min_frames": 4,
52
+ "patch_size": 16,
53
+ "resample": 3,
54
+ "rescale_factor": 0.00392156862745098,
55
+ "return_metadata": false,
56
+ "size": {
57
+ "longest_edge": 25165824,
58
+ "shortest_edge": 4096
59
+ },
60
+ "temporal_patch_size": 2,
61
+ "video_processor_type": "Qwen3VLVideoProcessor"
62
+ }
63
+ }
recipe.yaml ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ default_stage:
2
+ default_modifiers:
3
+ QuantizationModifier:
4
+ targets: [Linear]
5
+ ignore: [lm_head, 're:.*visual.*', 're:.*in_proj_a$', 're:.*in_proj_b$']
6
+ scheme: NVFP4
7
+ bypass_divisibility_checks: false
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3cf4da4ae047ec5d7e962adf31efe1c9db4b33ae3079bdf6518e716bf25c6dbd
3
+ size 19989623
tokenizer_config.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|im_end|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": true,
13
+ "model_max_length": 262144,
14
+ "model_specific_special_tokens": {
15
+ "audio_bos_token": "<|audio_start|>",
16
+ "audio_eos_token": "<|audio_end|>",
17
+ "audio_token": "<|audio_pad|>",
18
+ "image_token": "<|image_pad|>",
19
+ "video_token": "<|video_pad|>",
20
+ "vision_bos_token": "<|vision_start|>",
21
+ "vision_eos_token": "<|vision_end|>"
22
+ },
23
+ "pad_token": "<|endoftext|>",
24
+ "padding_side": "right",
25
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
26
+ "processor_class": "Qwen3VLProcessor",
27
+ "split_special_tokens": false,
28
+ "unk_token": null,
29
+ "video_token": "<|video_pad|>",
30
+ "vision_bos_token": "<|vision_start|>",
31
+ "vision_eos_token": "<|vision_end|>",
32
+ "tokenizer_class": "Qwen2TokenizerFast"
33
+ }