Hugging Face's logo

llmfan46
/
Qwen3.6-27B-uncensored-heretic-v2-GPTQ-Int4

Image-Text-to-Text
Transformers
Safetensors
qwen3_5
heretic
uncensored
decensored
abliterated
mpoa
conversational
4-bit precision
gptq
Model card Files
xet
 Community
1
llmfan46 commited on
Commit
b5944d7
·
verified ·
1 Parent(s): 95e2463

Upload folder using huggingface_hub

Browse files
Files changed (16) hide show
  1. .gitattributes +1 -0
  2. README.md +1236 -0
  3. chat_template.jinja +221 -0
  4. config.json +194 -0
  5. generation_config.json +13 -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.safetensors.index.json +0 -0
  12. processor_config.json +60 -0
  13. quant_log.csv +401 -0
  14. quantize_config.json +47 -0
  15. tokenizer.json +3 -0
  16. tokenizer_config.json +37 -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 +1,1236 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: transformers
3
+ license: apache-2.0
4
+ license_link: https://huggingface.co/Qwen/Qwen3.6-27B/blob/main/LICENSE
5
+ pipeline_tag: image-text-to-text
6
+ tags:
7
+ - heretic
8
+ - uncensored
9
+ - decensored
10
+ - abliterated
11
+ - mpoa
12
+ base_model:
13
+ - llmfan46/Qwen3.6-27B-uncensored-heretic-v2
14
+ ---
15
+ <div style="background-color: #ff4444; color: white; padding: 20px; border-radius: 10px; text-align: center; margin: 20px 0;">
16
+ <h2 style="color: white; margin: 0 0 10px 0;">🚨⚠️ I HAVE REACHED HUGGING FACE'S FREE STORAGE LIMIT ⚠️🚨</h2>
17
+ <p style="font-size: 18px; margin: 0 0 15px 0;">I can no longer upload new models unless I can cover the cost of additional storage.<br>I host <b>70+ free models</b> as an independent contributor and this work is unpaid.<br><b>Without your support, no more new models can be uploaded.</b></p>
18
+ <p style="font-size: 20px; margin: 0;">
19
+ <a href="https://patreon.com/LLMfan46" style="color: white; text-decoration: underline;">🎉 Patreon (Monthly)</a> &nbsp;|&nbsp;
20
+ <a href="https://ko-fi.com/llmfan46" style="color: white; text-decoration: underline;">☕ Ko-fi (One-time)</a>
21
+ </p>
22
+ <p style="font-size: 16px; margin: 10px 0 0 0;">Every contribution goes directly toward Hugging Face storage fees to keep models free for everyone.</p>
23
+ </div>
24
+
25
+ ---
26
+
27
+ ### **94% fewer refusals** (6/100 Uncensored vs 92/100 Original) while preserving model quality (0.0021 KL divergence).
28
+
29
+ ## ❤️ Support My Work
30
+ Creating these models takes significant time, work and compute. If you find them useful consider supporting me:
31
+
32
+ ![image/png](https://huggingface.co/llmfan46/Omega-Darker-Gaslight_The-Final-Forgotten-Fever-Dream-24B-ultra-uncensored-heretic-v1/resolve/main/waifu001.webp)
33
+
34
+ | Platform | Link | What you get |
35
+ |----------|------|--------------|
36
+ | 🎉 Patreon | [Monthly support](https://patreon.com/LLMfan46) | Priority model requests |
37
+ | ☕ Ko-fi | [One-time tip](https://ko-fi.com/llmfan46) | My eternal gratitude |
38
+
39
+ Your help will motivate me and would go into further improving my workflow and coverings fees for storage, compute and may even help uncensoring bigger model with rental Cloud GPUs.
40
+
41
+ -----
42
+
43
+ GPTQ-Int4 / 4-bit quantization of [llmfan46/Qwen3.6-27B-uncensored-heretic-v2](https://huggingface.co/llmfan46/Qwen3.6-27B-uncensored-heretic-v2).
44
+
45
+ # This is a decensored version of [Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B), made using [Heretic](https://github.com/p-e-w/heretic) v1.2.0 with a variant of the [Magnitude-Preserving Orthogonal Ablation (MPOA)](https://huggingface.co/blog/grimjim/norm-preserving-biprojected-abliteration) method
46
+
47
+ ## Abliteration parameters
48
+
49
+ | Parameter | Value |
50
+ | :-------- | :---: |
51
+ | **direction_index** | 30.38 |
52
+ | **attn.out_proj.max_weight** | 1.58 |
53
+ | **attn.out_proj.max_weight_position** | 38.93 |
54
+ | **attn.out_proj.min_weight** | 1.51 |
55
+ | **attn.out_proj.min_weight_distance** | 32.78 |
56
+ | **mlp.down_proj.max_weight** | 1.80 |
57
+ | **mlp.down_proj.max_weight_position** | 41.28 |
58
+ | **mlp.down_proj.min_weight** | 0.54 |
59
+ | **mlp.down_proj.min_weight_distance** | 43.66 |
60
+ | **attn.o_proj.max_weight** | 1.99 |
61
+ | **attn.o_proj.max_weight_position** | 48.06 |
62
+ | **attn.o_proj.min_weight** | 1.75 |
63
+ | **attn.o_proj.min_weight_distance** | 39.00 |
64
+
65
+ ## Targeted components
66
+
67
+ * attn.o_proj
68
+ * attn.out_proj
69
+ * mlp.down_proj
70
+
71
+ ## Performance
72
+
73
+ | Metric | This model | Original model ([Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B)) |
74
+ | :----- | :--------: | :---------------------------: |
75
+ | **KL divergence** | <span style="color:darkgoldenrod">0.0021</span> | 0 *(by definition)* |
76
+ | **Refusals** | ✅ <span style="color:darkgreen">6/100</span> | ❌ <span style="color:blue">92/100</span> |
77
+
78
+ Lower refusals indicate fewer content restrictions, while lower KL divergence indicates more closeness to the original model's baseline. Higher refusals cause more rejections, objections, pushbacks, lecturing, censorship, softening and deflections.
79
+
80
+ ## MMLU test results:
81
+
82
+ <span style="color:blue">Original:</span>
83
+
84
+ ============================================================
85
+
86
+ - Total questions: 7021
87
+
88
+ - Correct: 6084
89
+
90
+ - Accuracy: 0.8665 (86.65%)
91
+
92
+ - Parse failures: 0
93
+
94
+ ============================================================
95
+
96
+ Tested subject scores:
97
+ - professional_law: 0.7580 (595/785)
98
+ - moral_scenarios: 0.7715 (341/442)
99
+ - miscellaneous: 0.9399 (360/383)
100
+ - professional_psychology: 0.8987 (284/316)
101
+ - high_school_psychology: 0.9704 (262/270)
102
+ - high_school_macroeconomics: 0.9289 (183/197)
103
+ - elementary_mathematics: 0.8967 (165/184)
104
+ - moral_disputes: 0.8621 (150/174)
105
+ - prehistory: 0.8953 (154/172)
106
+ - philosophy: 0.8868 (141/159)
107
+ - high_school_biology: 0.9539 (145/152)
108
+ - professional_accounting: 0.8531 (122/143)
109
+ - clinical_knowledge: 0.9071 (127/140)
110
+ - high_school_microeconomics: 0.9706 (132/136)
111
+ - nutrition: 0.9111 (123/135)
112
+ - professional_medicine: 0.9478 (127/134)
113
+ - conceptual_physics: 0.9141 (117/128)
114
+ - high_school_mathematics: 0.6378 (81/127)
115
+ - human_aging: 0.8190 (95/116)
116
+ - security_studies: 0.9107 (102/112)
117
+ - high_school_statistics: 0.9009 (100/111)
118
+ - marketing: 0.9633 (105/109)
119
+ - high_school_world_history: 0.9623 (102/106)
120
+ - sociology: 0.9417 (97/103)
121
+ - high_school_government_and_politics: 0.9802 (99/101)
122
+ - high_school_geography: 0.9697 (96/99)
123
+ - high_school_chemistry: 0.8041 (78/97)
124
+ - high_school_us_history: 0.9895 (94/95)
125
+ - virology: 0.5506 (49/89)
126
+ - college_medicine: 0.8409 (74/88)
127
+ - world_religions: 0.9091 (80/88)
128
+ - high_school_physics: 0.8571 (72/84)
129
+ - electrical_engineering: 0.8642 (70/81)
130
+ - astronomy: 0.9747 (77/79)
131
+ - logical_fallacies: 0.9342 (71/76)
132
+ - high_school_european_history: 0.9452 (69/73)
133
+ - anatomy: 0.8169 (58/71)
134
+ - college_biology: 0.9375 (60/64)
135
+ - human_sexuality: 0.8906 (57/64)
136
+ - formal_logic: 0.7969 (51/64)
137
+ - public_relations: 0.7377 (45/61)
138
+ - international_law: 0.9000 (54/60)
139
+ - college_physics: 0.7719 (44/57)
140
+ - college_mathematics: 0.7636 (42/55)
141
+ - econometrics: 0.7963 (43/54)
142
+ - jurisprudence: 0.9057 (48/53)
143
+ - high_school_computer_science: 0.9615 (50/52)
144
+ - machine_learning: 0.8269 (43/52)
145
+ - medical_genetics: 0.9412 (48/51)
146
+ - global_facts: 0.6275 (32/51)
147
+ - management: 0.9000 (45/50)
148
+ - us_foreign_policy: 0.9600 (48/50)
149
+ - college_chemistry: 0.7021 (33/47)
150
+ - abstract_algebra: 0.7021 (33/47)
151
+ - business_ethics: 0.8261 (38/46)
152
+ - college_computer_science: 0.8222 (37/45)
153
+ - computer_security: 0.8372 (36/43)
154
+
155
+
156
+ <span style="color:darkgreen">Heretic:</span>
157
+
158
+ ============================================================
159
+
160
+ - Total questions: 7021
161
+
162
+ - Correct: 6011
163
+
164
+ - Accuracy: 0.8561 (85.61%)
165
+
166
+ - Parse failures: 0
167
+
168
+ ============================================================
169
+
170
+ Tested subject scores:
171
+ - professional_law: 0.7146 (561/785)
172
+ - moral_scenarios: 0.7466 (330/442)
173
+ - miscellaneous: 0.9347 (358/383)
174
+ - professional_psychology: 0.9019 (285/316)
175
+ - high_school_psychology: 0.9704 (262/270)
176
+ - high_school_macroeconomics: 0.9391 (185/197)
177
+ - elementary_mathematics: 0.8967 (165/184)
178
+ - moral_disputes: 0.8621 (150/174)
179
+ - prehistory: 0.8779 (151/172)
180
+ - philosophy: 0.8931 (142/159)
181
+ - high_school_biology: 0.9539 (145/152)
182
+ - professional_accounting: 0.8182 (117/143)
183
+ - clinical_knowledge: 0.9143 (128/140)
184
+ - high_school_microeconomics: 0.9706 (132/136)
185
+ - nutrition: 0.8815 (119/135)
186
+ - professional_medicine: 0.9478 (127/134)
187
+ - conceptual_physics: 0.9141 (117/128)
188
+ - high_school_mathematics: 0.6299 (80/127)
189
+ - human_aging: 0.8103 (94/116)
190
+ - security_studies: 0.8661 (97/112)
191
+ - high_school_statistics: 0.8829 (98/111)
192
+ - marketing: 0.9633 (105/109)
193
+ - high_school_world_history: 0.9528 (101/106)
194
+ - sociology: 0.9417 (97/103)
195
+ - high_school_government_and_politics: 0.9604 (97/101)
196
+ - high_school_geography: 0.9697 (96/99)
197
+ - high_school_chemistry: 0.8041 (78/97)
198
+ - high_school_us_history: 0.9474 (90/95)
199
+ - virology: 0.5281 (47/89)
200
+ - college_medicine: 0.8523 (75/88)
201
+ - world_religions: 0.9205 (81/88)
202
+ - high_school_physics: 0.8571 (72/84)
203
+ - electrical_engineering: 0.8395 (68/81)
204
+ - astronomy: 0.9747 (77/79)
205
+ - logical_fallacies: 0.9342 (71/76)
206
+ - high_school_european_history: 0.9452 (69/73)
207
+ - anatomy: 0.8169 (58/71)
208
+ - college_biology: 0.9531 (61/64)
209
+ - human_sexuality: 0.8750 (56/64)
210
+ - formal_logic: 0.7812 (50/64)
211
+ - public_relations: 0.7377 (45/61)
212
+ - international_law: 0.9167 (55/60)
213
+ - college_physics: 0.8070 (46/57)
214
+ - college_mathematics: 0.7455 (41/55)
215
+ - econometrics: 0.8333 (45/54)
216
+ - jurisprudence: 0.8868 (47/53)
217
+ - high_school_computer_science: 0.9615 (50/52)
218
+ - machine_learning: 0.7692 (40/52)
219
+ - medical_genetics: 0.9412 (48/51)
220
+ - global_facts: 0.6471 (33/51)
221
+ - management: 0.9000 (45/50)
222
+ - us_foreign_policy: 0.9800 (49/50)
223
+ - college_chemistry: 0.6596 (31/47)
224
+ - abstract_algebra: 0.7021 (33/47)
225
+ - business_ethics: 0.7826 (36/46)
226
+ - college_computer_science: 0.8444 (38/45)
227
+ - computer_security: 0.8605 (37/43)
228
+
229
+ MMLU - Massive Multitask Language Understanding, multiple-choice questions across 57 subjects (math, history, law, medicine, etc.).
230
+
231
+ -----
232
+
233
+
234
+ # Qwen3.6-27B
235
+
236
+ <img width="400px" src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.6/logo.png">
237
+
238
+ [![Qwen Chat](https://img.shields.io/badge/%F0%9F%92%9C%EF%B8%8F%20Qwen%20Chat%20-536af5)](https://chat.qwen.ai)
239
+
240
+ > [!Note]
241
+ > This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format.
242
+ >
243
+ > These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, KTransformers, etc.
244
+
245
+ Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience.
246
+
247
+ ## Qwen3.6 Highlights
248
+
249
+ This release delivers substantial upgrades, particularly in
250
+
251
+ - **Agentic Coding:** the model now handles frontend workflows and repository-level reasoning with greater fluency and precision.
252
+ - **Thinking Preservation:** we've introduced a new option to retain reasoning context from historical messages, streamlining iterative development and reducing overhead.
253
+
254
+ ![Benchmark Results](https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3.6/Figures/qwen3.6_27b_score.png)
255
+
256
+ For more details, please refer to our blog post [Qwen3.6-27B](https://qwen.ai/blog?id=qwen3.6-27b).
257
+
258
+ ## Model Overview
259
+
260
+ - Type: Causal Language Model with Vision Encoder
261
+ - Training Stage: Pre-training & Post-training
262
+ - Language Model
263
+ - Number of Parameters: 27B
264
+ - Hidden Dimension: 5120
265
+ - Token Embedding: 248320 (Padded)
266
+ - Number of Layers: 64
267
+ - Hidden Layout: 16 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN))
268
+ - Gated DeltaNet:
269
+ - Number of Linear Attention Heads: 48 for V and 16 for QK
270
+ - Head Dimension: 128
271
+ - Gated Attention:
272
+ - Number of Attention Heads: 24 for Q and 4 for KV
273
+ - Head Dimension: 256
274
+ - Rotary Position Embedding Dimension: 64
275
+ - Feed Forward Network:
276
+ - Intermediate Dimension: 17408
277
+ - LM Output: 248320 (Padded)
278
+ - MTP: trained with multi-steps
279
+ - Context Length: 262,144 natively and extensible up to 1,010,000 tokens.
280
+
281
+
282
+ ## Benchmark Results
283
+
284
+ ### Language
285
+
286
+ <div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;max-width:1000px;margin:0 auto;padding:16px 0">
287
+ <table style="width:100%;border-collapse:collapse;font-size:13px">
288
+ <thead><tr>
289
+ <th style="padding:10px 7px;text-align:left;font-weight:600;border-bottom:2px solid #7c3aed;color:#7c3aed"></th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-27B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-397B-A17B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Gemma4-31B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Claude 4.5 Opus</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-35B-A3B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-27B</th></tr></thead>
290
+ <tbody>
291
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Coding Agent</td></tr>
292
+ <tr>
293
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Verified</td>
294
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.0</td>
295
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.2</td>
296
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.0</td>
297
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.9</td>
298
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">73.4</td>
299
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.2</td>
300
+ </tr>
301
+ <tr>
302
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Pro</td>
303
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.2</td>
304
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.9</td>
305
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">35.7</td>
306
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">57.1</td>
307
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">49.5</td>
308
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">53.5</td>
309
+ </tr>
310
+ <tr>
311
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SWE-bench Multilingual</td>
312
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.3</td>
313
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">69.3</td>
314
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.7</td>
315
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.5</td>
316
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.2</td>
317
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">71.3</td>
318
+ </tr>
319
+ <tr>
320
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Terminal-Bench 2.0</td>
321
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.6</td>
322
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.5</td>
323
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">42.9</td>
324
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.3</td>
325
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.5</td>
326
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.3</td>
327
+ </tr>
328
+ <tr>
329
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SkillsBench <sub><small>Avg5</small></sub></td>
330
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">27.2</td>
331
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">30.0</td>
332
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">23.6</td>
333
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">45.3</td>
334
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.7</td>
335
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.2</td>
336
+ </tr>
337
+ <tr>
338
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">QwenWebBench</td>
339
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1068</td>
340
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1186</td>
341
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1197</td>
342
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1536</td>
343
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1397</td>
344
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">1487</td>
345
+ </tr>
346
+ <tr>
347
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">NL2Repo</td>
348
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">27.3</td>
349
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">32.2</td>
350
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">15.5</td>
351
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">43.2</td>
352
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">29.4</td>
353
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">36.2</td>
354
+ </tr>
355
+ <tr>
356
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Claw-Eval <sub><small>Avg</small></sub></td>
357
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.3</td>
358
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.7</td>
359
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.5</td>
360
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.6</td>
361
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.7</td>
362
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.4</td>
363
+ </tr>
364
+ <tr>
365
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">Claw-Eval <sub><small>Pass^3</small></sub></td>
366
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">46.2</td>
367
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">48.1</td>
368
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">25.0</td>
369
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">59.6</td>
370
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">50.0</td>
371
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">60.6</td>
372
+ </tr>
373
+ <tr>
374
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">QwenClawBench</td>
375
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.2</td>
376
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">51.8</td>
377
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">41.7</td>
378
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.3</td>
379
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.6</td>
380
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">53.4</td>
381
+ </tr>
382
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Knowledge</td></tr>
383
+ <tr>
384
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMLU-Pro</td>
385
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.1</td>
386
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td>
387
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.2</td>
388
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.5</td>
389
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.2</td>
390
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.2</td>
391
+ </tr>
392
+ <tr>
393
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMLU-Redux</td>
394
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.2</td>
395
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.9</td>
396
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.7</td>
397
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">95.6</td>
398
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td>
399
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.5</td>
400
+ </tr>
401
+ <tr>
402
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SuperGPQA</td>
403
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.6</td>
404
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.4</td>
405
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.7</td>
406
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.6</td>
407
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.7</td>
408
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">66.0</td>
409
+ </tr>
410
+ <tr>
411
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">C-Eval</td>
412
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.5</td>
413
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.0</td>
414
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.6</td>
415
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.2</td>
416
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.0</td>
417
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">91.4</td>
418
+ </tr>
419
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">STEM & Reasoning</td></tr>
420
+ <tr>
421
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">GPQA Diamond</td>
422
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.5</td>
423
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.4</td>
424
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>
425
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.0</td>
426
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.0</td>
427
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td>
428
+ </tr>
429
+ <tr>
430
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HLE</td>
431
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">24.3</td>
432
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">28.7</td>
433
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">19.5</td>
434
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">30.8</td>
435
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">21.4</td>
436
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">24.0</td>
437
+ </tr>
438
+ <tr>
439
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">LiveCodeBench v6</td>
440
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td>
441
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.6</td>
442
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.0</td>
443
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.8</td>
444
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.4</td>
445
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.9</td>
446
+ </tr>
447
+ <tr>
448
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Feb 25</td>
449
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.0</td>
450
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.8</td>
451
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">88.7</td>
452
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.9</td>
453
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.7</td>
454
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.8</td>
455
+ </tr>
456
+ <tr>
457
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Nov 25</td>
458
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.8</td>
459
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.7</td>
460
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.5</td>
461
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td>
462
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.1</td>
463
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.7</td>
464
+ </tr>
465
+ <tr>
466
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">HMMT Feb 26</td>
467
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>
468
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.9</td>
469
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.2</td>
470
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.3</td>
471
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.6</td>
472
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>
473
+ </tr>
474
+ <tr>
475
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">IMOAnswerBench</td>
476
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.9</td>
477
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.9</td>
478
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.5</td>
479
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.0</td>
480
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.9</td>
481
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.8</td>
482
+ </tr>
483
+ <tr>
484
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">AIME26</td>
485
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.6</td>
486
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.3</td>
487
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.2</td>
488
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">95.1</td>
489
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.7</td>
490
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.1</td>
491
+ </tr>
492
+ </tbody>
493
+ </table>
494
+
495
+ <p style="margin-top:12px;font-size:10px;opacity:0.7">
496
+ * SWE-Bench Series: Internal agent scaffold (bash + file-edit tools); temp=1.0, top_p=0.95, 200K context window. We correct some problematic tasks in the public set of SWE-bench Pro and evaluate all baselines on the refined benchmark.<br/>
497
+ * Terminal-Bench 2.0: Harbor/Terminus-2 harness; 3h timeout, 32 CPU/48 GB RAM; temp=1.0, top_p=0.95, top_k=20, max_tokens=80K, 256K ctx; avg of 5 runs.<br/>
498
+ * SkillsBench: Evaluated via OpenCode on 78 tasks (self-contained subset, excluding API-dependent tasks); avg of 5 runs.<br/>
499
+ * NL2Repo: Others are evaluated via Claude Code (temp=1.0, top_p=0.95, max_turns=900).<br/>
500
+ * QwenClawBench: A real-user-distribution Claw agent benchmark; temp=0.6, 256K ctx.<br/>
501
+ * QwenWebBench: An internal front-end code generation benchmark; bilingual (EN/CN), 7 categories (Web Design, Web Apps, Games, SVG, Data Visualization, Animation, and 3D); auto-render + multimodal judge (code/visual correctness); BT/Elo rating system.<br/>
502
+ * AIME 26: We use the full AIME 2026 (I & II), where the scores may differ from Qwen 3.5 notes.
503
+ </p>
504
+
505
+ </div>
506
+
507
+
508
+ ### Vision Language
509
+
510
+ <div style="font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif;max-width:1000px;margin:0 auto;padding:16px 0">
511
+ <table style="width:100%;border-collapse:collapse;font-size:13px">
512
+ <thead><tr>
513
+ <th style="padding:10px 7px;text-align:left;font-weight:600;border-bottom:2px solid #7c3aed;color:#7c3aed"></th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-27B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.5-397B-A17B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Gemma4-31B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Claude 4.5 Opus</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-35B-A3B</th><th style="padding:10px 7px;text-align:center;font-weight:500;border-bottom:2px solid #7c3aed;color:#7c3aed;font-size: 14px;">Qwen3.6-27B</th></tr></thead>
514
+ <tbody>
515
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">STEM & Puzzle</td></tr>
516
+ <tr>
517
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMMU</td>
518
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.3</td>
519
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.0</td>
520
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.4</td>
521
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td>
522
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.7</td>
523
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.9</td>
524
+ </tr>
525
+ <tr>
526
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMMU-Pro</td>
527
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.0</td>
528
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.0</td>
529
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.9</td>
530
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.6</td>
531
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.3</td>
532
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.8</td>
533
+ </tr>
534
+ <tr>
535
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MathVista <sub><small>mini</small></sub></td>
536
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.8</td>
537
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
538
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.3</td>
539
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
540
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.4</td>
541
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.4</td>
542
+ </tr>
543
+ <tr>
544
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">DynaMath</td>
545
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.7</td>
546
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.3</td>
547
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.5</td>
548
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.7</td>
549
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.8</td>
550
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.6</td>
551
+ </tr>
552
+ <tr>
553
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VlmsAreBlind</td>
554
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.9</td>
555
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
556
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.2</td>
557
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
558
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.6</td>
559
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.0</td>
560
+ </tr>
561
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">General VQA</td></tr>
562
+ <tr>
563
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RealWorldQA</td>
564
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.7</td>
565
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.9</td>
566
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">72.3</td>
567
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.0</td>
568
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.3</td>
569
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.1</td>
570
+ </tr>
571
+ <tr>
572
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMStar</td>
573
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.0</td>
574
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.8</td>
575
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.3</td>
576
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">73.2</td>
577
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.7</td>
578
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.4</td>
579
+ </tr>
580
+ <tr>
581
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MMBench<sub><small>EN-DEV-v1.1</small></sub></td>
582
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.6</td>
583
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
584
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.9</td>
585
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
586
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.8</td>
587
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.3</td>
588
+ </tr>
589
+ <tr>
590
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">SimpleVQA</td>
591
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">56.0</td>
592
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.1</td>
593
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">52.9</td>
594
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">65.7</td>
595
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">58.9</td>
596
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">56.1</td>
597
+ </tr>
598
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Document Understanding</td></tr>
599
+ <tr>
600
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CharXiv <sub><small>RQ</small></sub></td>
601
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">79.5</td>
602
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">80.8</td>
603
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.9</td>
604
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">68.5</td>
605
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.0</td>
606
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">78.4</td>
607
+ </tr>
608
+ <tr>
609
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CC-OCR</td>
610
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.0</td>
611
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.0</td>
612
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.7</td>
613
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">76.9</td>
614
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.9</td>
615
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.2</td>
616
+ </tr>
617
+ <tr>
618
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">OCRBench</td>
619
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.4</td>
620
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
621
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.1</td>
622
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
623
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.0</td>
624
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">89.4</td>
625
+ </tr>
626
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Spatial Intelligence</td></tr>
627
+ <tr>
628
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">ERQA</td>
629
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">60.5</td>
630
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.5</td>
631
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">57.5</td>
632
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">46.8</td>
633
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">61.8</td>
634
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">62.5</td>
635
+ </tr>
636
+ <tr>
637
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">CountBench</td>
638
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.8</td>
639
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.2</td>
640
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.1</td>
641
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.6</td>
642
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">96.1</td>
643
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">97.8</td>
644
+ </tr>
645
+ <tr>
646
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RefCOCO <sub><small>avg</small></sub></td>
647
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.9</td>
648
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.3</td>
649
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
650
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
651
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.0</td>
652
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">92.5</td>
653
+ </tr>
654
+ <tr>
655
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">EmbSpatialBench</td>
656
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.5</td>
657
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
658
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
659
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
660
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.3</td>
661
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.6</td>
662
+ </tr>
663
+ <tr>
664
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">RefSpatialBench</td>
665
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.7</td>
666
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
667
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">4.7</td>
668
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
669
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.3</td>
670
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.0</td>
671
+ </tr>
672
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Video Understanding</td></tr>
673
+ <tr>
674
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VideoMME<sub><small>(w sub.)</sub></small></td>
675
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.0</td>
676
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.5</td>
677
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
678
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.7</td>
679
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.6</td>
680
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">87.7</td>
681
+ </tr>
682
+ <tr>
683
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">VideoMMMU</td>
684
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">82.3</td>
685
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.7</td>
686
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.6</td>
687
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.4</td>
688
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">83.7</td>
689
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">84.4</td>
690
+ </tr>
691
+ <tr>
692
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MLVU</td>
693
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">85.9</td>
694
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.7</td>
695
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
696
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">81.7</td>
697
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.2</td>
698
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">86.6</td>
699
+ </tr>
700
+ <tr>
701
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">MVBench</td>
702
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.6</td>
703
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">77.6</td>
704
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
705
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.2</td>
706
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">74.6</td>
707
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">75.5</td>
708
+ </tr>
709
+ <tr><td colspan="7" style="padding:8px 12px;font-weight:600;color:#7c3aed;border-bottom:1px solid rgba(124, 58, 237, 0.2);background:rgba(124, 58, 237, 0.1)">Visual Agent</td></tr>
710
+ <tr>
711
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">V*</td>
712
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">93.7</td>
713
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">95.8</td>
714
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
715
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">67.0</td>
716
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">90.1</td>
717
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">94.7</td>
718
+ </tr>
719
+ <tr>
720
+ <td style="padding:7px 7px;padding-left:20px;border-bottom:1px solid rgba(128, 128, 128, 0.15);">AndroidWorld</td>
721
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">64.2</td>
722
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
723
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
724
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
725
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">--</td>
726
+ <td style="padding:7px 7px;text-align:center;border-bottom:1px solid rgba(128, 128, 128, 0.15)">70.3</td>
727
+ </tr>
728
+ </tbody>
729
+ </table>
730
+
731
+ <p style="margin-top:12px;font-size:10px;opacity:0.7">
732
+ * Empty cells (--) indicate scores not yet available or not applicable.
733
+ </p>
734
+
735
+ </div>
736
+
737
+
738
+ ## Quickstart
739
+
740
+ For streamlined integration, we recommend using Qwen3.6 via APIs. Below is a guide to use Qwen3.6 via OpenAI-compatible API.
741
+
742
+ ### Serving Qwen3.6
743
+
744
+ Qwen3.6 can be served via APIs with popular inference frameworks.
745
+ In the following, we show example commands to launch OpenAI-Compatible API servers for Qwen3.6 models.
746
+
747
+ > [!Important]
748
+ > Inference efficiency and throughput vary significantly across frameworks.
749
+ > We recommend using the latest framework versions to ensure optimal performance and compatibility.
750
+ > For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, KTransformers or vLLM are strongly recommended.
751
+
752
+ > [!Important]
753
+ > The model has a default context length of 262,144 tokens.
754
+ > If you encounter out-of-memory (OOM) errors, consider reducing the context window.
755
+ > However, because Qwen3.6 leverages extended context for complex tasks, we advise maintaining a context length of at least 128K tokens to preserve thinking capabilities.
756
+
757
+ #### SGLang
758
+
759
+ [SGLang](https://github.com/sgl-project/sglang) is a fast serving framework for large language models and vision language models.
760
+ `sglang>=0.5.10` is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:
761
+ ```shell
762
+ uv pip install sglang[all]
763
+ ```
764
+ See [its documentation](https://docs.sglang.ai/get_started/install.html) for more details.
765
+
766
+ The following will create API endpoints at `http://localhost:8000/v1`:
767
+
768
+ - **Standard Version**: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
769
+
770
+ ```shell
771
+ python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3
772
+ ```
773
+
774
+ - **Tool Use**: To support tool use, you can use the following command.
775
+
776
+ ```shell
777
+ python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --tool-call-parser qwen3_coder
778
+ ```
779
+
780
+ - **Multi-Token Prediction (MTP)**: The following command is recommended for MTP:
781
+
782
+ ```shell
783
+ python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --speculative-algo NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4
784
+ ```
785
+
786
+ For detailed deployment guide, see the [SGLang Qwen3.5 Cookbook](https://lmsysorg.mintlify.app/cookbook/llm/Qwen/Qwen3.5).
787
+
788
+ #### vLLM
789
+
790
+ [vLLM](https://github.com/vllm-project/vllm) is a high-throughput and memory-efficient inference and serving engine for LLMs.
791
+ `vllm>=0.19.0` is recommended for Qwen3.6, which can be installed using the following command in a fresh environment:
792
+ ```shell
793
+ uv pip install vllm --torch-backend=auto
794
+ ```
795
+ See [its documentation](https://docs.vllm.ai/en/stable/getting_started/installation/index.html) for more details.
796
+
797
+
798
+ The following will create API endpoints at `http://localhost:8000/v1`:
799
+
800
+ - **Standard Version**: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
801
+
802
+ ```shell
803
+ vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3
804
+ ```
805
+
806
+ - **Tool Call**: To support tool use, you can use the following command.
807
+
808
+ ```shell
809
+ vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder
810
+ ```
811
+
812
+ - **Multi-Token Prediction (MTP)**: The following command is recommended for MTP:
813
+
814
+ ```shell
815
+ vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}'
816
+ ```
817
+
818
+ - **Text-Only**: The following command skips the vision encoder and multimodal profiling to free up memory for additional KV cache:
819
+
820
+ ```shell
821
+ vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only
822
+ ```
823
+
824
+ For detailed deployment guide, see the [vLLM Qwen3.5 Recipe](https://docs.vllm.ai/projects/recipes/en/latest/Qwen/Qwen3.5.html).
825
+
826
+ #### KTransformers
827
+
828
+ [KTransformers](https://github.com/kvcache-ai/ktransformers) is a flexible framework for experiencing cutting-edge LLM inference optimizations with CPU-GPU heterogeneous computing.
829
+ For running Qwen3.6 with KTransformers, see the [KTransformers Deployment Guide](https://github.com/kvcache-ai/ktransformers/blob/main/doc/en/Qwen3.5.md).
830
+
831
+ #### Hugging Face Transformers
832
+
833
+ Hugging Face Transformers contains a _lightweight_ server which can be used for quick testing and moderate load deployment.
834
+ The latest `transformers` is required for Qwen3.6:
835
+ ```shell
836
+ pip install "transformers[serving]"
837
+ ```
838
+ See [its documentation](https://huggingface.co/docs/transformers/main/serving) for more details. Please also make sure torchvision and pillow are installed.
839
+
840
+ Then, run `transformers serve` to launch a server with API endpoints at `http://localhost:8000/v1`; it will place the model on accelerators if available:
841
+ ```shell
842
+ transformers serve Qwen/Qwen3.6-27B --port 8000 --continuous-batching
843
+ ```
844
+
845
+ ### Using Qwen3.6 via the Chat Completions API
846
+
847
+ The chat completions API is accessible via standard HTTP requests or OpenAI SDKs.
848
+ Here, we show examples using the OpenAI Python SDK.
849
+
850
+ Before starting, make sure it is installed and the API key and the API base URL is configured, e.g.:
851
+ ```shell
852
+ pip install -U openai
853
+
854
+ # Set the following accordingly
855
+ export OPENAI_BASE_URL="http://localhost:8000/v1"
856
+ export OPENAI_API_KEY="EMPTY"
857
+ ```
858
+
859
+ > [!Tip]
860
+ > We recommend using the following set of sampling parameters for generation
861
+ > - Thinking mode for general tasks: `temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0`
862
+ > - Thinking mode for precise coding tasks (e.g. WebDev): `temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0`
863
+ > - Instruct (or non-thinking) mode: `temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0`
864
+ >
865
+ > Please note that the support for sampling parameters varies according to inference frameworks.
866
+
867
+ > [!Important]
868
+ > Qwen3.6 models operate in thinking mode by default, generating thinking content signified by `<think>\n...</think>\n\n` before producing the final responses.
869
+ > To disable thinking content and obtain direct response, refer to the examples [here](#instruct-or-non-thinking-mode).
870
+
871
+
872
+ #### Text-Only Input
873
+
874
+ ```python
875
+ from openai import OpenAI
876
+ # Configured by environment variables
877
+ client = OpenAI()
878
+
879
+ messages = [
880
+ {"role": "user", "content": "Type \"I love Qwen3.6\" backwards"},
881
+ ]
882
+
883
+ chat_response = client.chat.completions.create(
884
+ model="Qwen/Qwen3.6-27B",
885
+ messages=messages,
886
+ max_tokens=81920,
887
+ temperature=1.0,
888
+ top_p=0.95,
889
+ presence_penalty=0.0,
890
+ extra_body={
891
+ "top_k": 20,
892
+ },
893
+ )
894
+ print("Chat response:", chat_response)
895
+ ```
896
+
897
+
898
+ #### Image Input
899
+
900
+ ```python
901
+ from openai import OpenAI
902
+ # Configured by environment variables
903
+ client = OpenAI()
904
+
905
+ messages = [
906
+ {
907
+ "role": "user",
908
+ "content": [
909
+ {
910
+ "type": "image_url",
911
+ "image_url": {
912
+ "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/CI_Demo/mathv-1327.jpg"
913
+ }
914
+ },
915
+ {
916
+ "type": "text",
917
+ "text": "The centres of the four illustrated circles are in the corners of the square. The two big circles touch each other and also the two little circles. With which factor do you have to multiply the radii of the little circles to obtain the radius of the big circles?\nChoices:\n(A) $\\frac{2}{9}$\n(B) $\\sqrt{5}$\n(C) $0.8 \\cdot \\pi$\n(D) 2.5\n(E) $1+\\sqrt{2}$"
918
+ }
919
+ ]
920
+ }
921
+ ]
922
+
923
+ response = client.chat.completions.create(
924
+ model="Qwen/Qwen3.6-27B",
925
+ messages=messages,
926
+ max_tokens=81920,
927
+ temperature=1.0,
928
+ top_p=0.95,
929
+ presence_penalty=0.0,
930
+ extra_body={
931
+ "top_k": 20,
932
+ },
933
+ )
934
+ print("Chat response:", chat_response)
935
+ ```
936
+
937
+ #### Video Input
938
+
939
+ ```python
940
+ from openai import OpenAI
941
+ # Configured by environment variables
942
+ client = OpenAI()
943
+
944
+ messages = [
945
+ {
946
+ "role": "user",
947
+ "content": [
948
+ {
949
+ "type": "video_url",
950
+ "video_url": {
951
+ "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/video/N1cdUjctpG8.mp4"
952
+ }
953
+ },
954
+ {
955
+ "type": "text",
956
+ "text": "How many porcelain jars were discovered in the niches located in the primary chamber of the tomb?"
957
+ }
958
+ ]
959
+ }
960
+ ]
961
+
962
+ # When vLLM is launched with `--media-io-kwargs '{"video": {"num_frames": -1}}'`,
963
+ # video frame sampling can be configured via `extra_body` (e.g., by setting `fps`).
964
+ # This feature is currently supported only in vLLM.
965
+ #
966
+ # By default, `fps=2` and `do_sample_frames=True`.
967
+ # With `do_sample_frames=True`, you can customize the `fps` value to set your desired video sampling rate.
968
+ response = client.chat.completions.create(
969
+ model="Qwen/Qwen3.6-27B",
970
+ messages=messages,
971
+ max_tokens=81920,
972
+ temperature=1.0,
973
+ top_p=0.95,
974
+ presence_penalty=0.0,
975
+ extra_body={
976
+ "top_k": 20,
977
+ "mm_processor_kwargs": {"fps": 2, "do_sample_frames": True},
978
+ },
979
+ )
980
+
981
+ print("Chat response:", chat_response)
982
+ ```
983
+
984
+
985
+ #### Instruct (or Non-Thinking) Mode
986
+
987
+ > [!Important]
988
+ > Qwen3.6 does not officially support the soft switch of Qwen3, i.e., `/think` and `/nothink`.
989
+
990
+ Qwen3.6 will think by default before response.
991
+ You can obtain direct response from the model without thinking by configuring the API parameters.
992
+ For example,
993
+ ```python
994
+ from openai import OpenAI
995
+ # Configured by environment variables
996
+ client = OpenAI()
997
+
998
+ messages = [
999
+ {
1000
+ "role": "user",
1001
+ "content": [
1002
+ {
1003
+ "type": "image_url",
1004
+ "image_url": {
1005
+ "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.6/demo/RealWorld/RealWorld-04.png"
1006
+ }
1007
+ },
1008
+ {
1009
+ "type": "text",
1010
+ "text": "Where is this?"
1011
+ }
1012
+ ]
1013
+ }
1014
+ ]
1015
+
1016
+ chat_response = client.chat.completions.create(
1017
+ model="Qwen/Qwen3.6-27B",
1018
+ messages=messages,
1019
+ max_tokens=32768,
1020
+ temperature=0.7,
1021
+ top_p=0.8,
1022
+ presence_penalty=1.5,
1023
+ extra_body={
1024
+ "top_k": 20,
1025
+ "chat_template_kwargs": {"enable_thinking": False},
1026
+ },
1027
+ )
1028
+ print("Chat response:", chat_response)
1029
+ ```
1030
+
1031
+ > [!Note]
1032
+ > If you are using APIs from Alibaba Cloud Model Studio, in addition to changing `model`, please use `"enable_thinking": False` instead of `"chat_template_kwargs": {"enable_thinking": False}`.
1033
+
1034
+ #### Preserve Thinking
1035
+
1036
+ By default, only the thinking blocks generated in handling the latest user message is retained, resulting in a pattern commonly as interleaved thinking.
1037
+ Qwen3.6 has been additionally trained to preserve and leverage thinking traces from historical messages.
1038
+ You can enable this behavior by setting the `preserve_thinking` option:
1039
+ ```python
1040
+ from openai import OpenAI
1041
+ # Configured by environment variables
1042
+ client = OpenAI()
1043
+
1044
+ messages = [...]
1045
+
1046
+ chat_response = client.chat.completions.create(
1047
+ model="Qwen/Qwen3.6-27B",
1048
+ messages=messages,
1049
+ max_tokens=32768,
1050
+ temperature=0.6,
1051
+ top_p=0.95,
1052
+ presence_penalty=0.0,
1053
+ extra_body={
1054
+ "top_k": 20,
1055
+ "chat_template_kwargs": {"preserve_thinking": True},
1056
+ },
1057
+ )
1058
+ print("Chat response:", chat_response)
1059
+ ```
1060
+
1061
+ > [!Note]
1062
+ > If you are using APIs from Alibaba Cloud Model Studio, in addition to changing `model`, please use `"preserve_thinking": True` instead of `"chat_template_kwargs": {"preserve_thinking": False}`.
1063
+
1064
+
1065
+ This capability is particularly beneficial for agent scenarios, where maintaining full reasoning context can enhance decision consistency and, in many cases, reduce overall token consumption by minimizing redundant reasoning. Additionally, it can improve KV cache utilization, optimizing inference efficiency in both thinking and non-thinking modes.
1066
+
1067
+
1068
+ ## Agentic Usage
1069
+
1070
+ Qwen3.6 excels in tool calling capabilities.
1071
+
1072
+ ### Qwen-Agent
1073
+
1074
+ We recommend using [Qwen-Agent](https://github.com/QwenLM/Qwen-Agent) to quickly build Agent applications with Qwen3.6.
1075
+
1076
+ To define the available tools, you can use the MCP configuration file, use the integrated tool of Qwen-Agent, or integrate other tools by yourself.
1077
+ ```python
1078
+ import os
1079
+ from qwen_agent.agents import Assistant
1080
+
1081
+ # Define LLM
1082
+ # Using Alibaba Cloud Model Studio
1083
+ llm_cfg = {
1084
+ # Use the OpenAI-compatible model service provided by DashScope:
1085
+ 'model': 'qwen3.6-27b',
1086
+ 'model_type': 'qwenvl_oai',
1087
+ 'model_server': 'https://dashscope.aliyuncs.com/compatible-mode/v1',
1088
+ 'api_key': os.getenv('DASHSCOPE_API_KEY'),
1089
+
1090
+ 'generate_cfg': {
1091
+ 'use_raw_api': True,
1092
+ # When using Dash Scope OAI API, pass the parameter of whether to enable thinking mode in this way
1093
+ 'extra_body': {
1094
+ 'enable_thinking': True,
1095
+ 'preserve_thinking': True,
1096
+ },
1097
+ },
1098
+ }
1099
+
1100
+ # Using OpenAI-compatible API endpoint.
1101
+ # functionality of the deployment frameworks and let Qwen-Agent automate the related operations.
1102
+ #
1103
+ # llm_cfg = {
1104
+ # # Use your own model service compatible with OpenAI API by vLLM/SGLang:
1105
+ # 'model': 'Qwen/Qwen3.6-27B',
1106
+ # 'model_type': 'qwenvl_oai',
1107
+ # 'model_server': 'http://localhost:8000/v1', # api_base
1108
+ # 'api_key': 'EMPTY',
1109
+ #
1110
+ # 'generate_cfg': {
1111
+ # 'use_raw_api': True,
1112
+ # # When using vLLM/SGLang OAI API, pass the parameter of whether to enable thinking mode in this way
1113
+ # 'extra_body': {
1114
+ # 'chat_template_kwargs': {'enable_thinking': True, 'preserve_thinking': True}
1115
+ # },
1116
+ # },
1117
+ # }
1118
+
1119
+ # Define Tools
1120
+ tools = [
1121
+ {'mcpServers': { # You can specify the MCP configuration file
1122
+ "filesystem": {
1123
+ "command": "npx",
1124
+ "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/xxxx/Desktop"]
1125
+ }
1126
+ }
1127
+ }
1128
+ ]
1129
+
1130
+ # Define Agent
1131
+ bot = Assistant(llm=llm_cfg, function_list=tools)
1132
+
1133
+ # Streaming generation
1134
+ messages = [{'role': 'user', 'content': 'Help me organize my desktop.'}]
1135
+ for responses in bot.run(messages=messages):
1136
+ pass
1137
+ print(responses)
1138
+
1139
+ # Streaming generation
1140
+ messages = [{'role': 'user', 'content': 'Develop a dog website and save it on the desktop'}]
1141
+ for responses in bot.run(messages=messages):
1142
+ pass
1143
+ print(responses)
1144
+ ```
1145
+
1146
+ ### Qwen Code
1147
+
1148
+
1149
+ [Qwen Code](https://github.com/QwenLM/qwen-code) is an open-source AI agent for the terminal, optimized for Qwen models. It helps you understand large codebases, automate tedious work, and ship faster.
1150
+
1151
+ For more information, please refer to [Qwen Code](https://qwenlm.github.io/qwen-code-docs/).
1152
+
1153
+ ## Processing Ultra-Long Texts
1154
+
1155
+ Qwen3.6 natively supports context lengths of up to 262,144 tokens.
1156
+ For long-horizon tasks where the total length (including both input and output) exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively., e.g., YaRN.
1157
+
1158
+ YaRN is currently supported by several inference frameworks, e.g., `transformers`, `vllm`, `ktransformers` and `sglang`.
1159
+ In general, there are two approaches to enabling YaRN for supported frameworks:
1160
+
1161
+ - Modifying the model configuration file:
1162
+ In the `config.json` file, change the `rope_parameters` fields in `text_config` to:
1163
+ ```json
1164
+ {
1165
+ "mrope_interleaved": true,
1166
+ "mrope_section": [
1167
+ 11,
1168
+ 11,
1169
+ 10
1170
+ ],
1171
+ "rope_type": "yarn",
1172
+ "rope_theta": 10000000,
1173
+ "partial_rotary_factor": 0.25,
1174
+ "factor": 4.0,
1175
+ "original_max_position_embeddings": 262144,
1176
+ }
1177
+ ```
1178
+
1179
+ - Passing command line arguments:
1180
+
1181
+ For `vllm`, you can use
1182
+ ```shell
1183
+ VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1010000
1184
+ ```
1185
+
1186
+ For `sglang` and `ktransformers`, you can use
1187
+ ```shell
1188
+ SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server ... --json-model-override-args '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --context-length 1010000
1189
+ ```
1190
+
1191
+ > [!NOTE]
1192
+ > All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, **potentially impacting performance on shorter texts.**
1193
+ > We advise modifying the `rope_parameters` configuration only when processing long contexts is required.
1194
+ > It is also recommended to modify the `factor` as needed. For example, if the typical context length for your application is 524,288 tokens, it would be better to set `factor` as 2.0.
1195
+
1196
+ ## Best Practices
1197
+
1198
+ To achieve optimal performance, we recommend the following settings:
1199
+
1200
+ 1. **Sampling Parameters**:
1201
+ - We suggest using the following sets of sampling parameters depending on the mode and task type:
1202
+ - **Thinking mode for general tasks**:
1203
+ `temperature=1.0`, `top_p=0.95`, `top_k=20`, `min_p=0.0`, `presence_penalty=0.0`, `repetition_penalty=1.0`
1204
+ - **Thinking mode for precise coding tasks (e.g., WebDev)**:
1205
+ `temperature=0.6`, `top_p=0.95`, `top_k=20`, `min_p=0.0`, `presence_penalty=0.0`, `repetition_penalty=1.0`
1206
+ - **Instruct (or non-thinking) mode**:
1207
+ `temperature=0.7`, `top_p=0.80`, `top_k=20`, `min_p=0.0`, `presence_penalty=1.5`, `repetition_penalty=1.0`
1208
+ - For supported frameworks, you can adjust the `presence_penalty` parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
1209
+
1210
+ 2. **Adequate Output Length**: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 81,920 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.
1211
+
1212
+ 3. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.
1213
+ - **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
1214
+ - **Multiple-Choice Questions**: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the `answer` field with only the choice letter, e.g., `"answer": "C"`."
1215
+
1216
+ 4. **Long Video Understanding**: To optimize inference efficiency for plain text and images, the `size` parameter in the released `video_preprocessor_config.json` is conservatively configured. It is recommended to set the `longest_edge` parameter in the video_preprocessor_config file to 469,762,048 (corresponding to 224k video tokens) to enable higher frame-rate sampling for hour-scale videos and thereby achieve superior performance. For example,
1217
+ ```json
1218
+ {"longest_edge": 469762048, "shortest_edge": 4096}
1219
+ ```
1220
+
1221
+ Alternatively, override the default values via engine startup parameters. For implementation details, refer to: [vLLM](https://github.com/vllm-project/vllm/pull/34330) / [SGLang](https://github.com/sgl-project/sglang/pull/18467).
1222
+
1223
+
1224
+ ### Citation
1225
+
1226
+ If you find our work helpful, feel free to give us a cite.
1227
+
1228
+ ```bibtex
1229
+ @misc{qwen3.6-27b,
1230
+ title = {{Qwen3.6-27B}: Flagship-Level Coding in a {27B} Dense Model},
1231
+ author = {{Qwen Team}},
1232
+ month = {April},
1233
+ year = {2026},
1234
+ url = {https://qwen.ai/blog?id=qwen3.6-27b}
1235
+ }
1236
+ ```
chat_template.jinja ADDED
@@ -0,0 +1,221 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id is defined and add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id is defined and add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- set ns_flags = namespace(enable_thinking=true) %}
43
+ {%- if enable_thinking is defined %}
44
+ {%- set ns_flags.enable_thinking = enable_thinking %}
45
+ {%- endif %}
46
+ {%- if not messages %}
47
+ {{- raise_exception('No messages provided.') }}
48
+ {%- endif %}
49
+ {%- if add_generation_prompt is defined and add_generation_prompt and continue_final_message is defined and continue_final_message %}
50
+ {{- raise_exception('add_generation_prompt and continue_final_message cannot both be true.') }}
51
+ {%- endif %}
52
+ {%- if tools and tools is iterable and tools is not mapping %}
53
+ {{- '<|im_start|>system\n' }}
54
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
55
+ {%- for tool in tools %}
56
+ {{- "\n" }}
57
+ {{- tool | tojson }}
58
+ {%- endfor %}
59
+ {{- "\n</tools>" }}
60
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
61
+ {%- if messages[0].role == 'system' or messages[0].role == 'developer' %}
62
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
63
+ {%- if '<|think_off|>' in content %}
64
+ {%- set ns_flags.enable_thinking = false %}
65
+ {%- set content = content.replace('<|think_off|>', '') %}
66
+ {%- endif %}
67
+ {%- if '<|think_on|>' in content %}
68
+ {%- set ns_flags.enable_thinking = true %}
69
+ {%- set content = content.replace('<|think_on|>', '') %}
70
+ {%- endif %}
71
+ {%- set content = content.strip() %}
72
+ {%- if content %}
73
+ {{- '\n\n' + content }}
74
+ {%- endif %}
75
+ {%- endif %}
76
+ {{- '<|im_end|>\n' }}
77
+ {%- else %}
78
+ {%- if messages[0].role == 'system' or messages[0].role == 'developer' %}
79
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
80
+ {%- if '<|think_off|>' in content %}
81
+ {%- set ns_flags.enable_thinking = false %}
82
+ {%- set content = content.replace('<|think_off|>', '') %}
83
+ {%- endif %}
84
+ {%- if '<|think_on|>' in content %}
85
+ {%- set ns_flags.enable_thinking = true %}
86
+ {%- set content = content.replace('<|think_on|>', '') %}
87
+ {%- endif %}
88
+ {%- set content = content.strip() %}
89
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
90
+ {%- endif %}
91
+ {%- endif %}
92
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
93
+ {%- for message in messages[::-1] %}
94
+ {%- set index = (messages|length - 1) - loop.index0 %}
95
+ {%- if ns.multi_step_tool and message.role == "user" %}
96
+ {%- set content = render_content(message.content, false)|trim %}
97
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
98
+ {%- set ns.multi_step_tool = false %}
99
+ {%- set ns.last_query_index = index %}
100
+ {%- endif %}
101
+ {%- endif %}
102
+ {%- endfor %}
103
+ {%- if ns.multi_step_tool %}
104
+ {%- set ns.last_query_index = messages|length - 1 %}
105
+ {%- endif %}
106
+ {%- for message in messages %}
107
+ {%- set content = render_content(message.content, true)|trim %}
108
+ {%- set content = content.replace('<|think_off|>', '').replace('<|think_on|>', '') %}
109
+ {%- set content = content.strip() %}
110
+ {%- if message.role == "system" or message.role == "developer" %}
111
+ {%- if not loop.first %}
112
+ {%- set sys_content = render_content(message.content, false, true)|trim %}
113
+ {%- set sys_content = sys_content.replace('<|think_off|>', '').replace('<|think_on|>', '')|trim %}
114
+ {{- '<|im_start|>system\n' + sys_content + '<|im_end|>' + '\n' }}
115
+ {%- endif %}
116
+ {%- elif message.role == "user" %}
117
+ {{- '<|im_start|>' + message.role + '\n' + (content if content else ' ') + '<|im_end|>' + '\n' }}
118
+ {%- elif message.role == "assistant" %}
119
+ {%- set reasoning_content = '' %}
120
+ {%- if message.reasoning_content is string %}
121
+ {%- set reasoning_content = message.reasoning_content %}
122
+ {%- else %}
123
+ {%- set has_think_tag = false %}
124
+ {%- set think_start_token = '<think>' %}
125
+ {%- set think_end_token = '</think>' %}
126
+ {%- if '</think>' in content %}
127
+ {%- set has_think_tag = true %}
128
+ {%- elif '</thinking>' in content %}
129
+ {%- set has_think_tag = true %}
130
+ {%- set think_start_token = '<thinking>' %}
131
+ {%- set think_end_token = '</thinking>' %}
132
+ {%- elif '<think>' in content %}
133
+ {%- set reasoning_content = content.split('<think>')[-1].lstrip('\n') %}
134
+ {%- set content = '' %}
135
+ {%- elif '<thinking>' in content %}
136
+ {%- set reasoning_content = content.split('<thinking>')[-1].lstrip('\n') %}
137
+ {%- set content = '' %}
138
+ {%- endif %}
139
+ {%- if has_think_tag %}
140
+ {%- set reasoning_content = content.split(think_end_token)[0].rstrip('\n').split(think_start_token)[-1].lstrip('\n') %}
141
+ {%- set content = content.split(think_end_token)[-1].lstrip('\n') %}
142
+ {%- endif %}
143
+ {%- endif %}
144
+ {%- set reasoning_content = reasoning_content|trim %}
145
+ {%- set show_think = false %}
146
+ {%- if loop.index0 > ns.last_query_index %}
147
+ {%- set show_think = true %}
148
+ {%- elif ns_flags.enable_thinking and (preserve_thinking is undefined or preserve_thinking is true) and reasoning_content|length > 0 %}
149
+ {%- set show_think = true %}
150
+ {%- endif %}
151
+ {%- if show_think %}
152
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
153
+ {%- else %}
154
+ {{- '<|im_start|>' + message.role + '\n' + content }}
155
+ {%- endif %}
156
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
157
+ {%- for tool_call in message.tool_calls %}
158
+ {%- if tool_call.function is defined %}
159
+ {%- set tool_call = tool_call.function %}
160
+ {%- endif %}
161
+ {%- if loop.first %}
162
+ {%- if content|trim %}
163
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
164
+ {%- else %}
165
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
166
+ {%- endif %}
167
+ {%- else %}
168
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
169
+ {%- endif %}
170
+ {%- if tool_call.arguments is defined and tool_call.arguments is mapping %}
171
+ {%- if tool_call.arguments|length > 0 %}
172
+ {%- for args_name in tool_call.arguments %}
173
+ {%- set args_value = tool_call.arguments[args_name] %}
174
+ {{- '<parameter=' + args_name + '>\n' }}
175
+ {%- set args_value = args_value | string if args_value is string else args_value | tojson %}
176
+ {{- args_value }}
177
+ {{- '\n</parameter>\n' }}
178
+ {%- endfor %}
179
+ {%- endif %}
180
+ {%- elif tool_call.arguments is defined and tool_call.arguments is string %}
181
+ {%- if tool_call.arguments|trim|length > 0 %}
182
+ {#- Note: raw JSON string arguments are emitted as-is and will not match
183
+ the XML parameter format in the tool instructions. Normalize arguments
184
+ to a dict in your serving layer before applying this template. -#}
185
+ {{- tool_call.arguments }}
186
+ {{- '\n' }}
187
+ {%- endif %}
188
+ {%- endif %}
189
+ {{- '</function>\n</tool_call>' }}
190
+ {%- endfor %}
191
+ {%- endif %}
192
+ {%- if not (loop.last and continue_final_message is defined and continue_final_message is true) %}
193
+ {{- '<|im_end|>\n' }}
194
+ {%- endif %}
195
+ {%- elif message.role == "tool" %}
196
+ {%- if not loop.previtem or (loop.previtem.role != "tool" and loop.previtem.role != "assistant") %}
197
+ {{- raise_exception('A tool message must follow an assistant or tool message.') }}
198
+ {%- endif %}
199
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
200
+ {{- '<|im_start|>user' }}
201
+ {%- endif %}
202
+ {{- '\n<tool_response>\n' }}
203
+ {{- content }}
204
+ {{- '\n</tool_response>' }}
205
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
206
+ {{- '<|im_end|>\n' }}
207
+ {%- elif loop.last %}
208
+ {{- '<|im_end|>\n' }}
209
+ {%- endif %}
210
+ {%- else %}
211
+ {{- raise_exception('Unexpected message role.') }}
212
+ {%- endif %}
213
+ {%- endfor %}
214
+ {%- if add_generation_prompt %}
215
+ {{- '<|im_start|>assistant\n' }}
216
+ {%- if ns_flags.enable_thinking is false %}
217
+ {{- '<think>\n\n</think>\n\n' }}
218
+ {%- else %}
219
+ {{- '<think>\n' }}
220
+ {%- endif %}
221
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,194 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3_5ForConditionalGeneration"
4
+ ],
5
+ "dtype": "bfloat16",
6
+ "image_token_id": 248056,
7
+ "language_model_only": false,
8
+ "model_type": "qwen3_5",
9
+ "quantization_config": {
10
+ "bits": 4,
11
+ "checkpoint_format": "gptq",
12
+ "desc_act": false,
13
+ "format": "gptq",
14
+ "group_size": 128,
15
+ "lm_head": false,
16
+ "meta": {
17
+ "act_group_aware": true,
18
+ "auto_forward_data_parallel": true,
19
+ "damp_auto_increment": 0.01,
20
+ "damp_percent": 0.05,
21
+ "dense_vram_strategy": "exclusive",
22
+ "dense_vram_strategy_devices": null,
23
+ "fallback": {
24
+ "smooth": null,
25
+ "strategy": "rtn",
26
+ "threshold": "0.5%"
27
+ },
28
+ "foem": null,
29
+ "gc_mode": "interval",
30
+ "gptaq": null,
31
+ "hessian": {
32
+ "chunk_bytes": null,
33
+ "chunk_size": null,
34
+ "staging_dtype": "float32"
35
+ },
36
+ "mock_quantization": false,
37
+ "moe_vram_strategy": "exclusive",
38
+ "moe_vram_strategy_devices": null,
39
+ "mse": 0.0,
40
+ "offload_to_disk": false,
41
+ "offload_to_disk_path": null,
42
+ "pack_impl": "cpu",
43
+ "quantizer": [
44
+ "gptqmodel:7.1.0-dev"
45
+ ],
46
+ "static_groups": false,
47
+ "true_sequential": true,
48
+ "uri": "https://github.com/modelcloud/gptqmodel",
49
+ "wait_for_submodule_finalizers": false
50
+ },
51
+ "method": "gptq",
52
+ "pack_dtype": "int32",
53
+ "quant_method": "gptq",
54
+ "sym": true
55
+ },
56
+ "rope_parameters": {
57
+ "rope_theta": 10000.0,
58
+ "rope_type": "default"
59
+ },
60
+ "text_config": {
61
+ "attention_bias": false,
62
+ "attention_dropout": 0.0,
63
+ "attn_output_gate": true,
64
+ "bos_token_id": 248044,
65
+ "dtype": "bfloat16",
66
+ "eos_token_id": 248044,
67
+ "full_attention_interval": 4,
68
+ "head_dim": 256,
69
+ "hidden_act": "silu",
70
+ "hidden_size": 5120,
71
+ "initializer_range": 0.02,
72
+ "intermediate_size": 17408,
73
+ "layer_types": [
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
+ "linear_attention",
91
+ "linear_attention",
92
+ "linear_attention",
93
+ "full_attention",
94
+ "linear_attention",
95
+ "linear_attention",
96
+ "linear_attention",
97
+ "full_attention",
98
+ "linear_attention",
99
+ "linear_attention",
100
+ "linear_attention",
101
+ "full_attention",
102
+ "linear_attention",
103
+ "linear_attention",
104
+ "linear_attention",
105
+ "full_attention",
106
+ "linear_attention",
107
+ "linear_attention",
108
+ "linear_attention",
109
+ "full_attention",
110
+ "linear_attention",
111
+ "linear_attention",
112
+ "linear_attention",
113
+ "full_attention",
114
+ "linear_attention",
115
+ "linear_attention",
116
+ "linear_attention",
117
+ "full_attention",
118
+ "linear_attention",
119
+ "linear_attention",
120
+ "linear_attention",
121
+ "full_attention",
122
+ "linear_attention",
123
+ "linear_attention",
124
+ "linear_attention",
125
+ "full_attention",
126
+ "linear_attention",
127
+ "linear_attention",
128
+ "linear_attention",
129
+ "full_attention",
130
+ "linear_attention",
131
+ "linear_attention",
132
+ "linear_attention",
133
+ "full_attention",
134
+ "linear_attention",
135
+ "linear_attention",
136
+ "linear_attention",
137
+ "full_attention"
138
+ ],
139
+ "linear_conv_kernel_dim": 4,
140
+ "linear_key_head_dim": 128,
141
+ "linear_num_key_heads": 16,
142
+ "linear_num_value_heads": 48,
143
+ "linear_value_head_dim": 128,
144
+ "mamba_ssm_dtype": "float32",
145
+ "max_position_embeddings": 262144,
146
+ "model_type": "qwen3_5_text",
147
+ "mtp_num_hidden_layers": 1,
148
+ "mtp_use_dedicated_embeddings": false,
149
+ "num_attention_heads": 24,
150
+ "num_hidden_layers": 64,
151
+ "num_key_value_heads": 4,
152
+ "output_gate_type": "swish",
153
+ "pad_token_id": 248044,
154
+ "partial_rotary_factor": 0.25,
155
+ "rms_norm_eps": 1e-06,
156
+ "rope_parameters": {
157
+ "mrope_interleaved": true,
158
+ "mrope_section": [
159
+ 11,
160
+ 11,
161
+ 10
162
+ ],
163
+ "partial_rotary_factor": 0.25,
164
+ "rope_theta": 10000000,
165
+ "rope_type": "default"
166
+ },
167
+ "tie_word_embeddings": false,
168
+ "use_cache": true,
169
+ "vocab_size": 248320
170
+ },
171
+ "tie_word_embeddings": false,
172
+ "transformers_version": "5.7.0",
173
+ "use_cache": false,
174
+ "video_token_id": 248057,
175
+ "vision_config": {
176
+ "deepstack_visual_indexes": [],
177
+ "depth": 27,
178
+ "dtype": "bfloat16",
179
+ "hidden_act": "gelu_pytorch_tanh",
180
+ "hidden_size": 1152,
181
+ "in_channels": 3,
182
+ "initializer_range": 0.02,
183
+ "intermediate_size": 4304,
184
+ "model_type": "qwen3_5_vision",
185
+ "num_heads": 16,
186
+ "num_position_embeddings": 2304,
187
+ "out_hidden_size": 5120,
188
+ "patch_size": 16,
189
+ "spatial_merge_size": 2,
190
+ "temporal_patch_size": 2
191
+ },
192
+ "vision_end_token_id": 248054,
193
+ "vision_start_token_id": 248053
194
+ }
generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 248044,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 248046,
6
+ 248044
7
+ ],
8
+ "pad_token_id": 248044,
9
+ "temperature": 1.0,
10
+ "top_k": 20,
11
+ "top_p": 0.95,
12
+ "transformers_version": "5.7.0"
13
+ }
model-00001-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:028ad551521a430a223ccbfecfcfd0444e9b6fa932dbbc8916f7b18dea0886f5
3
+ size 3516829256
model-00002-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1fcf6f42252e1d0a18c8320c9768a328b1a35bdf0aa86c7f060014a66b69a70d
3
+ size 4278206560
model-00003-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4c00665c4653885c549c6e7a98ee4207fdea7b12e6c43c4f74416f6d00b8dd89
3
+ size 4258595144
model-00004-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e6f6d03224e52ea86e79f9713fe42efdd74e111fc9abbdee2b610b617770459
3
+ size 4284981464
model-00005-of-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c84e04e49cd465c1633422cb3b38234c204d2e63dcbf1fde08562c878db1fdc5
3
+ size 2371438160
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
processor_config.json ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_processor": {
3
+ "do_convert_rgb": true,
4
+ "do_normalize": true,
5
+ "do_rescale": true,
6
+ "do_resize": true,
7
+ "image_mean": [
8
+ 0.5,
9
+ 0.5,
10
+ 0.5
11
+ ],
12
+ "image_processor_type": "Qwen2VLImageProcessor",
13
+ "image_std": [
14
+ 0.5,
15
+ 0.5,
16
+ 0.5
17
+ ],
18
+ "merge_size": 2,
19
+ "patch_size": 16,
20
+ "resample": 3,
21
+ "rescale_factor": 0.00392156862745098,
22
+ "size": {
23
+ "longest_edge": 16777216,
24
+ "shortest_edge": 65536
25
+ },
26
+ "temporal_patch_size": 2
27
+ },
28
+ "processor_class": "Qwen3VLProcessor",
29
+ "video_processor": {
30
+ "do_convert_rgb": true,
31
+ "do_normalize": true,
32
+ "do_rescale": true,
33
+ "do_resize": true,
34
+ "do_sample_frames": true,
35
+ "fps": 2,
36
+ "image_mean": [
37
+ 0.5,
38
+ 0.5,
39
+ 0.5
40
+ ],
41
+ "image_std": [
42
+ 0.5,
43
+ 0.5,
44
+ 0.5
45
+ ],
46
+ "max_frames": 768,
47
+ "merge_size": 2,
48
+ "min_frames": 4,
49
+ "patch_size": 16,
50
+ "resample": 3,
51
+ "rescale_factor": 0.00392156862745098,
52
+ "return_metadata": false,
53
+ "size": {
54
+ "longest_edge": 25165824,
55
+ "shortest_edge": 4096
56
+ },
57
+ "temporal_patch_size": 2,
58
+ "video_processor_type": "Qwen3VLVideoProcessor"
59
+ }
60
+ }
quant_log.csv ADDED
@@ -0,0 +1,401 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ layer,module,loss,samples,damp,time
2
+ 0,linear_attn.in_proj_qkv,0.0001124585,0.05000,1.369
3
+ 0,linear_attn.in_proj_z,0.0000735147,0.05000,1.191
4
+ 0,linear_attn.out_proj,0.0000000458,0.05000,1.428
5
+ 0,mlp.gate_proj,0.0000007344,0.05000,2.445
6
+ 0,mlp.up_proj,0.0000006729,0.05000,2.500
7
+ 0,mlp.down_proj,0.0000000073,0.05000,4.765
8
+ 1,linear_attn.in_proj_qkv,0.0000051857,0.05000,1.199
9
+ 1,linear_attn.in_proj_z,0.0000031970,0.05000,1.210
10
+ 1,linear_attn.out_proj,0.0000000167,0.05000,1.433
11
+ 1,mlp.gate_proj,0.0000015898,0.05000,2.528
12
+ 1,mlp.up_proj,0.0000015084,0.05000,2.543
13
+ 1,mlp.down_proj,0.0000000132,0.05000,4.929
14
+ 2,linear_attn.in_proj_qkv,0.0000079052,0.05000,1.302
15
+ 2,linear_attn.in_proj_z,0.0000047881,0.05000,1.278
16
+ 2,linear_attn.out_proj,0.0000000299,0.05000,1.442
17
+ 2,mlp.gate_proj,0.0000029350,0.05000,2.342
18
+ 2,mlp.up_proj,0.0000027572,0.05000,2.355
19
+ 2,mlp.down_proj,0.0000000538,0.05000,4.624
20
+ 3,self_attn.k_proj,0.0000057155,0.05000,3.372
21
+ 3,self_attn.q_proj,0.0000756238,0.05000,3.694
22
+ 3,self_attn.v_proj,0.0000054922,0.05000,3.707
23
+ 3,self_attn.o_proj,0.0000000235,0.05000,1.367
24
+ 3,mlp.gate_proj,0.0000036935,0.05000,2.377
25
+ 3,mlp.up_proj,0.0000035451,0.05000,2.400
26
+ 3,mlp.down_proj,0.0000000426,0.05000,4.655
27
+ 4,linear_attn.in_proj_qkv,0.0000117883,0.05000,1.201
28
+ 4,linear_attn.in_proj_z,0.0000075562,0.05000,1.291
29
+ 4,linear_attn.out_proj,0.0000000496,0.05000,1.401
30
+ 4,mlp.up_proj,0.0000048453,0.05000,2.258
31
+ 4,mlp.gate_proj,0.0000050513,0.05000,2.274
32
+ 4,mlp.down_proj,0.0000000585,0.05000,4.679
33
+ 5,linear_attn.in_proj_qkv,0.0000148360,0.05000,1.199
34
+ 5,linear_attn.in_proj_z,0.0000096042,0.05000,1.184
35
+ 5,linear_attn.out_proj,0.0000000584,0.05000,1.375
36
+ 5,mlp.gate_proj,0.0000064174,0.05000,2.241
37
+ 5,mlp.up_proj,0.0000060350,0.05000,2.241
38
+ 5,mlp.down_proj,0.0000000876,0.05000,4.673
39
+ 6,linear_attn.in_proj_qkv,0.0000215601,0.05000,1.341
40
+ 6,linear_attn.in_proj_z,0.0000128220,0.05000,1.136
41
+ 6,linear_attn.out_proj,0.0000000832,0.05000,1.369
42
+ 6,mlp.up_proj,0.0000079926,0.05000,2.254
43
+ 6,mlp.gate_proj,0.0000086342,0.05000,2.254
44
+ 6,mlp.down_proj,0.0000001483,0.05000,4.765
45
+ 7,self_attn.v_proj,0.0000050359,0.05000,3.457
46
+ 7,self_attn.q_proj,0.0000669005,0.05000,3.757
47
+ 7,self_attn.k_proj,0.0000056382,0.05000,3.788
48
+ 7,self_attn.o_proj,0.0000000580,0.05000,1.396
49
+ 7,mlp.up_proj,0.0000091083,0.05000,2.420
50
+ 7,mlp.gate_proj,0.0000097984,0.05000,2.424
51
+ 7,mlp.down_proj,0.0000001704,0.05000,4.695
52
+ 8,linear_attn.in_proj_qkv,0.0000238544,0.05000,1.249
53
+ 8,linear_attn.in_proj_z,0.0000149754,0.05000,1.211
54
+ 8,linear_attn.out_proj,0.0000001199,0.05000,1.382
55
+ 8,mlp.gate_proj,0.0000109753,0.05000,2.423
56
+ 8,mlp.up_proj,0.0000102813,0.05000,2.451
57
+ 8,mlp.down_proj,0.0000002006,0.05000,4.778
58
+ 9,linear_attn.in_proj_qkv,0.0000242076,0.05000,1.302
59
+ 9,linear_attn.in_proj_z,0.0000146433,0.05000,1.231
60
+ 9,linear_attn.out_proj,0.0000001254,0.05000,1.364
61
+ 9,mlp.gate_proj,0.0000119103,0.05000,2.403
62
+ 9,mlp.up_proj,0.0000112151,0.05000,2.430
63
+ 9,mlp.down_proj,0.0000002206,0.05000,4.728
64
+ 10,linear_attn.in_proj_qkv,0.0000252392,0.05000,1.407
65
+ 10,linear_attn.in_proj_z,0.0000148109,0.05000,1.161
66
+ 10,linear_attn.out_proj,0.0000001429,0.05000,1.446
67
+ 10,mlp.gate_proj,0.0000122284,0.05000,2.563
68
+ 10,mlp.up_proj,0.0000115203,0.05000,2.572
69
+ 10,mlp.down_proj,0.0000002357,0.05000,4.856
70
+ 11,self_attn.q_proj,0.0000610761,0.05000,3.384
71
+ 11,self_attn.k_proj,0.0000055117,0.05000,3.450
72
+ 11,self_attn.v_proj,0.0000047976,0.05000,3.593
73
+ 11,self_attn.o_proj,0.0000000999,0.05000,1.393
74
+ 11,mlp.up_proj,0.0000122271,0.05000,2.266
75
+ 11,mlp.gate_proj,0.0000128432,0.05000,2.276
76
+ 11,mlp.down_proj,0.0000002628,0.05000,4.761
77
+ 12,linear_attn.in_proj_qkv,0.0000278871,0.05000,1.260
78
+ 12,linear_attn.in_proj_z,0.0000163464,0.05000,1.301
79
+ 12,linear_attn.out_proj,0.0000002194,0.05000,1.407
80
+ 12,mlp.gate_proj,0.0000132430,0.05000,2.460
81
+ 12,mlp.up_proj,0.0000125775,0.05000,2.469
82
+ 12,mlp.down_proj,0.0000002945,0.05000,4.735
83
+ 13,linear_attn.in_proj_qkv,0.0000290394,0.05000,1.366
84
+ 13,linear_attn.in_proj_z,0.0000162143,0.05000,1.203
85
+ 13,linear_attn.out_proj,0.0000002337,0.05000,1.397
86
+ 13,mlp.up_proj,0.0000134144,0.05000,2.302
87
+ 13,mlp.gate_proj,0.0000140844,0.05000,2.307
88
+ 13,mlp.down_proj,0.0000003324,0.05000,4.695
89
+ 14,linear_attn.in_proj_qkv,0.0000319660,0.05000,1.291
90
+ 14,linear_attn.in_proj_z,0.0000182989,0.05000,1.219
91
+ 14,linear_attn.out_proj,0.0000002754,0.05000,1.376
92
+ 14,mlp.gate_proj,0.0000148546,0.05000,2.412
93
+ 14,mlp.up_proj,0.0000141748,0.05000,2.435
94
+ 14,mlp.down_proj,0.0000003675,0.05000,4.692
95
+ 15,self_attn.v_proj,0.0000045519,0.05000,3.499
96
+ 15,self_attn.q_proj,0.0000530139,0.05000,3.574
97
+ 15,self_attn.k_proj,0.0000048562,0.05000,3.807
98
+ 15,self_attn.o_proj,0.0000001643,0.05000,1.378
99
+ 15,mlp.gate_proj,0.0000168037,0.05000,2.309
100
+ 15,mlp.up_proj,0.0000161847,0.05000,2.333
101
+ 15,mlp.down_proj,0.0000004292,0.05000,4.652
102
+ 16,linear_attn.in_proj_qkv,0.0000341283,0.05000,1.364
103
+ 16,linear_attn.in_proj_z,0.0000192766,0.05000,1.269
104
+ 16,linear_attn.out_proj,0.0000003231,0.05000,1.362
105
+ 16,mlp.gate_proj,0.0000184563,0.05000,2.292
106
+ 16,mlp.up_proj,0.0000176501,0.05000,2.340
107
+ 16,mlp.down_proj,0.0000005006,0.05000,4.720
108
+ 17,linear_attn.in_proj_qkv,0.0000386339,0.05000,1.352
109
+ 17,linear_attn.in_proj_z,0.0000193993,0.05000,1.188
110
+ 17,linear_attn.out_proj,0.0000004226,0.05000,1.390
111
+ 17,mlp.gate_proj,0.0000210040,0.05000,2.420
112
+ 17,mlp.up_proj,0.0000200822,0.05000,2.439
113
+ 17,mlp.down_proj,0.0000006504,0.05000,4.841
114
+ 18,linear_attn.in_proj_qkv,0.0000415584,0.05000,1.203
115
+ 18,linear_attn.in_proj_z,0.0000226482,0.05000,1.195
116
+ 18,linear_attn.out_proj,0.0000004800,0.05000,1.374
117
+ 18,mlp.up_proj,0.0000238993,0.05000,2.257
118
+ 18,mlp.gate_proj,0.0000259059,0.05000,2.268
119
+ 18,mlp.down_proj,0.0000010430,0.05000,4.726
120
+ 19,self_attn.k_proj,0.0000064666,0.05000,3.490
121
+ 19,self_attn.v_proj,0.0000073474,0.05000,3.939
122
+ 19,self_attn.q_proj,0.0000663701,0.05000,4.029
123
+ 19,self_attn.o_proj,0.0000006422,0.05000,1.379
124
+ 19,mlp.gate_proj,0.0000311597,0.05000,2.389
125
+ 19,mlp.up_proj,0.0000293290,0.05000,2.414
126
+ 19,mlp.down_proj,0.0000013996,0.05000,4.705
127
+ 20,linear_attn.in_proj_qkv,0.0000734079,0.05000,1.191
128
+ 20,linear_attn.in_proj_z,0.0000405261,0.05000,1.316
129
+ 20,linear_attn.out_proj,0.0000007682,0.05000,1.433
130
+ 20,mlp.up_proj,0.0000348975,0.05000,2.315
131
+ 20,mlp.gate_proj,0.0000378113,0.05000,2.323
132
+ 20,mlp.down_proj,0.0000016595,0.05000,4.780
133
+ 21,linear_attn.in_proj_qkv,0.0000917496,0.05000,1.270
134
+ 21,linear_attn.in_proj_z,0.0000575525,0.05000,1.200
135
+ 21,linear_attn.out_proj,0.0000009235,0.05000,1.415
136
+ 21,mlp.up_proj,0.0000384960,0.05000,2.299
137
+ 21,mlp.gate_proj,0.0000414019,0.05000,2.303
138
+ 21,mlp.down_proj,0.0000017634,0.05000,4.689
139
+ 22,linear_attn.in_proj_qkv,0.0000793031,0.05000,1.332
140
+ 22,linear_attn.in_proj_z,0.0000478564,0.05000,1.219
141
+ 22,linear_attn.out_proj,0.0000007876,0.05000,1.396
142
+ 22,mlp.up_proj,0.0000432892,0.05000,2.377
143
+ 22,mlp.gate_proj,0.0000486733,0.05000,2.383
144
+ 22,mlp.down_proj,0.0000019806,0.05000,4.699
145
+ 23,self_attn.v_proj,0.0000093535,0.05000,3.424
146
+ 23,self_attn.k_proj,0.0000092182,0.05000,3.684
147
+ 23,self_attn.q_proj,0.0000858245,0.05000,3.697
148
+ 23,self_attn.o_proj,0.0000007971,0.05000,1.378
149
+ 23,mlp.gate_proj,0.0000496822,0.05000,2.304
150
+ 23,mlp.up_proj,0.0000456048,0.05000,2.311
151
+ 23,mlp.down_proj,0.0000020382,0.05000,4.784
152
+ 24,linear_attn.in_proj_qkv,0.0000772304,0.05000,1.342
153
+ 24,linear_attn.in_proj_z,0.0000429458,0.05000,1.166
154
+ 24,linear_attn.out_proj,0.0000010391,0.05000,1.391
155
+ 24,mlp.gate_proj,0.0000537335,0.05000,2.338
156
+ 24,mlp.up_proj,0.0000491605,0.05000,2.362
157
+ 24,mlp.down_proj,0.0000023205,0.05000,4.669
158
+ 25,linear_attn.in_proj_qkv,0.0000778156,0.05000,1.265
159
+ 25,linear_attn.in_proj_z,0.0000403616,0.05000,1.231
160
+ 25,linear_attn.out_proj,0.0000010801,0.05000,1.469
161
+ 25,mlp.gate_proj,0.0000513127,0.05000,2.357
162
+ 25,mlp.up_proj,0.0000496487,0.05000,2.421
163
+ 25,mlp.down_proj,0.0000022580,0.05000,4.664
164
+ 26,linear_attn.in_proj_qkv,0.0000708801,0.05000,1.363
165
+ 26,linear_attn.in_proj_z,0.0000351736,0.05000,1.192
166
+ 26,linear_attn.out_proj,0.0000011845,0.05000,1.398
167
+ 26,mlp.gate_proj,0.0000475717,0.05000,2.395
168
+ 26,mlp.up_proj,0.0000487775,0.05000,2.394
169
+ 26,mlp.down_proj,0.0000023169,0.05000,4.663
170
+ 27,self_attn.k_proj,0.0000084618,0.05000,3.420
171
+ 27,self_attn.v_proj,0.0000073161,0.05000,3.630
172
+ 27,self_attn.q_proj,0.0000777619,0.05000,3.640
173
+ 27,self_attn.o_proj,0.0000015292,0.05000,1.407
174
+ 27,mlp.gate_proj,0.0000475408,0.05000,2.253
175
+ 27,mlp.up_proj,0.0000502094,0.05000,2.269
176
+ 27,mlp.down_proj,0.0000023378,0.05000,4.663
177
+ 28,linear_attn.in_proj_qkv,0.0000672106,0.05000,1.263
178
+ 28,linear_attn.in_proj_z,0.0000312117,0.05000,1.218
179
+ 28,linear_attn.out_proj,0.0000014488,0.05000,1.370
180
+ 28,mlp.gate_proj,0.0000449930,0.05000,2.256
181
+ 28,mlp.up_proj,0.0000483526,0.05000,2.256
182
+ 28,mlp.down_proj,0.0000023035,0.05000,4.652
183
+ 29,linear_attn.in_proj_qkv,0.0000708439,0.05000,1.329
184
+ 29,linear_attn.in_proj_z,0.0000324049,0.05000,1.171
185
+ 29,linear_attn.out_proj,0.0000012436,0.05000,1.387
186
+ 29,mlp.gate_proj,0.0000442164,0.05000,2.279
187
+ 29,mlp.up_proj,0.0000478293,0.05000,2.288
188
+ 29,mlp.down_proj,0.0000022463,0.05000,4.685
189
+ 30,linear_attn.in_proj_qkv,0.0000709262,0.05000,1.307
190
+ 30,linear_attn.in_proj_z,0.0000330901,0.05000,1.140
191
+ 30,linear_attn.out_proj,0.0000016106,0.05000,1.370
192
+ 30,mlp.gate_proj,0.0000431725,0.05000,2.288
193
+ 30,mlp.up_proj,0.0000478322,0.05000,2.310
194
+ 30,mlp.down_proj,0.0000022741,0.05000,4.642
195
+ 31,self_attn.v_proj,0.0000086143,0.05000,3.363
196
+ 31,self_attn.q_proj,0.0000707655,0.05000,3.403
197
+ 31,self_attn.k_proj,0.0000078561,0.05000,3.523
198
+ 31,self_attn.o_proj,0.0000018328,0.05000,1.365
199
+ 31,mlp.gate_proj,0.0000457723,0.05000,2.302
200
+ 31,mlp.up_proj,0.0000515885,0.05000,2.354
201
+ 31,mlp.down_proj,0.0000024259,0.05000,4.656
202
+ 32,linear_attn.in_proj_qkv,0.0000750446,0.05000,1.256
203
+ 32,linear_attn.in_proj_z,0.0000326144,0.05000,1.287
204
+ 32,linear_attn.out_proj,0.0000016018,0.05000,1.374
205
+ 32,mlp.up_proj,0.0000546413,0.05000,2.324
206
+ 32,mlp.gate_proj,0.0000484193,0.05000,2.328
207
+ 32,mlp.down_proj,0.0000025691,0.05000,4.651
208
+ 33,linear_attn.in_proj_qkv,0.0000882211,0.05000,1.445
209
+ 33,linear_attn.in_proj_z,0.0000362461,0.05000,1.159
210
+ 33,linear_attn.out_proj,0.0000018170,0.05000,1.382
211
+ 33,mlp.up_proj,0.0000568260,0.05000,2.312
212
+ 33,mlp.gate_proj,0.0000503239,0.05000,2.318
213
+ 33,mlp.down_proj,0.0000026316,0.05000,4.695
214
+ 34,linear_attn.in_proj_qkv,0.0000855749,0.05000,1.269
215
+ 34,linear_attn.in_proj_z,0.0000388334,0.05000,1.280
216
+ 34,linear_attn.out_proj,0.0000018901,0.05000,1.376
217
+ 34,mlp.up_proj,0.0000608969,0.05000,2.323
218
+ 34,mlp.gate_proj,0.0000559209,0.05000,2.345
219
+ 34,mlp.down_proj,0.0000035411,0.05000,4.660
220
+ 35,self_attn.k_proj,0.0000096211,0.05000,3.485
221
+ 35,self_attn.v_proj,0.0000134193,0.05000,3.688
222
+ 35,self_attn.q_proj,0.0000877188,0.05000,3.895
223
+ 35,self_attn.o_proj,0.0000029999,0.05000,1.410
224
+ 35,mlp.gate_proj,0.0000583476,0.05000,2.547
225
+ 35,mlp.up_proj,0.0000624397,0.05000,2.590
226
+ 35,mlp.down_proj,0.0000042472,0.05000,4.804
227
+ 36,linear_attn.in_proj_qkv,0.0001224222,0.05000,1.366
228
+ 36,linear_attn.in_proj_z,0.0000610235,0.05000,1.220
229
+ 36,linear_attn.out_proj,0.0000018739,0.05000,1.438
230
+ 36,mlp.gate_proj,0.0000661158,0.05000,2.499
231
+ 36,mlp.up_proj,0.0000630097,0.05000,2.509
232
+ 36,mlp.down_proj,0.0000040550,0.05000,4.936
233
+ 37,linear_attn.in_proj_qkv,0.0001350269,0.05000,1.324
234
+ 37,linear_attn.in_proj_z,0.0000763220,0.05000,1.278
235
+ 37,linear_attn.out_proj,0.0000019218,0.05000,1.363
236
+ 37,mlp.up_proj,0.0000636777,0.05000,2.427
237
+ 37,mlp.gate_proj,0.0000674593,0.05000,2.442
238
+ 37,mlp.down_proj,0.0000037380,0.05000,4.749
239
+ 38,linear_attn.in_proj_qkv,0.0001201361,0.05000,1.434
240
+ 38,linear_attn.in_proj_z,0.0000698241,0.05000,1.207
241
+ 38,linear_attn.out_proj,0.0000013292,0.05000,1.407
242
+ 38,mlp.gate_proj,0.0000896982,0.05000,2.680
243
+ 38,mlp.up_proj,0.0000720469,0.05000,2.698
244
+ 38,mlp.down_proj,0.0000038076,0.05000,4.846
245
+ 39,self_attn.v_proj,0.0000141818,0.05000,3.635
246
+ 39,self_attn.k_proj,0.0000119403,0.05000,3.995
247
+ 39,self_attn.q_proj,0.0001056009,0.05000,4.007
248
+ 39,self_attn.o_proj,0.0000018453,0.05000,1.415
249
+ 39,mlp.gate_proj,0.0000797957,0.05000,2.222
250
+ 39,mlp.up_proj,0.0000692360,0.05000,2.226
251
+ 39,mlp.down_proj,0.0000034702,0.05000,4.638
252
+ 40,linear_attn.in_proj_qkv,0.0001151452,0.05000,1.339
253
+ 40,linear_attn.in_proj_z,0.0000635157,0.05000,1.218
254
+ 40,linear_attn.out_proj,0.0000016607,0.05000,1.403
255
+ 40,mlp.gate_proj,0.0000860928,0.05000,2.522
256
+ 40,mlp.up_proj,0.0000728400,0.05000,2.543
257
+ 40,mlp.down_proj,0.0000036189,0.05000,4.690
258
+ 41,linear_attn.in_proj_qkv,0.0001032554,0.05000,1.282
259
+ 41,linear_attn.in_proj_z,0.0000565361,0.05000,1.241
260
+ 41,linear_attn.out_proj,0.0000014398,0.05000,1.415
261
+ 41,mlp.up_proj,0.0000698027,0.05000,2.881
262
+ 41,mlp.gate_proj,0.0000737049,0.05000,2.900
263
+ 41,mlp.down_proj,0.0000034730,0.05000,4.659
264
+ 42,linear_attn.in_proj_qkv,0.0000963521,0.05000,1.276
265
+ 42,linear_attn.in_proj_z,0.0000505140,0.05000,1.166
266
+ 42,linear_attn.out_proj,0.0000017249,0.05000,1.364
267
+ 42,mlp.up_proj,0.0000682909,0.05000,2.863
268
+ 42,mlp.gate_proj,0.0000668445,0.05000,2.868
269
+ 42,mlp.down_proj,0.0000036960,0.05000,4.754
270
+ 43,self_attn.k_proj,0.0000116194,0.05000,3.609
271
+ 43,self_attn.q_proj,0.0000980658,0.05000,3.772
272
+ 43,self_attn.v_proj,0.0000133358,0.05000,3.839
273
+ 43,self_attn.o_proj,0.0000026923,0.05000,1.392
274
+ 43,mlp.up_proj,0.0000684222,0.05000,2.338
275
+ 43,mlp.gate_proj,0.0000641078,0.05000,2.341
276
+ 43,mlp.down_proj,0.0000038280,0.05000,4.833
277
+ 44,linear_attn.in_proj_qkv,0.0000890969,0.05000,1.218
278
+ 44,linear_attn.in_proj_z,0.0000435154,0.05000,1.261
279
+ 44,linear_attn.out_proj,0.0000024255,0.05000,1.378
280
+ 44,mlp.up_proj,0.0000686872,0.05000,2.306
281
+ 44,mlp.gate_proj,0.0000635907,0.05000,2.306
282
+ 44,mlp.down_proj,0.0000042407,0.05000,4.631
283
+ 45,linear_attn.in_proj_qkv,0.0000886524,0.05000,1.392
284
+ 45,linear_attn.in_proj_z,0.0000432914,0.05000,1.187
285
+ 45,linear_attn.out_proj,0.0000019384,0.05000,1.390
286
+ 45,mlp.gate_proj,0.0000630996,0.05000,2.223
287
+ 45,mlp.up_proj,0.0000686800,0.05000,2.223
288
+ 45,mlp.down_proj,0.0000042392,0.05000,4.680
289
+ 46,linear_attn.in_proj_qkv,0.0000929781,0.05000,1.296
290
+ 46,linear_attn.in_proj_z,0.0000459551,0.05000,1.209
291
+ 46,linear_attn.out_proj,0.0000030155,0.05000,1.378
292
+ 46,mlp.gate_proj,0.0000619233,0.05000,2.401
293
+ 46,mlp.up_proj,0.0000694591,0.05000,2.467
294
+ 46,mlp.down_proj,0.0000045538,0.05000,4.727
295
+ 47,self_attn.v_proj,0.0000162491,0.05000,3.410
296
+ 47,self_attn.q_proj,0.0000999248,0.05000,3.448
297
+ 47,self_attn.k_proj,0.0000107605,0.05000,3.652
298
+ 47,self_attn.o_proj,0.0000027216,0.05000,1.383
299
+ 47,mlp.gate_proj,0.0000676714,0.05000,2.470
300
+ 47,mlp.up_proj,0.0000753666,0.05000,2.554
301
+ 47,mlp.down_proj,0.0000052757,0.05000,4.737
302
+ 48,linear_attn.in_proj_qkv,0.0001006849,0.05000,1.258
303
+ 48,linear_attn.in_proj_z,0.0000478416,0.05000,1.311
304
+ 48,linear_attn.out_proj,0.0000031839,0.05000,1.418
305
+ 48,mlp.gate_proj,0.0000734429,0.05000,2.322
306
+ 48,mlp.up_proj,0.0000812822,0.05000,2.347
307
+ 48,mlp.down_proj,0.0000064582,0.05000,4.709
308
+ 49,linear_attn.in_proj_qkv,0.0001202173,0.05000,1.351
309
+ 49,linear_attn.in_proj_z,0.0000503558,0.05000,1.187
310
+ 49,linear_attn.out_proj,0.0000040884,0.05000,1.394
311
+ 49,mlp.gate_proj,0.0000807752,0.05000,2.374
312
+ 49,mlp.up_proj,0.0000870323,0.05000,2.385
313
+ 49,mlp.down_proj,0.0000083465,0.05000,4.705
314
+ 50,linear_attn.in_proj_qkv,0.0001237247,0.05000,1.274
315
+ 50,linear_attn.in_proj_z,0.0000565736,0.05000,1.183
316
+ 50,linear_attn.out_proj,0.0000059434,0.05000,1.387
317
+ 50,mlp.up_proj,0.0000996032,0.05000,2.315
318
+ 50,mlp.gate_proj,0.0000981337,0.05000,2.343
319
+ 50,mlp.down_proj,0.0000144299,0.05000,4.786
320
+ 51,self_attn.k_proj,0.0000140152,0.05000,3.275
321
+ 51,self_attn.v_proj,0.0000325511,0.05000,3.557
322
+ 51,self_attn.q_proj,0.0001265508,0.05000,3.704
323
+ 51,self_attn.o_proj,0.0000075998,0.05000,1.407
324
+ 51,mlp.gate_proj,0.0001048252,0.05000,2.279
325
+ 51,mlp.up_proj,0.0001101668,0.05000,2.298
326
+ 51,mlp.down_proj,0.0000185332,0.05000,4.741
327
+ 52,linear_attn.in_proj_qkv,0.0001891212,0.05000,1.292
328
+ 52,linear_attn.in_proj_z,0.0000844733,0.05000,1.359
329
+ 52,linear_attn.out_proj,0.0000079041,0.05000,1.442
330
+ 52,mlp.up_proj,0.0001203507,0.05000,2.408
331
+ 52,mlp.gate_proj,0.0001297232,0.05000,2.413
332
+ 52,mlp.down_proj,0.0000190583,0.05000,4.833
333
+ 53,linear_attn.in_proj_qkv,0.0001876515,0.05000,1.242
334
+ 53,linear_attn.in_proj_z,0.0000906394,0.05000,1.260
335
+ 53,linear_attn.out_proj,0.0000087621,0.05000,1.464
336
+ 53,mlp.gate_proj,0.0001458285,0.05000,2.460
337
+ 53,mlp.up_proj,0.0001293711,0.05000,2.462
338
+ 53,mlp.down_proj,0.0000210388,0.05000,4.869
339
+ 54,linear_attn.in_proj_qkv,0.0001827447,0.05000,1.235
340
+ 54,linear_attn.in_proj_z,0.0000971079,0.05000,1.239
341
+ 54,linear_attn.out_proj,0.0000083235,0.05000,1.363
342
+ 54,mlp.gate_proj,0.0001865210,0.05000,2.565
343
+ 54,mlp.up_proj,0.0001556885,0.05000,2.572
344
+ 54,mlp.down_proj,0.0000314462,0.05000,4.856
345
+ 55,self_attn.v_proj,0.0000439367,0.05000,3.638
346
+ 55,self_attn.q_proj,0.0001741279,0.05000,3.985
347
+ 55,self_attn.k_proj,0.0000213388,0.05000,3.993
348
+ 55,self_attn.o_proj,0.0000090859,0.05000,1.436
349
+ 55,mlp.up_proj,0.0001559785,0.05000,2.319
350
+ 55,mlp.gate_proj,0.0001770042,0.05000,2.325
351
+ 55,mlp.down_proj,0.0000267854,0.05000,4.691
352
+ 56,linear_attn.in_proj_qkv,0.0002152424,0.05000,1.322
353
+ 56,linear_attn.in_proj_z,0.0001060118,0.05000,1.194
354
+ 56,linear_attn.out_proj,0.0000146725,0.05000,1.399
355
+ 56,mlp.gate_proj,0.0001977834,0.05000,2.335
356
+ 56,mlp.up_proj,0.0001702014,0.05000,2.352
357
+ 56,mlp.down_proj,0.0000282588,0.05000,4.758
358
+ 57,linear_attn.in_proj_qkv,0.0002348663,0.05000,1.276
359
+ 57,linear_attn.in_proj_z,0.0001137585,0.05000,1.241
360
+ 57,linear_attn.out_proj,0.0000124321,0.05000,1.379
361
+ 57,mlp.gate_proj,0.0002321251,0.05000,2.608
362
+ 57,mlp.up_proj,0.0001968419,0.05000,2.609
363
+ 57,mlp.down_proj,0.0000292723,0.05000,4.739
364
+ 58,linear_attn.in_proj_qkv,0.0002240436,0.05000,1.338
365
+ 58,linear_attn.in_proj_z,0.0001192428,0.05000,1.200
366
+ 58,linear_attn.out_proj,0.0000115405,0.05000,1.427
367
+ 58,mlp.up_proj,0.0002276784,0.05000,2.465
368
+ 58,mlp.gate_proj,0.0002692953,0.05000,2.472
369
+ 58,mlp.down_proj,0.0000364566,0.05000,4.839
370
+ 59,self_attn.k_proj,0.0000280561,0.05000,3.574
371
+ 59,self_attn.q_proj,0.0002173350,0.05000,3.767
372
+ 59,self_attn.v_proj,0.0000956889,0.05000,3.820
373
+ 59,self_attn.o_proj,0.0000275526,0.05000,1.467
374
+ 59,mlp.gate_proj,0.0002666235,0.05000,2.340
375
+ 59,mlp.up_proj,0.0002308083,0.05000,2.348
376
+ 59,mlp.down_proj,0.0000436368,0.05000,4.676
377
+ 60,linear_attn.in_proj_qkv,0.0002759045,0.05000,1.294
378
+ 60,linear_attn.in_proj_z,0.0001288455,0.05000,1.231
379
+ 60,linear_attn.out_proj,0.0000279598,0.05000,1.386
380
+ 60,mlp.up_proj,0.0002357491,0.05000,2.332
381
+ 60,mlp.gate_proj,0.0002701066,0.05000,2.332
382
+ 60,mlp.down_proj,0.0000516546,0.05000,4.627
383
+ 61,linear_attn.in_proj_qkv,0.0002005166,0.05000,1.246
384
+ 61,linear_attn.in_proj_z,0.0001136661,0.05000,1.265
385
+ 61,linear_attn.out_proj,0.0000253483,0.05000,1.371
386
+ 61,mlp.up_proj,0.0002504054,0.05000,2.290
387
+ 61,mlp.gate_proj,0.0002867618,0.05000,2.293
388
+ 61,mlp.down_proj,0.0000634011,0.05000,4.636
389
+ 62,linear_attn.in_proj_qkv,0.0002241477,0.05000,1.240
390
+ 62,linear_attn.in_proj_z,0.0001171058,0.05000,1.193
391
+ 62,linear_attn.out_proj,0.0000583888,0.05000,1.393
392
+ 62,mlp.gate_proj,0.0002625598,0.05000,2.323
393
+ 62,mlp.up_proj,0.0002305720,0.05000,2.342
394
+ 62,mlp.down_proj,0.0000942124,0.05000,4.698
395
+ 63,self_attn.q_proj,0.0001960612,0.05000,3.448
396
+ 63,self_attn.v_proj,0.0000792475,0.05000,3.448
397
+ 63,self_attn.k_proj,0.0000266499,0.05000,3.617
398
+ 63,self_attn.o_proj,0.0000696076,0.05000,1.386
399
+ 63,mlp.gate_proj,0.0002238317,0.05000,2.283
400
+ 63,mlp.up_proj,0.0001896259,0.05000,2.298
401
+ 63,mlp.down_proj,0.0001958388,0.05000,4.663
quantize_config.json ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bits": 4,
3
+ "group_size": 128,
4
+ "desc_act": false,
5
+ "lm_head": false,
6
+ "method": "gptq",
7
+ "quant_method": "gptq",
8
+ "format": "gptq",
9
+ "checkpoint_format": "gptq",
10
+ "pack_dtype": "int32",
11
+ "meta": {
12
+ "quantizer": [
13
+ "gptqmodel:7.1.0-dev"
14
+ ],
15
+ "uri": "https://github.com/modelcloud/gptqmodel",
16
+ "damp_percent": 0.05,
17
+ "damp_auto_increment": 0.01,
18
+ "static_groups": false,
19
+ "true_sequential": true,
20
+ "mse": 0.0,
21
+ "gptaq": null,
22
+ "foem": null,
23
+ "act_group_aware": true,
24
+ "fallback": {
25
+ "strategy": "rtn",
26
+ "threshold": "0.5%",
27
+ "smooth": null
28
+ },
29
+ "offload_to_disk": false,
30
+ "offload_to_disk_path": null,
31
+ "pack_impl": "cpu",
32
+ "gc_mode": "interval",
33
+ "wait_for_submodule_finalizers": false,
34
+ "auto_forward_data_parallel": true,
35
+ "dense_vram_strategy": "exclusive",
36
+ "dense_vram_strategy_devices": null,
37
+ "moe_vram_strategy": "exclusive",
38
+ "moe_vram_strategy_devices": null,
39
+ "mock_quantization": false,
40
+ "hessian": {
41
+ "chunk_size": null,
42
+ "chunk_bytes": null,
43
+ "staging_dtype": "float32"
44
+ }
45
+ },
46
+ "sym": true
47
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523
3
+ size 19989325
tokenizer_config.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "local_files_only": false,
14
+ "max_length": null,
15
+ "model_max_length": 262144,
16
+ "model_specific_special_tokens": {
17
+ "audio_bos_token": "<|audio_start|>",
18
+ "audio_eos_token": "<|audio_end|>",
19
+ "audio_token": "<|audio_pad|>",
20
+ "image_token": "<|image_pad|>",
21
+ "video_token": "<|video_pad|>",
22
+ "vision_bos_token": "<|vision_start|>",
23
+ "vision_eos_token": "<|vision_end|>"
24
+ },
25
+ "pad_to_multiple_of": null,
26
+ "pad_token": "<|endoftext|>",
27
+ "pad_token_type_id": 0,
28
+ "padding_side": "left",
29
+ "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+",
30
+ "split_special_tokens": false,
31
+ "tokenizer_class": "Qwen2TokenizerFast",
32
+ "unk_token": null,
33
+ "video_token": "<|video_pad|>",
34
+ "vision_bos_token": "<|vision_start|>",
35
+ "vision_eos_token": "<|vision_end|>",
36
+ "_commit_hash": null
37
+ }