Instructions to use darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic")
model = AutoModelForMultimodalLM.from_pretrained("darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic

SGLang

How to use darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Unsloth Studio

How to use darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic",
    max_seq_length=2048,
)

Docker Model Runner
How to use darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic with Docker Model Runner:
```
docker model run hf.co/darkc0de/XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic
```

XORTRON.CriminalComputing.Qwen3.5-27B-Claude-Opus-4.6-Distill-darkc0de-heretic

File size: 28,863 Bytes

---
base_model: unsloth/Qwen3.5-27B
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3.5
- heretic
- uncensored
- decensored
- abliterated
license: apache-2.0
datasets:
- crownelius/Opus-4.6-Reasoning-2100x-formatted
---
# This is a decensored version of [TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill](https://huggingface.co/TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill), made using [Heretic](https://github.com/p-e-w/heretic) v1.2.0

## Abliteration parameters

| Parameter | Value |
| :-------- | :---: |
| **direction_index** | 54.87 |
| **attn.o_proj.max_weight** | 1.15 |
| **attn.o_proj.max_weight_position** | 42.46 |
| **attn.o_proj.min_weight** | 1.12 |
| **attn.o_proj.min_weight_distance** | 36.42 |
| **mlp.down_proj.max_weight** | 1.49 |
| **mlp.down_proj.max_weight_position** | 45.06 |
| **mlp.down_proj.min_weight** | 1.42 |
| **mlp.down_proj.min_weight_distance** | 30.83 |

## Performance

| Metric | This model | Original model ([TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill](https://huggingface.co/TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill)) |
| :----- | :--------: | :---------------------------: |
| **KL divergence** | 0.0238 | 0 *(by definition)* |
| **Refusals** | 3/100 | 92/100 |

-----


# Qwen3.5 27B x Claude Opus 4.6

Big thanks to [@EclipseMist](https://huggingface.co/EclipseMist) for providing [the LoRAs](https://huggingface.co/EclipseMist/Qwen3.5-27b-Opus-4.6-Distill-LoRa) for this model

- 🧬 Datasets:
  - `crownelius/Opus-4.6-Reasoning-2100x-formatted`
  - Personal Claude Data provided by [@EclipseMist](https://huggingface.co/EclipseMist)

- 🏗 Base Model:
  - `unsloth/Qwen3.5-27B`
 
- &#9889; Use cases:
  - Coding
  - Creative Writing
  - Visual Understanding
  - General Purpose

## Citations and Contributions

- [@EclipseMist](https://huggingface.co/EclipseMist) - Training and Data Curation
- [@crownelius](https://huggingface.co/crownelius) - Data Curation
- [@unsloth](https://huggingface.co/unsloth) - This qwen3 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
- [@Qwen](https://huggingface.co/Qwen) - Providing a fantastic, native-multimodal base model

---

# Benchmarks

![alt="Benchmark score Chart"](benchmarks/all.png)

![alt="Benchmark comparison Chart"](benchmarks/all_alt.png)

| Benchmark             | TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill   | unsloth/Qwen3.5-27B   |
|:----------------------|:----------------------------------------------|:----------------------|
| arc_challenge         | **0.461**                                     | 0.435                 |
| gpqa_diamond_zeroshot | **0.283**                                     | **0.283**             |
| hellaswag             | **0.613**                                     | 0.574                 |
| mmlu                  | **0.233**                                     | 0.230                 |
| truthfulqa_mc2        | **0.610**                                     | 0.599                 |
| winogrande            | **0.769**                                     | 0.749                 |

<details>
<summary>Table</summary>

| Model                                       | Benchmark             |    Score |   Total Questions |   Total Correct |
|:--------------------------------------------|:----------------------|---------:|------------------:|----------------:|
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | arc_challenge         | 0.460751 |              1172 |             540 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | gpqa_diamond_zeroshot | 0.282828 |               198 |              56 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | hellaswag             | 0.612926 |             10042 |            6155 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | mmlu                  | 0.232944 |             14042 |            3271 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | truthfulqa_mc2        | 0.610146 |               817 |             498 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | winogrande            | 0.768745 |              1267 |             974 |
| unsloth/Qwen3.5-27B                         | arc_challenge         | 0.435154 |              1172 |             510 |
| unsloth/Qwen3.5-27B                         | gpqa_diamond_zeroshot | 0.282828 |               198 |              56 |
| unsloth/Qwen3.5-27B                         | hellaswag             | 0.574288 |             10042 |            5767 |
| unsloth/Qwen3.5-27B                         | mmlu                  | 0.229597 |             14042 |            3224 |
| unsloth/Qwen3.5-27B                         | truthfulqa_mc2        | 0.599243 |               817 |             489 |
| unsloth/Qwen3.5-27B                         | winogrande            | 0.749013 |              1267 |             949 |

</details>


## MMLU Subject Breakdown

![alt="MMLU Subject Breakdown"](benchmarks/mmlu.png)

<details>
<summary>Table</summary>
  
| Model                                       | Subject                             | Benchmark                                |    Score |   Total Questions |   Total Correct |
|:--------------------------------------------|:------------------------------------|:-----------------------------------------|---------:|------------------:|----------------:|
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | formal_logic                        | mmlu_formal_logic                        | 0.285714 |               126 |              36 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_european_history        | mmlu_high_school_european_history        | 0.224242 |               165 |              37 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_us_history              | mmlu_high_school_us_history              | 0.240196 |               204 |              49 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_world_history           | mmlu_high_school_world_history           | 0.274262 |               237 |              65 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | international_law                   | mmlu_international_law                   | 0.239669 |               121 |              29 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | jurisprudence                       | mmlu_jurisprudence                       | 0.268519 |               108 |              29 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | logical_fallacies                   | mmlu_logical_fallacies                   | 0.226994 |               163 |              37 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | moral_disputes                      | mmlu_moral_disputes                      | 0.263006 |               346 |              91 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | moral_scenarios                     | mmlu_moral_scenarios                     | 0.237989 |               895 |             213 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | philosophy                          | mmlu_philosophy                          | 0.192926 |               311 |              59 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | prehistory                          | mmlu_prehistory                          | 0.209877 |               324 |              68 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | professional_law                    | mmlu_professional_law                    | 0.245111 |              1534 |             376 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | world_religions                     | mmlu_world_religions                     | 0.321637 |               171 |              55 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | business_ethics                     | mmlu_business_ethics                     | 0.3      |               100 |              30 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | clinical_knowledge                  | mmlu_clinical_knowledge                  | 0.222642 |               265 |              59 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | college_medicine                    | mmlu_college_medicine                    | 0.208092 |               173 |              36 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | global_facts                        | mmlu_global_facts                        | 0.19     |               100 |              19 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | human_aging                         | mmlu_human_aging                         | 0.313901 |               223 |              70 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | management                          | mmlu_management                          | 0.194175 |               103 |              20 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | marketing                           | mmlu_marketing                           | 0.290598 |               234 |              68 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | medical_genetics                    | mmlu_medical_genetics                    | 0.29     |               100 |              28 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | miscellaneous                       | mmlu_miscellaneous                       | 0.259259 |               783 |             203 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | nutrition                           | mmlu_nutrition                           | 0.222222 |               306 |              68 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | professional_accounting             | mmlu_professional_accounting             | 0.230496 |               282 |              65 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | professional_medicine               | mmlu_professional_medicine               | 0.183824 |               272 |              50 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | virology                            | mmlu_virology                            | 0.283133 |               166 |              47 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | econometrics                        | mmlu_econometrics                        | 0.236842 |               114 |              27 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_geography               | mmlu_high_school_geography               | 0.171717 |               198 |              34 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_government_and_politics | mmlu_high_school_government_and_politics | 0.196891 |               193 |              38 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_macroeconomics          | mmlu_high_school_macroeconomics          | 0.205128 |               390 |              80 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_microeconomics          | mmlu_high_school_microeconomics          | 0.210084 |               238 |              50 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_psychology              | mmlu_high_school_psychology              | 0.201835 |               545 |             110 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | human_sexuality                     | mmlu_human_sexuality                     | 0.259542 |               131 |              34 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | professional_psychology             | mmlu_professional_psychology             | 0.25817  |               612 |             158 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | public_relations                    | mmlu_public_relations                    | 0.236364 |               110 |              26 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | security_studies                    | mmlu_security_studies                    | 0.191837 |               245 |              47 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | sociology                           | mmlu_sociology                           | 0.268657 |               201 |              54 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | us_foreign_policy                   | mmlu_us_foreign_policy                   | 0.27     |               100 |              27 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | abstract_algebra                    | mmlu_abstract_algebra                    | 0.22     |               100 |              22 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | anatomy                             | mmlu_anatomy                             | 0.2      |               135 |              27 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | astronomy                           | mmlu_astronomy                           | 0.177632 |               152 |              27 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | college_biology                     | mmlu_college_biology                     | 0.263889 |               144 |              38 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | college_chemistry                   | mmlu_college_chemistry                   | 0.19     |               100 |              19 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | college_computer_science            | mmlu_college_computer_science            | 0.26     |               100 |              26 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | college_mathematics                 | mmlu_college_mathematics                 | 0.21     |               100 |              21 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | college_physics                     | mmlu_college_physics                     | 0.215686 |               102 |              22 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | computer_security                   | mmlu_computer_security                   | 0.28     |               100 |              28 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | conceptual_physics                  | mmlu_conceptual_physics                  | 0.26383  |               235 |              62 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | electrical_engineering              | mmlu_electrical_engineering              | 0.241379 |               145 |              35 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | elementary_mathematics              | mmlu_elementary_mathematics              | 0.21164  |               378 |              80 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_biology                 | mmlu_high_school_biology                 | 0.190323 |               310 |              58 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_chemistry               | mmlu_high_school_chemistry               | 0.152709 |               203 |              31 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_computer_science        | mmlu_high_school_computer_science        | 0.25     |               100 |              25 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_mathematics             | mmlu_high_school_mathematics             | 0.211111 |               270 |              57 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_physics                 | mmlu_high_school_physics                 | 0.198675 |               151 |              29 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | high_school_statistics              | mmlu_high_school_statistics              | 0.152778 |               216 |              33 |
| TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill | machine_learning                    | mmlu_machine_learning                    | 0.3125   |               112 |              35 |
| unsloth/Qwen3.5-27B                         | formal_logic                        | mmlu_formal_logic                        | 0.285714 |               126 |              36 |
| unsloth/Qwen3.5-27B                         | high_school_european_history        | mmlu_high_school_european_history        | 0.218182 |               165 |              36 |
| unsloth/Qwen3.5-27B                         | high_school_us_history              | mmlu_high_school_us_history              | 0.25     |               204 |              51 |
| unsloth/Qwen3.5-27B                         | high_school_world_history           | mmlu_high_school_world_history           | 0.270042 |               237 |              63 |
| unsloth/Qwen3.5-27B                         | international_law                   | mmlu_international_law                   | 0.239669 |               121 |              29 |
| unsloth/Qwen3.5-27B                         | jurisprudence                       | mmlu_jurisprudence                       | 0.259259 |               108 |              28 |
| unsloth/Qwen3.5-27B                         | logical_fallacies                   | mmlu_logical_fallacies                   | 0.220859 |               163 |              36 |
| unsloth/Qwen3.5-27B                         | moral_disputes                      | mmlu_moral_disputes                      | 0.248555 |               346 |              86 |
| unsloth/Qwen3.5-27B                         | moral_scenarios                     | mmlu_moral_scenarios                     | 0.237989 |               895 |             213 |
| unsloth/Qwen3.5-27B                         | philosophy                          | mmlu_philosophy                          | 0.186495 |               311 |              58 |
| unsloth/Qwen3.5-27B                         | prehistory                          | mmlu_prehistory                          | 0.216049 |               324 |              70 |
| unsloth/Qwen3.5-27B                         | professional_law                    | mmlu_professional_law                    | 0.245763 |              1534 |             377 |
| unsloth/Qwen3.5-27B                         | world_religions                     | mmlu_world_religions                     | 0.321637 |               171 |              55 |
| unsloth/Qwen3.5-27B                         | business_ethics                     | mmlu_business_ethics                     | 0.3      |               100 |              30 |
| unsloth/Qwen3.5-27B                         | clinical_knowledge                  | mmlu_clinical_knowledge                  | 0.215094 |               265 |              57 |
| unsloth/Qwen3.5-27B                         | college_medicine                    | mmlu_college_medicine                    | 0.208092 |               173 |              36 |
| unsloth/Qwen3.5-27B                         | global_facts                        | mmlu_global_facts                        | 0.18     |               100 |              18 |
| unsloth/Qwen3.5-27B                         | human_aging                         | mmlu_human_aging                         | 0.313901 |               223 |              70 |
| unsloth/Qwen3.5-27B                         | management                          | mmlu_management                          | 0.174757 |               103 |              18 |
| unsloth/Qwen3.5-27B                         | marketing                           | mmlu_marketing                           | 0.290598 |               234 |              68 |
| unsloth/Qwen3.5-27B                         | medical_genetics                    | mmlu_medical_genetics                    | 0.3      |               100 |              30 |
| unsloth/Qwen3.5-27B                         | miscellaneous                       | mmlu_miscellaneous                       | 0.240102 |               783 |             188 |
| unsloth/Qwen3.5-27B                         | nutrition                           | mmlu_nutrition                           | 0.22549  |               306 |              69 |
| unsloth/Qwen3.5-27B                         | professional_accounting             | mmlu_professional_accounting             | 0.234043 |               282 |              66 |
| unsloth/Qwen3.5-27B                         | professional_medicine               | mmlu_professional_medicine               | 0.183824 |               272 |              50 |
| unsloth/Qwen3.5-27B                         | virology                            | mmlu_virology                            | 0.283133 |               166 |              47 |
| unsloth/Qwen3.5-27B                         | econometrics                        | mmlu_econometrics                        | 0.236842 |               114 |              27 |
| unsloth/Qwen3.5-27B                         | high_school_geography               | mmlu_high_school_geography               | 0.176768 |               198 |              35 |
| unsloth/Qwen3.5-27B                         | high_school_government_and_politics | mmlu_high_school_government_and_politics | 0.196891 |               193 |              38 |
| unsloth/Qwen3.5-27B                         | high_school_macroeconomics          | mmlu_high_school_macroeconomics          | 0.202564 |               390 |              79 |
| unsloth/Qwen3.5-27B                         | high_school_microeconomics          | mmlu_high_school_microeconomics          | 0.210084 |               238 |              50 |
| unsloth/Qwen3.5-27B                         | high_school_psychology              | mmlu_high_school_psychology              | 0.192661 |               545 |             105 |
| unsloth/Qwen3.5-27B                         | human_sexuality                     | mmlu_human_sexuality                     | 0.259542 |               131 |              34 |
| unsloth/Qwen3.5-27B                         | professional_psychology             | mmlu_professional_psychology             | 0.25     |               612 |             153 |
| unsloth/Qwen3.5-27B                         | public_relations                    | mmlu_public_relations                    | 0.218182 |               110 |              24 |
| unsloth/Qwen3.5-27B                         | security_studies                    | mmlu_security_studies                    | 0.187755 |               245 |              46 |
| unsloth/Qwen3.5-27B                         | sociology                           | mmlu_sociology                           | 0.243781 |               201 |              49 |
| unsloth/Qwen3.5-27B                         | us_foreign_policy                   | mmlu_us_foreign_policy                   | 0.28     |               100 |              28 |
| unsloth/Qwen3.5-27B                         | abstract_algebra                    | mmlu_abstract_algebra                    | 0.22     |               100 |              22 |
| unsloth/Qwen3.5-27B                         | anatomy                             | mmlu_anatomy                             | 0.185185 |               135 |              25 |
| unsloth/Qwen3.5-27B                         | astronomy                           | mmlu_astronomy                           | 0.177632 |               152 |              27 |
| unsloth/Qwen3.5-27B                         | college_biology                     | mmlu_college_biology                     | 0.256944 |               144 |              37 |
| unsloth/Qwen3.5-27B                         | college_chemistry                   | mmlu_college_chemistry                   | 0.2      |               100 |              20 |
| unsloth/Qwen3.5-27B                         | college_computer_science            | mmlu_college_computer_science            | 0.26     |               100 |              26 |
| unsloth/Qwen3.5-27B                         | college_mathematics                 | mmlu_college_mathematics                 | 0.21     |               100 |              21 |
| unsloth/Qwen3.5-27B                         | college_physics                     | mmlu_college_physics                     | 0.215686 |               102 |              22 |
| unsloth/Qwen3.5-27B                         | computer_security                   | mmlu_computer_security                   | 0.28     |               100 |              28 |
| unsloth/Qwen3.5-27B                         | conceptual_physics                  | mmlu_conceptual_physics                  | 0.26383  |               235 |              62 |
| unsloth/Qwen3.5-27B                         | electrical_engineering              | mmlu_electrical_engineering              | 0.241379 |               145 |              35 |
| unsloth/Qwen3.5-27B                         | elementary_mathematics              | mmlu_elementary_mathematics              | 0.208995 |               378 |              79 |
| unsloth/Qwen3.5-27B                         | high_school_biology                 | mmlu_high_school_biology                 | 0.177419 |               310 |              55 |
| unsloth/Qwen3.5-27B                         | high_school_chemistry               | mmlu_high_school_chemistry               | 0.152709 |               203 |              31 |
| unsloth/Qwen3.5-27B                         | high_school_computer_science        | mmlu_high_school_computer_science        | 0.25     |               100 |              25 |
| unsloth/Qwen3.5-27B                         | high_school_mathematics             | mmlu_high_school_mathematics             | 0.211111 |               270 |              57 |
| unsloth/Qwen3.5-27B                         | high_school_physics                 | mmlu_high_school_physics                 | 0.198675 |               151 |              29 |
| unsloth/Qwen3.5-27B                         | high_school_statistics              | mmlu_high_school_statistics              | 0.152778 |               216 |              33 |
| unsloth/Qwen3.5-27B                         | machine_learning                    | mmlu_machine_learning                    | 0.3125   |               112 |              35 |

</details>

---

# The following best practices recommended by Qwen

## Best Practices

To achieve optimal performance, we recommend the following settings:

1. **Sampling Parameters**:  
   - We suggest using the following sets of sampling parameters depending on the mode and task type:  
     - **Thinking mode for general tasks**:  
       `temperature=1.0`, `top_p=0.95`, `top_k=20`, `min_p=0.0`, `presence_penalty=1.5`, `repetition_penalty=1.0`  
     - **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`  
     - **Instruct (or non-thinking) mode for general tasks**:  
       `temperature=0.7`, `top_p=0.8`, `top_k=20`, `min_p=0.0`, `presence_penalty=1.5`, `repetition_penalty=1.0`  
     - **Instruct (or non-thinking) mode for reasoning tasks**:  
       `temperature=1.0`, `top_p=1.0`, `top_k=40`, `min_p=0.0`, `presence_penalty=2.0`, `repetition_penalty=1.0`  
   - 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.

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.

3. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.
   - **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
   - **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"`."

4. **No Thinking Content in History**: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. It is implemented in the provided chat template in Jinja2. However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.

5. **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,
    ```json
    {"longest_edge": 469762048, "shortest_edge": 4096}
    ```

    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).