Instructions to use khazarai/Qwen3.8-max-Reasoning-Distilled with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use khazarai/Qwen3.8-max-Reasoning-Distilled with Transformers:

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

pipe = pipeline("image-text-to-text", model="khazarai/Qwen3.8-max-Reasoning-Distilled")
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("khazarai/Qwen3.8-max-Reasoning-Distilled")
model = AutoModelForMultimodalLM.from_pretrained("khazarai/Qwen3.8-max-Reasoning-Distilled", 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 khazarai/Qwen3.8-max-Reasoning-Distilled with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "khazarai/Qwen3.8-max-Reasoning-Distilled"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "khazarai/Qwen3.8-max-Reasoning-Distilled",
		"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/khazarai/Qwen3.8-max-Reasoning-Distilled

SGLang

How to use khazarai/Qwen3.8-max-Reasoning-Distilled 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 "khazarai/Qwen3.8-max-Reasoning-Distilled" \
    --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": "khazarai/Qwen3.8-max-Reasoning-Distilled",
		"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 "khazarai/Qwen3.8-max-Reasoning-Distilled" \
        --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": "khazarai/Qwen3.8-max-Reasoning-Distilled",
		"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 Desktop
Docker Model Runner
How to use khazarai/Qwen3.8-max-Reasoning-Distilled with Docker Model Runner:
```
docker model run hf.co/khazarai/Qwen3.8-max-Reasoning-Distilled
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Qwen3.8-max-Reasoning-Distilled

Qwen3.8-max-Reasoning-Distilled is a compact, high-efficiency language model fine-tuned using Knowledge Distillation from Qwen3.8-max.

It specifically addresses the common issue of overthinking loops and repetitive hesitations found in smaller raw reasoning models. By learning directly from high-quality reasoning traces, this model delivers concise, structured, and decisive Chain-of-Thought (CoT) outputs without wasting tokens on circular verification.

Key Features

Streamlined Chain-of-Thought (CoT): Replaces verbose, rambling reasoning paths with direct, step-by-step logic.
Elimination of Overthinking: Suppresses repetitive inner monologue loops (e.g., "Wait, let me re-verify...") that consume unnecessary tokens.
Higher Inference Efficiency: Drastically reduces total generated token count while maintaining or improving final answer accuracy.
Distilled Logic from Qwen3.8-max: Captures the complex evaluation heuristics of the teacher model into a smaller, faster student architecture.

Reasoning Quality Comparison

Below is a comparison highlighting how distillation improves reasoning structure and eliminates repetitive loops:

Feature	Qwen3.5-0.8B	Qwen3.8-max-Reasoning-Distilled
Logic Flow	Frequent self-doubt, circular checks, hesitation	Linear, structured, and hypothesis-driven
Token Economy	High token overhead spent on repetitive verification	Low token overhead with concise step evaluation
Option Analysis	Reiterates choices multiple times without deciding	Evaluates each option once with clear justification
Decision Speed	Slow convergence to final answer	Fast, decisive output generation

Example Trace

Task: Identify the beneficial effect of carpooling from multiple choice options.

Qwen3.5-0.8B: "Evaluate option D... wait, let me check option E... actually, let me re-evaluate D... wait, is there a trick? Let me double-check..." (Overthinking Loop)
Qwen3.8-max-Reasoning-Distilled: "Systematically evaluates choices A through H in a single pass $\rightarrow$ identifies option D as the primary environmental benefit $\rightarrow$ concludes decisively." (Clean & Direct)

Limitations

While the model minimizes overthinking, extremely complex multi-step mathematical problems may still require prompting for explicit scratchpad steps. Always verify outputs when using the model for domain-critical tasks (e.g., medical, legal, or financial decisions).