Qwen
/

Qwen3-235B-A22B

Text Generation

Model card Files Files and versions

Instructions to use Qwen/Qwen3-235B-A22B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Qwen/Qwen3-235B-A22B with Transformers:

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

pipe = pipeline("text-generation", model="Qwen/Qwen3-235B-A22B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-235B-A22B")
model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-235B-A22B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Qwen/Qwen3-235B-A22B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Qwen/Qwen3-235B-A22B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Qwen/Qwen3-235B-A22B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Qwen/Qwen3-235B-A22B

How to use Qwen/Qwen3-235B-A22B 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 "Qwen/Qwen3-235B-A22B" \
    --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": "Qwen/Qwen3-235B-A22B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "Qwen/Qwen3-235B-A22B" \
        --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": "Qwen/Qwen3-235B-A22B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Qwen/Qwen3-235B-A22B with Docker Model Runner:
```
docker model run hf.co/Qwen/Qwen3-235B-A22B
```

Resources

View closed (3)

Listed on OpenModelMap

#48 opened 20 days ago by

Listed on OpenModelMap

#47 opened about 1 month ago by

TemporalMesh Transformer: 29.4 PPL at 48% compute — beats Mamba, new open-source architecture

#46 opened about 2 months ago by

Add EvalEval community eval results

#45 opened about 2 months ago by

Add SWE-Bench Pro evaluation results

#44 opened 5 months ago by

What is the minimum hardware needed to run Qwen3-235B-A22B with vLLM?

#43 opened 8 months ago by

yarn scale to 122,880 context length

#41 opened about 1 year ago by

chat_template issue

#40 opened about 1 year ago by

dannt

#39 opened about 1 year ago by

Add assistant mask support to Qwen3-235B-A22B

#38 opened about 1 year ago by

system prompt suggestion

#37 opened about 1 year ago by

Publish to GitHub Marketplace

#36 opened about 1 year ago by

The Qwen3-235B-A22B model is not as effective as the Qwen3-32B model.

#35 opened about 1 year ago by

Video?

#34 opened about 1 year ago by

关于Qwen3-235B-A22B在文学创作中“思维链”模式对输出风格影响的观察与建议

#33 opened about 1 year ago by

In complex reasoning tasks Qwen3 is far behind QwQ

#32 opened about 1 year ago by

I know not related but saw This Star Wars picture on Facebook and i thought it is code so trying to say Hi to Someone named John!

#31 opened about 1 year ago by

Not sure Clocks keep coming up and keep getting the run around might need time stamping confirmations?

#30 opened about 1 year ago by

Not Sure Orginal to Persian to Binary?

#29 opened about 1 year ago by

Qwen3 is simply amazing.

#28 opened about 1 year ago by

Upload b891b3c3bb6a146a8e809bb72a06d101.png

#27 opened about 1 year ago by

Add image visual recognition output just like qwen 2.5 vl-32b instruct

#26 opened about 1 year ago by

English to French - French to English based on Meta & HuggingFace Chat Bot

#25 opened about 1 year ago by

Qwen3 幻觉太高了，比 Qwen 2.5 差太多了

#24 opened about 1 year ago by

Upload 3 files

#23 opened about 1 year ago by

Upload 3 files

#22 opened about 1 year ago by

Model keeps talking about Cumhurbaşkanlığı Sarayı when speaking Turkish

#21 opened about 1 year ago by

Qwen3 not Using Tools in Complex Prompts Unlike QwQ-32B

#20 opened about 1 year ago by

Thanks a lot for this release

#19 opened about 1 year ago by

Does anyone feel Qwen3 often fails to follow instructions accurately?

#18 opened about 1 year ago by

Two of the base models are missing

#17 opened about 1 year ago by deleted

Qwen is loosing broad knowledge since Qwen2.

#16 opened about 1 year ago by

GPQA perf for DSV3-Base seems wrong

#15 opened about 1 year ago by

72B-MoE

#13 opened about 1 year ago by

235B会放出来Base模型吗？

#12 opened about 1 year ago by

看模型介绍和模型结构里面没有关于vision encoder的部分，但是在qwen的在线模型服务界面可以用这个模型去看图片，想问下视觉部分是复用了哪个vision encoder呢？

#11 opened about 1 year ago by

有用4张H20实践过的大佬吗

#10 opened about 1 year ago by

8张80G显存的8卡A100能部署不？

#9 opened about 1 year ago by

User rating and reviews of Qwen3 App and Qwen3 Model

#8 opened about 1 year ago by

是不是奖励函数没有ngram重复度惩罚

#7 opened about 1 year ago by

🚀[Fine-tuning] Qwen3-MoE Megatron Training Implementation and Best Practices👋

#6 opened about 1 year ago by

【Evaluation】Best practice for evaluating Qwen3 !!

#5 opened about 1 year ago by

Please upload the base model for this one

#4 opened about 1 year ago by

GPTQ/AWQ

#3 opened about 1 year ago by

Add languages tag

#2 opened about 1 year ago by

de-francophones