Instructions to use Qwen/Qwen3-235B-A22B-Instruct-2507 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Qwen/Qwen3-235B-A22B-Instruct-2507 with Transformers:

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

pipe = pipeline("text-generation", model="Qwen/Qwen3-235B-A22B-Instruct-2507")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-235B-A22B-Instruct-2507")
model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-235B-A22B-Instruct-2507", 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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Qwen/Qwen3-235B-A22B-Instruct-2507 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-Instruct-2507"
# 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-Instruct-2507",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Qwen/Qwen3-235B-A22B-Instruct-2507

SGLang

How to use Qwen/Qwen3-235B-A22B-Instruct-2507 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-Instruct-2507" \
    --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-Instruct-2507",
		"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-Instruct-2507" \
        --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-Instruct-2507",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Qwen/Qwen3-235B-A22B-Instruct-2507 with Docker Model Runner:
```
docker model run hf.co/Qwen/Qwen3-235B-A22B-Instruct-2507
```

Good idea to remove the hybrid thinking mode

#16

by rtzurtz - opened Jul 22, 2025

Discussion

rtzurtz

Jul 22, 2025

•

edited Jul 22, 2025

(hybrid = thinking/reasoning + normal mode in one model. thinking/reasoning can be switched on or off in the prompt (/no_think in that case).)
I noticed Qwen3-32B worse evaluation results (in no think mode) vs Qwen2.5-32B in Falcon-H1 evaluation testing first. This is indeed maybe due to the hybrid thinking mode. So it's a no-brainer to remove the hybrid mode (if that's the reason for the improvement) and thank you for doing so, when the scores increase by that much.
Another reason is that SSD space is cheap, so if one needs the reasoning model, one can simply download it and use it and it will perform better than a hybrid thinking LLM.

PS: Many, including me, hope you release the other smaller Qwen3s, each as non-reasoning and reasoning (reasoning of this 235B ofc too).

sneng

Jul 23, 2025

I noticed that latest models, such as this Qwen3 instruct model and Kimi K2, emphasize their non-thinking mode performance. Can anyone shed some light on why the community is preferring non-thinking models recently?

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment