Instructions to use Qwen/Qwen3-32B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Qwen/Qwen3-32B with Transformers:

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

pipe = pipeline("text-generation", model="Qwen/Qwen3-32B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-32B")
model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-32B")
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

vLLM

How to use Qwen/Qwen3-32B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Qwen/Qwen3-32B"
# 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/Qwen3-32B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Qwen/Qwen3-32B

SGLang

How to use Qwen/Qwen3-32B 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/Qwen3-32B" \
    --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/Qwen3-32B",
		"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/Qwen3-32B" \
        --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/Qwen3-32B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Qwen/Qwen3-32B with Docker Model Runner:
```
docker model run hf.co/Qwen/Qwen3-32B
```

Base Model?

by Downtown-Case - opened Apr 28, 2025

Discussion

Downtown-Case

Apr 28, 2025

It appears the base model for the 30B MoE was uploaded, but there's none for this one.

Is that an oversight?

AuriAetherwiing

Apr 28, 2025

It seems like this model is the only one lacking a base variant. That's a shame, I could really use it.

KnutJaegersberg

Qwen org Apr 29, 2025

yep the base model would be great for the 32b variant

reign12

Apr 29, 2025

this is true! need the base model!

guang

Apr 29, 2025

is a base model or instruct model???

Downtown-Case

May 2, 2025

Bump. Any word on this?

tmostak

May 5, 2025

Adding to this, we really could use the base model as well.

liheng

May 9, 2025

Indeed, a 32B base model is really needed.

eramax

May 13, 2025

please release the base model for 32b qwen3

KnutJaegersberg

Qwen org Jun 3, 2025

the paper

Writing-Zero: Bridge the Gap Between Non-verifiable Problems and Verifiable Rewards

contains evidence that it exists

Vikashsharmask

Jun 4, 2025

Hello

MikeDd2025

Jun 6, 2025

is it out?

KnutJaegersberg

Qwen org Jun 11, 2025

people already do continued pretraining to restore a base model from the fine tuned version:

"Our Approach to DeepSeek-R1-0528-Distill-Qwen3-32B-Preview0-QAT:
Since Qwen3 did not provide a pre-trained base for its 32B model, our initial step was to perform additional pre-training on Qwen3-32B using a self-constructed multilingual pre-training dataset. This was done to restore a "pre-training style" model base as much as possible, ensuring that subsequent work would not be influenced by Qwen3's inherent SFT language style. This model will also be open-sourced in the future."

https://www.reddit.com/r/LocalLLaMA/comments/1l7mijq/i_found_a_deepseekr10528distillqwen332b/

daniel-dona

Jun 13, 2025

people already do continued pretraining to restore a base model from the fine tuned version:

"Our Approach to DeepSeek-R1-0528-Distill-Qwen3-32B-Preview0-QAT:
Since Qwen3 did not provide a pre-trained base for its 32B model, our initial step was to perform additional pre-training on Qwen3-32B using a self-constructed multilingual pre-training dataset. This was done to restore a "pre-training style" model base as much as possible, ensuring that subsequent work would not be influenced by Qwen3's inherent SFT language style. This model will also be open-sourced in the future."

https://www.reddit.com/r/LocalLLaMA/comments/1l7mijq/i_found_a_deepseekr10528distillqwen332b/

What's the difference between doing that and cold-starting the training?

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