Instructions to use drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4 with Transformers:

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

pipe = pipeline("text-generation", model="drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4")
model = AutoModelForCausalLM.from_pretrained("drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4")
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]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4

SGLang

How to use drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4 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 "drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4" \
    --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": "drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4",
		"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 "drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4" \
        --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": "drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4 with Docker Model Runner:
```
docker model run hf.co/drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4
```

DeepSeek-R1-Distill-Qwen-32B-NVFP4

File size: 1,774 Bytes

---
license: mit
license_link: https://opensource.org/license/mit
base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
tags:
  - quantized
  - 4-bit
  - int4
  - awq
language:
  - en
library_name: transformers
pipeline_tag: text-generation
---

# DeepSeek-R1-Distill-Qwen-32B-NVFP4

INT4 weight-only quantization of [`deepseek-ai/DeepSeek-R1-Distill-Qwen-32B`](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B).

DeepSeek-R1-Distill-Qwen-32B in NVFP4 W4A4. Native vLLM compressed-tensors. About 17 GB on disk.

| Property | Value |
|---|---|
| Base model | [deepseek-ai/DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) |
| Quantization | INT4 weight-only |
| Approx. on-disk size | ~20.7 GB |
| License | MIT License |
| Languages | English |

## Load (vLLM)

```bash
vllm serve drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4 \
  --max-model-len 32768 \
  --gpu-memory-utilization 0.94
```

```python
from vllm import LLM, SamplingParams
llm = LLM(model="drawais/DeepSeek-R1-Distill-Qwen-32B-NVFP4", max_model_len=32768)
print(llm.generate(["Hello!"], SamplingParams(max_tokens=128))[0].outputs[0].text)
```

## Footprint

~20.7 GB on disk. Recommended VRAM: enough headroom for KV cache.

## License & attribution

This artifact is a derivative work of [`deepseek-ai/DeepSeek-R1-Distill-Qwen-32B`](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B),
released by its original authors under the **MIT License**.

This artifact is distributed under the same license. The full license text is
included in [`LICENSE`](LICENSE), and required attribution is in [`NOTICE`](NOTICE).

License text: https://opensource.org/license/mit
Source model: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B