vllm-setup

bởi nvidia

Triển khai máy chủ suy luận vLLM trên NVIDIA DGX Station GB300 với container đã xác thực, nhắm mục tiêu GPU và các tham số tinh chỉnh. Sử dụng khi người dùng yêu cầu phục vụ một…

npx skills add https://github.com/nvidia/dgx-spark-playbooks --skill vllm-setup

Tải ZIP GitHub

1.3k

Dùng thử kỹ năng nàyĐược tài trợ

vLLM Setup on DGX Station

Deploy a vLLM inference server on DGX Station with validated configuration.

Steps

Find the GB300 GPU index. Run:
```
nvidia-smi --query-gpu=index,name --format=csv,noheader
```
Identify the device index for the GB300 (typically device 1). Use this index for --gpus below. Do NOT use --gpus all — mixed coherency will cause CUDA failures.
Ask the user which model to serve. If they don't have a preference, suggest:
- nvidia/Qwen3-235B-A22B-NVFP4 — large MoE model, fits in 279 GB HBM
- meta-llama/Llama-3.1-70B-Instruct — solid general-purpose model
- Qwen/Qwen3-8B — small model for testing
Check if the user has an HF_TOKEN. Many models require HuggingFace authentication. The token must be passed inline with -e HF_TOKEN="..." — do not rely on shell export in background Docker tasks.

Deploy the container. Use this validated configuration:

docker pull nvcr.io/nvidia/vllm:26.01-py3

docker run -d \
  --name vllm-server \
  --gpus '"device=<GB300_INDEX>"' \
  --ipc host \
  --ulimit memlock=-1 \
  --ulimit stack=67108864 \
  -p 8000:8000 \
  -e HF_TOKEN="<TOKEN>" \
  -v "$HOME/.cache/huggingface/hub:/root/.cache/huggingface/hub" \
  nvcr.io/nvidia/vllm:26.01-py3 \
  vllm serve "<MODEL>" \
    --max-model-len 32768 \
    --gpu-memory-utilization 0.9

Container version: Use nvcr.io/nvidia/vllm:26.01-py3. Do NOT use 25.10 — it has a FlashInfer buffer overflow on DGX Station.

Wait for the server to be ready. Monitor logs:
```
docker logs -f vllm-server
```
Wait for the line indicating the server is listening on port 8000.

Test the server:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<MODEL>",
    "messages": [{"role": "user", "content": "Hello"}],
    "max_tokens": 64
  }'

Report the result to the user, including:
- Model loaded and serving on port 8000
- GPU memory utilization
- How to stop: docker stop vllm-server && docker rm vllm-server

Tuning parameters

Adjust these based on the user's workload:

Parameter	Default	Agent workloads	Throughput workloads
`--max-model-len`	32768	32768-65536	8192-16384
`--gpu-memory-utilization`	0.9	0.85-0.90	0.90-0.92
`--enable-prefix-caching`	off	Enable (multi-turn reuse)	Enable
`--max-num-seqs`	default	4-16 (lower latency)	32+ (higher throughput)