Setting Up vLLM with Hugging Face for Generative AI Projects

03 Aug, 2024

Installing

Just run:

git clone https://github.com/vllm-project/vllm.git
cd vllm
docker build -f Dockerfile.cpu -t vllm-cpu-env --shm-size=4g .

Once you get it installed, you can login to Hugging Face:

huggingface-cli login

You will pull the models from Hugging Face, so you need to be logged in.

Start vLLM server

Run:

docker run -it \
     -v ~/.cache/huggingface:/root/.cache/huggingface \
     --rm \
     --network=host \
     vllm-cpu-env \
     --model facebook/opt-125m \
     --api-key 1234banana

This will install Llama-3-Groq-8B-Tool-Use model and start the server.

Testing

You can curl the server to test it:

curl http://localhost:8000/v1/completions \
    -H "Content-Type: application/json" \
    -d '{
        "model": "facebook/opt-125m",
        "prompt": "Dario Meira seria muito",
        "max_tokens": 50,
        "temperature": 0,
        "api_key": "1234banana"
    }'

Bonus: using LiteLLM to consume vLLM

base_url = "http://localhost:8000/v1"
model_name = "facebook/opt-125m"
api_key = "s3cr3t"

# vLLM uses the OpenAI API, so we need to set the provider to "openai"
PROVIDER = "openai"

completion(
    model=f"{PROVIDER}/{model_name}",
    api_key=api_key,
    base_url=base_url,
    messages=messages,
)

Voila! You have vLLM running on your machine.

#Generative-Ai