vLLM

Overview

vLLM is a high-throughput inference engine for serving Large Language Models in production. It exposes an OpenAI-compatible API and uses PagedAttention for efficient memory management.

Official Website: https://vllm.ai Documentation: https://docs.vllm.ai

Key features

• PagedAttention — Efficient KV-cache memory management
• High throughput — Optimized continuous batching for production serving
• OpenAI-compatible — Drop-in /v1/chat/completions API
• Multi-GPU — Tensor and pipeline parallelism
• Any Hugging Face model — Serve any compatible open model

Local usage

vLLM is self-hosted. Start a server and call it directly — no API key is required for a local instance. The default endpoint is http://localhost:8000/v1.

# Start vLLM server
vllm serve meta-llama/Llama-3.1-8B-Instruct

# Call the local OpenAI-compatible endpoint
curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta-llama/Llama-3.1-8B-Instruct",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Using vLLM with Yunxin

When a vLLM backend is registered with Yunxin, address its models using the provider/model convention by prefixing with vllm/ (for example, vllm/meta-llama/Llama-3.1-8B-Instruct). Requests use the same Yunxin base URL (https://api.yuhuanstudio.com/v1) and your Yunxin sk- API key.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["YUNXIN_API_KEY"],
    base_url="https://api.yuhuanstudio.com/v1"
)

response = client.chat.completions.create(
    model="vllm/meta-llama/Llama-3.1-8B-Instruct",
    messages=[{"role": "user", "content": "Hello!"}]
)

print(response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.YUNXIN_API_KEY,
  baseURL: "https://api.yuhuanstudio.com/v1",
});

const response = await client.chat.completions.create({
  model: "vllm/meta-llama/Llama-3.1-8B-Instruct",
  messages: [{ role: "user", content: "Hello!" }],
});

console.log(response.choices[0].message.content);

curl https://api.yuhuanstudio.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $YUNXIN_API_KEY" \
  -d '{
    "model": "vllm/meta-llama/Llama-3.1-8B-Instruct",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Available models

vLLM serves whichever Hugging Face model you start the server with; capabilities depend on the model and your hardware. When registered with Yunxin, query the live catalog with the Models API:

curl "https://api.yuhuanstudio.com/v1/models?provider=vllm" \
  -H "Authorization: Bearer $YUNXIN_API_KEY"

Official resources

• vLLM Website
• Documentation
• GitHub