Description

This pull request adds a new example demonstrating how to useLightRAG with vLLM-served models for large language model inference, text embeddings, and reranking.
The example mirrors the structure and conventions of existing LightRAG demos (e.g. Gemini) while showcasing a fully self-hosted, OpenAI-compatible vLLM deployment.

Related Issues

• N/A (example contribution)

Changes Made

• Added a new example script:examples/lightrag_vllm_demo.py

• Demonstrates integration of:

  • vLLM-served LLM via OpenAI-compatible API
  • vLLM-served embedding model
  • Jina-compatible reranker served via vLLM

• Included clear setup instructions and required environment variables

• Aligned code structure and formatting with existing LightRAG example demos

• Provided a realistic hybrid query with reranking enabled

Checklist

• Changes tested locally
• Code reviewed
• Documentation updated (inline script docstring)
• Unit tests added (not applicable for example script)

Additional Notes

This example is intended to help users deploy LightRAG inself-hosted and on-prem environments using vLLM, without reliance on external proprietary APIs. It follows existing LightRAG patterns to ensure consistency and ease of adoption.