chore(llm): multi-stage building in Dockerfile

[imbajin]

follow #195

After the image uploaded, users could use the following cmd like:

docker run -d --name rag -p 8001:8001 \
  -v /path/to/.env:/home/work/hugegraph-llm/.env \
  -v /path/to/resources:/home/work/hugegraph-llm/src/hugegraph_llm/resources \
  hugegraph/graphrag:1.5.0

or configs in docker-compose.yml

    volumes:
      # Mount local '.env' file, could use ${ENV_FILE_PATH:-/dev/null} to avoid error
      - /path/to/.env:/home/work/hugegraph-llm/.env
      # Mount local resources file
      - /path/to/resources:/home/work/hugegraph-llm/src/hugegraph_llm/resources

or volume/configmap/pvc in k8s

Build/Test the image in root

# Build Image
docker build -f docker/Dockerfile.llm -t graphrag .

# Test/Run container
docker run -it --name rag -p 8001:8001 graphrag  bash

Note

Currently we store the vector data in local by faiss, should replace it as a separate processes/service make rag services stateless)
Or the graph database supports small-scale vector indexing itself

TODO:

• use https://github.com/Nuitka/Nuitka to improve performance and compress runtime dependencies