Graph Retrieval Augmented Generation

LLMにおけるRAGの仕組みを一通り理解すると、Graph RAGというものに興味を持ちました。
NVIDIA Geforce RTX5060Ti 16GB も購入したので、試してみました。
グラフィックボードを有効に動かすには、いろいろと設定が面倒でした。
いろんなサンプルを動かそうとしましたが、まだ新しいドライバなのでPythonのライブラリの依存性が合わず、思ったより苦労します。
そんな中下記サイトではDocker環境で全て用意していただいているため、順調に動作させることができました。

引用）https://highreso.jp/edgehub/machinelearning/graphrag.html

langchain.Dockerfile

FROM nvidia/cuda:12.1.0-cudnn8-devel-ubuntu22.04
RUN apt-get update && apt-get install -y python3-pip python3-venv git nano curl
RUN curl -fsSL https://ollama.com/install.sh | sh
WORKDIR /app
COPY . /app
RUN python3 -m venv /app/.venv
RUN echo "source /app/.venv/bin/activate" >> /root/.bashrc
RUN /app/.venv/bin/pip install Jupyter jupyterlab
RUN /app/.venv/bin/pip install ollama langchain-ollama langchain langsmith langchain-graphrag langchain-community langchain_huggingface langchain_core langchain-experimental neo4j langchain_neo4j tiktoken
CMD ["/bin/bash"]

FROM nvidia/cuda:12.1.0-cudnn8-devel-ubuntu22.04

RUN apt-get update && apt-get install -y python3-pip python3-venv git nano curl

RUN curl -fsSL https://ollama.com/install.sh | sh

WORKDIR /app

COPY . /app

RUN python3 -m venv /app/.venv

RUN echo "source /app/.venv/bin/activate" >> /root/.bashrc

RUN /app/.venv/bin/pip install Jupyter jupyterlab

RUN /app/.venv/bin/pip install ollama langchain-ollama langchain langsmith langchain-graphrag langchain-community langchain_huggingface langchain_core langchain-experimental neo4j langchain_neo4j tiktoken

CMD ["/bin/bash"]

neo4j.Dockerfile

FROM neo4j:5.25.1
RUN apt-get update && apt-get install -y curl && apt-get clean
ENV NEO4JLABS_PLUGINS '[ "apoc" ]'
ENV NEO4J_dbms_security_procedures_unrestricted apoc.*
RUN curl -L -o /var/lib/neo4j/plugins/apoc-5.25.1-core.jar \
    https://github.com/neo4j/apoc/releases/download/5.25.1/apoc-5.25.1-core.jar
EXPOSE 7474 7687
CMD ["neo4j"]

FROM neo4j:5.25.1

RUN apt-get update && apt-get install -y curl && apt-get clean

ENV NEO4JLABS_PLUGINS '[ "apoc" ]'

ENV NEO4J_dbms_security_procedures_unrestricted apoc.*

RUN curl -L -o /var/lib/neo4j/plugins/apoc-5.25.1-core.jar \

https://github.com/neo4j/apoc/releases/download/5.25.1/apoc-5.25.1-core.jar

EXPOSE 7474 7687

CMD ["neo4j"]

docker-compose.yml

services:
  langchain_neo4j:
    build:
      context: .
      dockerfile: langchain.Dockerfile
    image: langchain
    runtime: nvidia
    container_name: langchain
    ports:
      - "8888:8888"
    volumes:
      - .:/app/langchain
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    networks:
      - app-network
    depends_on:
      - neo4j
    command: >
      bash -c '/usr/local/bin/ollama serve & /app/.venv/bin/jupyter lab --ip="*" --port=8888 --NotebookApp.token="" --NotebookApp.password="" --no-browser --allow-root'

  neo4j:
    build:
      context: .
      dockerfile: neo4j.Dockerfile
    image: neo4j_with_apoc
    container_name: neo4j
    ports:
      - "7474:7474"
      - "7687:7687"
    volumes:
      - ./data:/data
      - ./plugins:/plugins
      - ./import:/import
    environment:
      - NEO4J_apoc_export_file_enabled=true
      - NEO4J_apoc_import_file_enabled=true
      - NEO4J_apoc_import_file_use__neo4j__config=true
      - NEO4J_dbms_security_procedures_unrestricted=apoc.*
      - NEO4J_dbms_memory_heap_initial__size=512m
      - NEO4J_dbms_memory_heap_max__size=2G
      - NEO4J_apoc_uuid_enabled=true
      - NEO4J_dbms_default__listen__address=0.0.0.0
      - NEO4J_dbms_allow__upgrade=true
      - NEO4J_dbms_default__database=neo4j
      - NEO4J_AUTH=neo4j/your_password
    restart: unless-stopped
    networks:
      - app-network

networks:
  app-network:
    driver: bridge

services:

langchain_neo4j:

build:

context: .

dockerfile: langchain.Dockerfile

image: langchain

runtime: nvidia

container_name: langchain

ports:

- "8888:8888"

volumes:

- .:/app/langchain

deploy:

resources:

reservations:

devices:

- driver: nvidia

count: 1

capabilities: [gpu]

networks:

- app-network

depends_on:

- neo4j

command: >

bash -c '/usr/local/bin/ollama serve & /app/.venv/bin/jupyter lab --ip="*" --port=8888 --NotebookApp.token="" --NotebookApp.password="" --no-browser --allow-root'

neo4j:

build:

context: .

dockerfile: neo4j.Dockerfile

image: neo4j_with_apoc

container_name: neo4j

ports:

- "7474:7474"

- "7687:7687"

volumes:

- ./data:/data

- ./plugins:/plugins