docker-compose.yml

# Docker Compose file for BGE Embedding Server

services:
  embed-server:
    build: .
    container_name: bge_embed_server
    environment:
      - EMBED_API_KEY=${EMBED_API_KEY}
      - MODEL_NAME=${MODEL_NAME}
      - BATCH_SIZE=${BATCH_SIZE:-8}
      - MAX_LENGTH=${MAX_LENGTH:-1024}
      - RETRY_ATTEMPTS=${RETRY_ATTEMPTS:-3}
      - RETRY_BACKOFF_SECONDS=${RETRY_BACKOFF_SECONDS:-2}
      - LOG_LEVEL=${LOG_LEVEL:-INFO}
      - PORT=${PORT}
      - HOST=${HOST}
      - WORKERS=${WORKERS}
      - DEVICE=${DEVICE}
    ports:
      - "${PORT:-7860}:7860"
    volumes:
      - ./:/app
    restart: unless-stopped
    # If i have GPU server i simply uncomment it to use gpu
    # deploy:
    #   resources:
    #     reservations:
    #       devices:
    #         - driver: nvidia
    #           count: all
    #           capabilities: [ gpu ]