docker-compose.yml

version: '3.8'

services:
  bitnet-web:
    build:
      context: .
      dockerfile: Dockerfile
    container_name: bitnet-chat
    ports:
      - "127.0.0.1:8080:8080"
    volumes:
      # Mount the model directory to avoid downloading inside the container
      - ./models:/app/models
      # Mount the build directory to persist compiled binaries
      - ./build:/app/build
    environment:
      - MODEL_PATH=/app/models/BitNet-b1.58-2B-4T/ggml-model-i2_s.gguf
      - CTX_SIZE=2048
      - N_PREDICT=512
      - TEMPERATURE=0.8
      - THREADS=4
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s

# Optional: Use for GPU acceleration with NVIDIA Docker
# Uncomment the following and use --gpus all flag when running
#   deploy:
#     resources:
#       reservations:
#         devices:
#           - driver: nvidia
#             count: 1
#             capabilities: [gpu]