version: '3.8'

services:
  llama-api:
    build: .
    ports:
      - "53693:53693"
    volumes:
      - ./models:/app/models
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    environment:
      - MODEL_PATH=/app/Llama-3.2-1B-Instruct-IQ3_M.gguf
      - USE_GPU=1
      - N_GPU_LAYERS=32
      - N_THREADS=4
      - N_CONTEXT=2048