services:

  eval-llm:
    build: .
    container_name: eval-llm-safety
    environment:
      - HF_TOKEN=${HF_TOKEN}  # Pass Hugging Face token from .env
    volumes:
      - ./src:/app  # Mount the current directory for code changes
    command: python /app/eval_llm.py --model_name "HuggingFaceTB/SmolLM2-135M-Instruct" --dataset_name "declare-lab/HarmfulQA" --num_samples 10 --temperature 0.5 --seed 42 --output_file results.json
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
  
  gradio-app:
    build: .
    container_name: eval-llm-safety-gradio
    environment:
      - HF_TOKEN=${HF_TOKEN}  # Pass Hugging Face token from .env
    volumes:
      - .:/app  # Mount the entire project directory
    ports:
      - "7861:7860"  # Expose Gradio port on 7861 to avoid conflict
    entrypoint: ["python3", "-u", "/app/src/app.py"]
    # add access to GPU if available
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]