compose.github.yaml

name: llmgoat-github
services:
  llmgoat-cpu:
    container_name: llmgoat-cpu
    image: ghcr.io/secforce/llmgoat-cpu:latest
    environment:
      - LLMGOAT_SERVER_HOST=0.0.0.0 # Customise the server host
      - LLMGOAT_SERVER_PORT=5000 # Customise the server port
      - LLMGOAT_DEFAULT_MODEL=gemma-2.gguf # Customise the initial model (should be available in the models folder)
      - LLMGOAT_N_THREADS=16 # Customise the number of threads to use (max value used is the total number of cores)
      - LLMGOAT_N_GPU_LAYERS=0 # Use this to enable GPU-usage. 20 is a good starting value
      - LLMGOAT_VERBOSE=0 # "1" for verbose, "0" for silent
    ports:
      - "${LLMGOAT_SERVER_PORT:-5000}:${LLMGOAT_SERVER_PORT:-5000}"
    volumes:
      - ${HOME}/.LLMGoat/models:/root/.LLMGoat/models
      - ${HOME}/.LLMGoat/cache:/root/.LLMGoat/cache # To avoid downloading each time the model for challenge A09

  llmgoat-gpu:
    container_name: llmgoat-gpu
    image: ghcr.io/secforce/llmgoat-gpu:latest
    environment:
      - LLMGOAT_SERVER_HOST=0.0.0.0 # Customise the server host
      - LLMGOAT_SERVER_PORT=5000 # Customise the server port
      - LLMGOAT_DEFAULT_MODEL=gemma-2.gguf # Customise the initial model (should be available in the models folder)
      - LLMGOAT_N_THREADS=16 # Customise the number of threads to use (max value used is the total number of cores)
      - LLMGOAT_N_GPU_LAYERS=20 # Use this to enable GPU-usage. 20 is a good starting value
      - LLMGOAT_VERBOSE=0 # "1" for verbose, "0" for silent
    ports:
      - "${LLMGOAT_SERVER_PORT:-5000}:${LLMGOAT_SERVER_PORT:-5000}"
    volumes:
      - ${HOME}/.LLMGoat/models:/root/.LLMGoat/models
      - ${HOME}/.LLMGoat/cache:/root/.LLMGoat/cache # To avoid downloading each time the model for challenge A09      
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities:
                - gpu