docker-compose.gpu.yml

# Override para servir un LLM REAL con vLLM (requiere GPU NVIDIA + NVIDIA Container Toolkit).
#   1. En .env:  LLM_PROVIDER=vllm   y   LLM_BASE_URL=http://vllm:8000/v1
#   2. docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d
#
# Añade el servidor vLLM oficial (API compatible con OpenAI). El gateway `llm` le habla.

services:
  vllm:
    image: vllm/vllm-openai:latest
    command: >
      --model mistralai/Mistral-7B-Instruct-v0.2
      --max-model-len 8192
    environment:
      HUGGING_FACE_HUB_TOKEN: ${HUGGING_FACE_HUB_TOKEN:-}
    volumes:
      - hf_cache:/root/.cache/huggingface
    ports:
      - "8002:8000"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

  # El gateway apunta al servidor vLLM
  llm:
    environment:
      LLM_PROVIDER: vllm
      LLM_BASE_URL: http://vllm:8000/v1
      LLM_MODEL: mistralai/Mistral-7B-Instruct-v0.2
    depends_on:
      - vllm

volumes:
  hf_cache: