# Paquet de déploiement CPU Ingenium Expert LMS 3B Instruct — candidat Ollama.
#
# Variante de docker-compose.yml qui remplace vLLM/GPU par Ollama/CPU. Le
# harnais reste l'unique point d'entrée public (127.0.0.1:8080) ; Ollama
# reste sur le réseau interne du compose et n'est jamais joignable
# directement.
#
# Usage :
#   GGUF_PATH=./ingenium-expert-lms-3b-instruct-grounded-v0.1.0-Q8_0.gguf docker compose -f docker-compose.cpu.yml up --build
#
#   curl -X POST http://127.0.0.1:8080/ask \
#     -H "Content-Type: application/json" \
#     -d '{"question": "Comment créer un utilisateur ?", "context": {...context-v1...}}'

services:
  ollama:
    build:
      context: .
      dockerfile: docker/ollama/Dockerfile
    environment:
      MODEL_NAME: ingenium-expert-lms-3b-instruct
    volumes:
      - ${GGUF_PATH:?GGUF_PATH requis}:/model/model.gguf:ro
      - ./deploy/ollama/Modelfile.ingenium-3b:/model/Modelfile:ro
    healthcheck:
      test: ["CMD-SHELL", "ollama list | grep -q \"$$MODEL_NAME\""]
      interval: 10s
      timeout: 5s
      retries: 60
      start_period: 120s

  harness:
    build:
      context: .
      dockerfile: harness/Dockerfile
    environment:
      VLLM_BASE_URL: http://ollama:11434/v1
      MODEL_NAME: ingenium-expert-lms-3b-instruct
      MAX_NEW_TOKENS: "128"
    ports:
      - "127.0.0.1:8080:8080"
    depends_on:
      ollama:
        condition: service_healthy