# Production stack: llama.cpp server with a Llama-3-8B GGUF model
# alongside the PII redactor service. Both speak HTTP.
#
# Usage:
#   1. Place a Llama-3-8B-Instruct GGUF in ./models/
#      e.g. Meta-Llama-3-8B-Instruct.Q4_K_M.gguf (~4.6 GB)
#      Available from: hf.co/QuantFactory/Meta-Llama-3-8B-Instruct-GGUF
#   2. cp .env.example .env  and fill in PIIR_API_KEY + PIIR_AUDIT_KEY
#   3. docker compose up -d
#   4. Test: curl localhost:8000/health
#
# Resource expectations:
#   - llama-server: ~8GB RAM, faster with GPU passthrough
#   - pii-redactor: ~200MB RAM, CPU only

services:
  llama-server:
    image: ghcr.io/ggerganov/llama.cpp:server
    container_name: pii-llama-server
    command: >
      -m /models/Meta-Llama-3-8B-Instruct.Q4_K_M.gguf
      --host 0.0.0.0
      --port 8080
      --ctx-size 8192
      --n-gpu-layers 0
    volumes:
      - ./models:/models:ro
    ports:
      - "8080:8080"
    healthcheck:
      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 60s
    restart: unless-stopped

  pii-redactor:
    build:
      context: ..
      dockerfile: docker/Dockerfile
    container_name: pii-redactor-api
    environment:
      PIIR_BACKEND: llama_cpp
      PIIR_LLAMA_CPP_URL: http://llama-server:8080
      PIIR_FAIL_ON_LLM_ERROR: "true"
      PIIR_REQUIRE_PRODUCTION_SAFETY: "true"
      PIIR_REQUIRE_API_KEY: "true"
      PIIR_API_KEY: ${PIIR_API_KEY:?Set PIIR_API_KEY}
      PIIR_REIDENTIFY_API_KEY: ${PIIR_REIDENTIFY_API_KEY:?Set PIIR_REIDENTIFY_API_KEY}
      PIIR_AUDIT_KEY: ${PIIR_AUDIT_KEY:?Set PIIR_AUDIT_KEY}
      PIIR_AUDIT_PATH: /data/audit.jsonl
      PIIR_AUDIT_ENABLED: "true"
      PIIR_MAX_TEXT_CHARS: ${PIIR_MAX_TEXT_CHARS:-200000}
      PIIR_MAX_BATCH_DOCS: ${PIIR_MAX_BATCH_DOCS:-1000}
      PIIR_MAX_CONCURRENCY: ${PIIR_MAX_CONCURRENCY:-8}
      PIIR_PLACEHOLDER_STYLE: numbered
    volumes:
      - ./data:/data
    ports:
      - "8000:8000"
    healthcheck:
      test: ["CMD-SHELL", "python -c \"import urllib.request; urllib.request.urlopen('http://localhost:8000/health', timeout=5).read()\" || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 30s
    depends_on:
      llama-server:
        condition: service_healthy
    restart: unless-stopped