# A local laya-serve built from upstream's Dockerfile (CPU torch), pinned to a known commit.
# Bump the ref deliberately and rerun the integration tests when upgrading.
#
#   docker compose up -d --wait        # first build takes a while
#   LAYA_URL=http://localhost:8000 composer test:integration
#
# The first request per checkpoint downloads its weights (~1 GB) into the model-cache volume.
# Set LAYA_PRELOAD=1 to download and load them at startup instead.
services:
  laya-serve:
    build:
      context: https://github.com/NandhaKishorM/laya.git#8c42cf35127448ad4b2dfc011ee087a9f134488e
    command: ["laya-serve"]
    ports:
      - "127.0.0.1:${LAYA_PORT:-8000}:8000"
    environment:
      LAYA_DEVICE: cpu
      LAYA_PRELOAD: "${LAYA_PRELOAD:-0}"
      LAYA_API_KEY: "${LAYA_API_KEY:-}"
      OMP_NUM_THREADS: "${OMP_NUM_THREADS:-4}"
    volumes:
      - model-cache:/home/laya/.cache/huggingface
    init: true
    healthcheck:
      test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8000/health', timeout=4)"]
      interval: 30s
      start_period: 10m
      start_interval: 2s

volumes:
  model-cache: