# ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF
# docker compose up -d   (the first start builds Strata 0.1.45 from GitHub, then downloads the weights and sets up)
# As measured: add "answer_room": false to /data/config/strata-coder-iq1_m.json after the first start, then docker compose restart
name: bench_qwen38_flash_next_coder
services:
  engine:
    build:
      context: https://github.com/Niko1221/Strata.git#v0.1.45
    image: strata:0.1.45
    container_name: bench-model
    gpus: all
    ipc: host
    ports:
    - 127.0.0.1:8000:8000
    ulimits:
      memlock: -1
    stop_grace_period: 60s
    environment:
      CUDA_VISIBLE_DEVICES: ${CUDA_VISIBLE_DEVICES:-0}
      FAMILY: coder
      MODEL: IQ1_M
      PORT: '8000'
      CONTEXT: '262144'
      VISION: gpu
      KV: int8
    volumes:
    - strata-data:/data
    healthcheck:
      test:
      - CMD
      - curl
      - -fsS
      - http://127.0.0.1:8000/health
      interval: 15s
      timeout: 5s
      retries: 60
      start_period: 7200s
    restart: 'no'
volumes:
  strata-data: {}