# MiroThinker-v1.0-30B-FP8 Docker Compose
#
# Usage (TP=2, recommended):
#   docker compose up
#
# Usage (single GPU, not recommended):
#   SINGLE_GPU=1 docker compose up

services:
  mirothinker:
    image: vllm/vllm-openai:v0.11.2
    ports:
      - "8000:8000"
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - HF_HOME=/root/.cache/huggingface
      # Set SINGLE_GPU=1 for single GPU mode (poor performance)
      - SINGLE_GPU=${SINGLE_GPU:-}
    volumes:
      # Cache downloaded models
      - hf_cache:/root/.cache/huggingface
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    shm_size: '16gb'
    command: >
      sh -c '
        if [ -n "$SINGLE_GPU" ]; then
          echo "WARNING: Single GPU mode - expect ~1-2 tok/s, 2K context max";
          python -m vllm.entrypoints.openai.api_server
            --model Doradus/MiroThinker-v1.0-30B-FP8
            --host 0.0.0.0
            --port 8000
            --tensor-parallel-size 1
            --max-model-len 2048
            --max-num-seqs 4
            --gpu-memory-utilization 0.95
            --enforce-eager
            --trust-remote-code;
        else
          python -m vllm.entrypoints.openai.api_server
            --model Doradus/MiroThinker-v1.0-30B-FP8
            --host 0.0.0.0
            --port 8000
            --tensor-parallel-size 2
            --max-model-len 32768
            --gpu-memory-utilization 0.90
            --trust-remote-code
            --enable-chunked-prefill;
        fi
      '
    healthcheck:
      test: ["CMD", "curl", "-sf", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 120s

volumes:
  hf_cache: