infra/verification-loop/docker-compose.yml

version: "3.8"

services:
  granite-verifier:
    image: ${VLLM_IMAGE_TAG}
    container_name: lean_llm_granite_verifier
    environment:
      - HF_HUB_OFFLINE=1
      - VLLM_ATTENTION_BACKEND=FLASH_ATTN
      - PYTHONHASHSEED=42
      - TOKENIZERS_PARALLELISM=false
    volumes:
      - ${SOVEREIGN_MODEL_STORE}/granite-4.1-8b-instruct:/models/granite-4.1:ro
      - ${BIFROST_AUDIT_SINK}/vllm:/var/log/vllm:rw
    ports:
      - "127.0.0.1:8100:8000"
    ipc: host
    shm_size: "16g"
    command: >
      --model /models/granite-4.1
      --host 0.0.0.0
      --port 8000
      --max-model-len ${GRANITE_CONTEXT_LEN}
      --max-num-seqs 8
      --gpu-memory-utilization 0.92
      --served-model-name sovereign-granite-4.1-verifier
      --enforce-eager
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 120s
    restart: unless-stopped

  lean4-kernel:
    image: ${LEAN4_IMAGE_TAG}
    container_name: lean_llm_lean4_kernel
    user: "1000:1000"
    read_only: true
    security_opt:
      - no-new-privileges:true
    cap_drop:
      - ALL
    tmpfs:
      - /tmp:exec,size=2g
    volumes:
      - ${VERIFICATION_PROJECT_ROOT}:/workspace/project:ro
      - ${SOVEREIGN_CACHE_STORE}/lake:/home/lean/.lake:rw
      - ${BIFROST_AUDIT_SINK}/lean4:/workspace/artifacts:rw
    working_dir: /workspace/project/lean4
    entrypoint: ["/bin/bash", "-c", "lake build && lake exe verify"]
    deploy:
      resources:
        limits:
          cpus: "4"
          memory: 8G

networks:
  default:
    name: lean_llm_verification_loop
    driver: bridge
    internal: true