# Copyright (C) 2025 Intel Corporation
# SPDX-License-Identifier: Apache-2.0

services:
  vllm-service:
    image: ${REGISTRY:-opea}/vllm-gaudi:${TAG:-latest}
    container_name: vllm-gaudi-server
    ports:
      - "8086:8000"
    volumes:
      - ${HF_CACHE_DIR}:/data
    environment:
      no_proxy: ${no_proxy}
      http_proxy: ${http_proxy}
      https_proxy: ${https_proxy}
      HF_TOKEN: ${HF_TOKEN}
      HUGGING_FACE_HUB_TOKEN: ${HF_TOKEN}
      HF_HOME: /data
      HABANA_VISIBLE_DEVICES: all
      OMPI_MCA_btl_vader_single_copy_mechanism: none
      LLM_MODEL_ID: ${LLM_MODEL_ID}
      VLLM_TORCH_PROFILER_DIR: "/mnt"
      VLLM_SKIP_WARMUP: true
      PT_HPU_ENABLE_LAZY_COLLECTIVES: true
    healthcheck:
      test: ["CMD-SHELL", "curl -f http://$host_ip:8086/health || exit 1"]
      interval: 10s
      timeout: 10s
      retries: 100
    runtime: habana
    cap_add:
      - SYS_NICE
    ipc: host
    command: --model $LLM_MODEL_ID --tensor-parallel-size 4 --host 0.0.0.0 --port 8000 --max-seq-len-to-capture $MAX_LEN