CMVR-AI-ANALYSIS/deploy/compose.sglang.yaml

services:
  qwen38-fast:
    container_name: cmvr-qwen38-fast
    image: quay.io/gpustack/runner:cuda13.0-sglang0.5.15.post1
    restart: unless-stopped
    ipc: host
    shm_size: 32gb
    ports:
      - "192.168.28.10:14081:30000"
    environment:
      SGLANG_USE_MODELSCOPE: "true"
      MODELSCOPE_CACHE: /root/.cache/modelscope
      MODELSCOPE_DOWNLOAD_PARALLEL_WORKERS: "4"
    volumes:
      - ../models/modelscope:/root/.cache/modelscope
    gpus:
      - driver: nvidia
        device_ids: ["2"]
        capabilities: [gpu]
    command:
      - python
      - -m
      - sglang.launch_server
      - --model-path
      - Qwen/Qwen3.8-27B-FP8
      - --served-model-name
      - Qwen/Qwen3.8-27B-FP8
      - --host
      - 0.0.0.0
      - --port
      - "30000"
      - --tp-size
      - "1"
      - --context-length
      - "65536"
      - --kv-cache-dtype
      - fp8_e4m3
      - --mem-fraction-static
      - "0.80"
      - --reasoning-parser
      - qwen3

  qwen38-accurate:
    container_name: cmvr-qwen38-accurate
    image: quay.io/gpustack/runner:cuda13.0-sglang0.5.15.post1
    restart: unless-stopped
    ipc: host
    shm_size: 32gb
    ports:
      - "192.168.28.10:14082:30000"
    environment:
      SGLANG_USE_MODELSCOPE: "true"
      MODELSCOPE_CACHE: /root/.cache/modelscope
      MODELSCOPE_DOWNLOAD_PARALLEL_WORKERS: "4"
    volumes:
      - ../models/modelscope:/root/.cache/modelscope
    gpus:
      - driver: nvidia
        device_ids: ["3"]
        capabilities: [gpu]
    command:
      - python
      - -m
      - sglang.launch_server
      - --model-path
      - Qwen/Qwen3.8-27B
      - --served-model-name
      - Qwen/Qwen3.8-27B
      - --host
      - 0.0.0.0
      - --port
      - "30000"
      - --tp-size
      - "1"
      - --context-length
      - "65536"
      - --kv-cache-dtype
      - fp8_e4m3
      - --mem-fraction-static
      - "0.84"
      - --reasoning-parser
      - qwen3