Qwen3.6-27B 模型效果实测

modelscope download --model Qwen/Qwen3.6-27B --local_dir /data/Qwen3.6-27B

services:
  qwen35:
    container_name: Qwen3.6-27B
    image: vllm/vllm-openai:v0.20-cu130
    restart: always

    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

    ipc: host
    shm_size: 64gb

    ports:
      - "8000:8000"

    volumes:
      - /data/Qwen3.6-27B/:/models/Qwen3.6-27B

    command: >
      /models/Qwen3.6-27B
      --served-model-name Qwen3.6-27B
      --host 0.0.0.0
      --port 8000
      --max-model-len 262144
      --tensor-parallel-size 2
      --gpu-memory-utilization 0.85
      --max-num-seqs 8 
      --enable-prefix-caching
      --enable-chunked-prefill
      --trust-remote-code
      --enable-auto-tool-choice
      --tool-call-parser qwen3_coder
      --performance-mode throughput
      --language-model-only
      --kv-cache-dtype auto
      --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":4}'

    logging:
      driver: "json-file"
      options:
        max-size: "100m"
        max-file: "3"