老师您好如果您当前使用的加速卡型号为K100，其支持的vllm版本最高为085，您可以换用该版本的镜像尝试，例如：docker pull image.sourcefind.cn:5000/dcu/admin/base/vllm:0.8.5-ubuntu22.04-dtk25.04.1-rc5-das1.6-py3.10-20250724

你的模型文件是是不是没下全 还有你物理机操作系统是什么 docker 版本是多少

services: qwen3-32b-tool: image: image.sourcefind.cn:5000/dcu/admin/base/vllm:0.9.2-ubuntu22.04-dtk25.04.2-das1.7-py3.10-20251203 container_name: qwen3-32b-tool privileged: true devices: - /dev/kfd - /dev/dri - /dev/mkfd group_add: - video cap_add: - SYS_PTRACE security_opt: - seccomp:unconfined environment: - HIP_VISIBLE_DEVICES=0,1 - VLLM_WORKER_MULTIPROC_METHOD=spawn - OMP_NUM_THREADS=16 # - VLLM_USE_TRITON_FLASH_ATTN=True # - VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 volumes: - my-pip-cache:/root/.cache/pip - /ssddata/aimodel:/ssddata/aimodel - /opt/hyhal:/opt/hyhal:ro #- /HB/models/tool-call-parser:/HB/models/tool-call-parser # - /HB/models/dist-packages/vllm:/usr/local/lib/python3.10/dist-packages/vllm shm_size: '64g' # 共享内存调大 api-key sk-Wp1****************KjUpPZD86 \ # command: bash # tty: true # stdin_open: true command: bash -c "export ALLREDUCE_STREAM_WITH_COMPUTE=1 && \ export VLLM_NUMA_BIND=1 && \ export VLLM_RANK0_NUMA=0 && \ export VLLM_RANK1_NUMA=1 && \ export VLLM_RANK2_NUMA=2 && \ export VLLM_RANK3_NUMA=3 && \ vllm serve /ssddata/aimodel/Qwen3-8B \ --served-model-name Qwen3-8B \ --host 0.0.0.0 \ --port 8000 \ --max-model-len 32768 \ --reasoning-parser qwen3 \ --gpu-memory-utilization 0.95 \ --tool-call-parser hermes \ --enforce-eager \ --enable-auto-tool-choice \ --max-num-batched-tokens 32768 \ --tensor-parallel-size 2 \ --trust-remote-code" ports: - "8000:8000" networks: - vllm ulimits: nproc: 65535 nofile: soft: 65535 hard: 65535 volumes: my-pip-cache: networks: vllm: external: true

可以参考这个 git 仓库 https://github.com/hikafeng/dcu-vllm.git

Worker failed with error 'HIP error: invalid device function