官方文档是在线安装的将在线安装的rpm包缓存后做成一个安装包将镜像下载后保存就可以完成离线安装。echo mkdir -p /software/vllmcd /software/vllmmkdir -p /yumrepocp vllm-rhaiis-3.3.0-rhel-9.6.tar /yumrepocd /yumrepotar -xvf vllm-rhaiis-3.3.0-rhel-9.6.tarcd /etc/yum.repos.dmv * /mntcat EOF /etc/yum.repos.d/vllm.repo[vllm]namevllmbaseurlfile:///yumrepo/vllmenabled1gpgcheck0EOFmkdir -p /isomkdir -p /software/osmv /software/vllm/rhel-9.6-x86_64-dvd.iso /software/osls -l /software/osmount /software/os/rhel-9.6-x86_64-dvd.iso /isocat EOF /etc/yum.repos.d/rhel9.repo[BaseOS]nameBaseOSbaseurlfile:///iso/BaseOSenable1gpgcheck0[APP]nameAPPbaseurlfile:///iso/AppStreamenable1gpgcheck0EOFyum clean allyum repolistyum search httpdecho return workplacecd /workcd /yumrepo/vllmdnf install *.rpm -yrpm -qa | grep nvidia-drivernvidia-smirebootnvidia-smicd /software/vllmcp model-vllm-Qwen2.5-0.5B-Instruct.tar /cd /tar -xvf model-vllm-Qwen2.5-0.5B-Instruct.tarcd /mkdir -p rhaiis-cache chmod grwX rhaiis-cachecd /software/vllmpodman load -i vllm-cuda-rhel9-3.3.0.tarpodman run --rm -it \--device nvidia.com/gpuall \--security-optlabeldisable \--shm-size4g -p 8000:8000 \--group-addvideo --group-addrender \--env HUGGING_FACE_HUB_TOKEN$HF_TOKEN \--env HF_HUB_OFFLINE1 \--env VLLM_LOGGING_LEVELDEBUG \-v /model/vllm:/model-input:Z \-v /rhaiis-cache:/opt/app-root/src/.cache:Z \registry.redhat.io/rhaiis/vllm-cuda-rhel9:3.3.0 \--model /model-input--tensor-parallel-size 1可运行在T4机器上------关键日志----------------------官方文档是在线安装的将在线安装的rpm包缓存后做成一个安装包将镜像下载后保存就可以完成离线安装。echo mkdir -p /software/vllmcd /software/vllmmkdir -p /yumrepocp vllm-rhaiis-3.3.0-rhel-9.6.tar /yumrepocd /yumrepotar -xvf vllm-rhaiis-3.3.0-rhel-9.6.tarcd /etc/yum.repos.dmv * /mntcat EOF /etc/yum.repos.d/vllm.repo[vllm]namevllmbaseurlfile:///yumrepo/vllmenabled1gpgcheck0EOFmkdir -p /isomkdir -p /software/osmv /software/vllm/rhel-9.6-x86_64-dvd.iso /software/osls -l /software/osmount /software/os/rhel-9.6-x86_64-dvd.iso /isocat EOF /etc/yum.repos.d/rhel9.repo[BaseOS]nameBaseOSbaseurlfile:///iso/BaseOSenable1gpgcheck0[APP]nameAPPbaseurlfile:///iso/AppStreamenable1gpgcheck0EOFyum clean allyum repolistyum search httpdecho return workplacecd /workcd /yumrepo/vllmdnf install *.rpm -yrpm -qa | grep nvidia-drivernvidia-smirebootnvidia-smicd /software/vllmcp model-vllm-Qwen2.5-0.5B-Instruct.tar /cd /tar -xvf model-vllm-Qwen2.5-0.5B-Instruct.tarcd /mkdir -p rhaiis-cache chmod grwX rhaiis-cachecd /software/vllmpodman load -i vllm-cuda-rhel9-3.3.0.tarpodman run --rm -it \--device nvidia.com/gpuall \--security-optlabeldisable \--shm-size4g -p 8000:8000 \--group-addvideo --group-addrender \--env HUGGING_FACE_HUB_TOKEN$HF_TOKEN \--env HF_HUB_OFFLINE1 \--env VLLM_LOGGING_LEVELDEBUG \-v /model/vllm:/model-input:Z \-v /rhaiis-cache:/opt/app-root/src/.cache:Z \registry.redhat.io/rhaiis/vllm-cuda-rhel9:3.3.0 \--model /model-input--tensor-parallel-size 1可运行在T4机器上