双机直连部署DeepSeek-V3.1-W8A8模型https://www.modelscope.cn/models/vllm-ascend/DeepSeek-V3.1-W8A8
镜像：https://docker.aityp.com/image/docker.io/vllm/vllm-openai:v0.10.2
运行报错：
RuntimeError: EI0006:[PID: 6317)2025-09-21-11:12:12.317.865 Getting socket times out, Reason: 1. The remote does not initiate a connect request, some NPus in the cluster are abnormal.
2.The remote does notinitiate aconnect request because the collective communication operator is started too late or is not startedby some NPU in the cluster.
3.The communication link is disconnected, (For example, the Ip addresses are not on the same network segment or the iLs configurations are inconsistent.)
查询了华为昇腾官网，hccl 通讯库异常原因为：npu 通讯和 tcl 规则不一致
这里我们检查过 npu 两两相连能够 ping 通，tcl 规则的 switch 值都为 0
后续考虑了更换 vllm 镜像，延长 hccl 超时时间， 使用 ray 集群，但都没有解决，麻烦帮忙看看
NPU之间可以互联，状态也都正常：
点赞
收藏
回复
分享
举报
已解决
双机直连部署DeepSeek-V3.1-W8A8模型https://www.modelscope.cn/models/vllm-ascend/DeepSeek-V3.1-W8A8
镜像：https://docker.aityp.com/image/docker.io/vllm/vllm-openai:v0.10.2
运行报错：
RuntimeError: EI0006:[PID: 6317)2025-09-21-11:12:12.317.865 Getting socket times out, Reason: 1. The remote does not initiate a connect request, some NPus in the cluster are abnormal.
2.The remote does notinitiate aconnect request because the collective communication operator is started too late or is not startedby some NPU in the cluster.
3.The communication link is disconnected, (For example, the Ip addresses are not on the same network segment or the iLs configurations are inconsistent.)
查询了华为昇腾官网，hccl 通讯库异常原因为：npu 通讯和 tcl 规则不一致
这里我们检查过 npu 两两相连能够 ping 通，tcl 规则的 switch 值都为 0
后续考虑了更换 vllm 镜像，延长 hccl 超时时间， 使用 ray 集群，但都没有解决，麻烦帮忙看看