300I PRO 4卡推理启动失败
收藏回复举报
300I PRO 4卡推理启动失败
t('forum.solved') 已解决
新人帖
发表于2025-07-12 14:22:40
0 查看

华为虚拟化FC上的虚拟机直通绑定了4块300I PRO卡,使用mindie:2.0.RC2-300I-Duo-py311-openeuler24.03-lts容器镜像运行Qwen32B,拉起服务时报错

cke_696.png

查看mindie/log/debug/mindie-batchscheduler*日志,报错如下:

RuntimeError: External Comm Manager: Create the hccl communication group failed. export ASDOPS_LOG_LEVEL=ERROR, export ASDOPS_LOG_TO_STDOUT=1 to see more details. Default log path is $HOME/atb/log. 

[2025-07-12 13:59:24.012+0800] [2984] [140318099883712] [batchscheduler] [ERROR] [model.py:65] : [MIE04E13030A] [Model] >>> return initialize error result: {'status': 'error', 'npuBlockNum': '0', 'cpuBlockNum': '0', 'memPoolId': '-1'} 

cke_4753.png

我要发帖子