报错信息
File "/usr/local/lib/python3.10/dist-packages/torch/__init__.py", line 236, in <module>
from torch._C import * # noqa: F403
ImportError: /opt/hpcx/ucc/lib/libucc.so.1: undefined symbol: ucs_mpool_params_reset
解决:
sudo apt purge hwloc-nox libhwloc-dev libhwloc-plugins
docker迁移后又一次报相同错
解决:
export LD_LIBRARY_PATH=/opt/hpcx/ucx/lib:$LD_LIBRARY_PATH