Closed AlfaranoAndrea closed 2 years ago
How could we re-produce this issue?
i found what was the issue: i built a 1.12.0 version of pytorch following your guide (https://github.com/xuhuisheng/rocm-build/tree/master/gfx803). I went back to your built torch 1.11.0 and torchvision don't give anymore this error. thanks!
Hi First of all thanks for all the support that do you give us! i'm trying to train in multigpu a torchvision project, and i'm getting this issue:
/home/fi/.local/lib/python3.8/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: /home/fi/.local/lib/python3.8/site-packages/torchvision/image.so: undefined symbol: _ZNK3c1010TensorImpl36is_contiguous_nondefault_policy_implENS_12MemoryFormatE
this led to GPU memory errors, with frequently computer freezing and shut down
Maybe rebuild torchvision from source can be a nice try?