ufoym/deepo:pytorch-py36-cu101 cuda.max_memory_allocated crash

[大地小神]

>>> import torch
>>> torch.__version__
'1.6.0.dev20200607+cu101'
>>> from torch.cuda import max_memory_allocated
>>> max_memory_allocated(0)
Segmentation fault (core dumped)

[大地小神]

it's my fault, i run docker without --runtime nvidia.

[大地小神]

reopen, it's not caused by missing the "--runtime nvidia", but caused by "torch" with version 1.6

torch1.5:

>>> import torch
>>> torch.__version__
'1.5.0.dev20200319'
>>> torch.version.cuda
'10.1'
>>> torch.cuda.max_memory_reserved(0)
0
>>> 

torch 1.6:

>>> import torch
>>> torch.__version__
'1.6.0.dev20200609+cu101'
>>> torch.version.cuda
'10.1'
>>> torch.cuda.max_memory_reserved(0)
Segmentation fault (core dumped)

[Ming]

Should be OK @ latest deepo images:

>>> import torch
>>> torch.__version__
'1.11.0.dev20211224+cu111'
>>> torch.version.cuda
'11.1'
>>> torch.cuda.max_memory_reserved(0)
'0'