Example on training with TPU does not run at all

[tungts1101]

I am currently try this colab notebook https://colab.research.google.com/github/PytorchLightning/lightning-tutorials/blob/publication/.notebooks/lightning_examples/mnist-tpu-training.ipynb#scrollTo=2772a2e1 provided by PL teams to get some experience with TPU training. But when I try to execute the third cell, there is some Import error with the _XLAC module.

[Programmer-RD-AI]

hi,
the problem seems to be with torch_xla,

try

!pip install cloud-tpu-client==0.10 https://storage.googleapis.com/tpu-pytorch/wheels/torch_xla-1.9-cp37-cp37m-linux_x86_64.whl

!pip install pytorch-lightning

if you have cuda 10.2
!pip3 install torch torchvision torchaudio

if you have cuda 11.1
!pip3 install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html

and you can add an issue to https://github.com/pytorch/xla

which is the library which seems to have the problem can you try these.

With best regards,
Ranuga

[Programmer-RD-AI]

hi, the problem seems to be with torch_xla,

try

!pip install cloud-tpu-client==0.10 https://storage.googleapis.com/tpu-pytorch/wheels/torch_xla-1.9-cp37-cp37m-linux_x86_64.whl

!pip install pytorch-lightning

if you have cuda 10.2 !pip3 install torch torchvision torchaudio

if you have cuda 11.1 !pip3 install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html

and you can add an issue to https://github.com/pytorch/xla

which is the library which seems to have the problem can you try these.

With best regards, Ranuga

I will check the notebook and try to resolve it

With best regards,
Ranuga

[Programmer-RD-AI]

!pip3 install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html

Check this issue also
https://stackoverflow.com/questions/68846290/colab-pytorch-importerror-usr-local-lib-python3-7-dist-packages-xlac-cpytho

[Programmer-RD-AI]

hi,

I found another issue is similar

check the issue below
pytorch/xla#2259

it seems to have the fix for your issue

With best regards,
Ranuga

[Programmer-RD-AI]

check this issue also

facebookresearch/maskrcnn-benchmark#891

[akihironitta]

@tungts1101 Hi, please have a look at #8315.

[kaushikb11]

Hi @tungts1101! this error is raised when the PyTorch and PyTorch xla are not of the same versions.
You could verify using pip list | grep torch and could install the latest versions for both!