AlexWertheim

Thanks for bringing this to our attention - I'll take a look. Could you please confirm that [this](https://github.com/pytorch/xla/blob/master/contrib/colab/multi-core-alexnet-fashion-mnist.ipynb) is the notebook to which you're referring?

Thanks very much for your data. I've reproduced your issue - right now, each core takes ~320 seconds on average to run. This is consistent across previous versions of PyTorch/XLA....

Thanks for your patience. At the moment, it appears that this is likely a colab issue, not a TPU or PyTorch/XLA issue. Running the same code natively on TPU VM,...

Indeed, I was able to reproduce the issue on my side - see the comment preceding my most recent comment - so this isn't anything related to your Colab account....

Sure, I can take a look!

> @AlexWertheim With your recent pr can we call this request done? Yeah, I think so. For reference, the PR in question can be seen [here](https://github.com/huggingface/transformers/pull/21406). @muellerzr can say better...

@alanwaketan nope, looks good to me. I'll work with @Liyang90 and @ronghanghu to collect information, but I can be ultimately responsible for updating the documentation. Thanks!