Carlos Mocholí comments

Results 427 comments of


                                            Carlos Mocholí

Add TPU v5 to CI

Closing since we don't have quota and this is not a priority

integration with new low lever _KinetoProfile api

Hi @guotuofeng! This PR is quite large and has no linked issue or discussion. Can you describe the advantages and the changes done? Also, is this blocked on something or...

Resolve collectives test hang

Thanks Ota! Doing great here, hope you are good too. We never continued this feature and the tests are skipped. I'll close this in the meantime

Fix `BatchSizeFinder` leaving model in train state

I would suggest calling `_scale_batch_size` directly with some unittest Mocks, and assert that the original training state is preserved

calling iter twice messes up dataloaders with queues

I didn't look too deeply. Couldn't we check `restarting` too for the `FitLoop`'s `iter` call? We have a lot of tests around this so If a solution passes them we...

ci: adding testing with M1 [1/2: without Fabric] [wip]

Those tests don't seem too big. Maybe there's a memory leak making later tests fail. It will need some research

PyTorch Lightning DDP crashes with unused parameters

2.0 with `ddp_find_unused_parameters_true` should match the speed of

PyTorch Lightning DDP crashes with unused parameters

Slowness might be caused by https://github.com/Lightning-AI/lightning/issues/17725. You can verify by setting `Trainer(logger=False)`

PyTorch Lightning DDP crashes with unused parameters

@awaelchli Since an exception gets raised anyway, do you think we could hook into it and print the list of unused parameters? This seems to be a recurring issue with...

Input validation for Trainer's `min_time` and `max_time` when passed as string

@kylebgorman Would you like to contribute this suggestion?