Kernel crash in Colab and Kaggle during training

[BenjaminMidtvedt]

Sometimes the kernel will crash in Colab and Kaggle during training. This is an issue on lightning's end, and can be traced back to the progress bar refresh rate.

We have currently implemented safeguards to reduce the refresh rate if we detect that the code is run on either Colab or Kaggle. If these do not work correctly (perhaps because the manner we detect this no longer works due to changes on the provider's end), one can manually set the refresh rate using methods:

trainer = dl.Trainer(...)
trainer.tqdm_progress_bar(refresh_rate=10)

or by disabling the progress bar completely:

trainer = dl.Trainer(...)
trainer.disbale_progress_bar()

[giovannivolpe]

This issue is resolved in PR #144

[cmanzo]

Successfully ran code that was previously giving problems on Colab and Kaggle using CPU.