Carlos Mocholí

> With LightningCLI it might make sense to extend SaveConfigCallback to do the logging of the hyperparameters. Yes, we could add it to the end of its `setup` hook: ```python...

Do these changes also suggest we should do some refactor with the `_LiteModule`? Now that the APIs match.

These new images will need to be pushed to the hub before merging this PR, or you will block master until the nightly job runs

> Do you think we should let pytest treat `DeprecationWarning` as errors just like `FutureWarning`? I tried it in the past and there's a TON of overrides we would need...

@akihironitta is this done?

The commits are not correct for rebasing where backwards compatibility is kept. @akihironitta Do you think you could help the author?

A user in Slack got confused by this > How do I do mixed precision training (CPU / GPU)? [...] Allowed precision values: ('16', '32', '64', 'bf16', 'mixed') suggest that...

> My suggestion for the API would be to have arguments... @wisecornelius I think that is very correct and concise but prone to errors from inexperienced users who don't understand...

Trying to revive this discussion, here's what I find most logical | Flags Before | Precision After | |:----------------------------------:|:-----------------:| | precision=16 | "16-mixed" | | precision=64 | "64-true" (true) |...

I prefer the simplicity of the string options, they are consistent with our other registry based trainer arguments. Also, passing a dict is very similar to passing separate arguments, and...