Justin Goodwin

I think, for now, we should provide this functionality as is. It provides a useful functionality that probably won't be used that much. > - Can the wrapper be designed...

Had to throw this at you, the interface made me try it 😵‍💫 ```python Module = builds(...) # lightning module LightningTrainer = like(Trainer)(gpus=2, strategy="ddp") LightningFit = just(LightningTrainer.fit(Module)) # automatically start...

Sorry for the delayed response. I'm finally going through your example here and see what you are trying to do. I think this is the equivalent using `like`: ```python from...

I'm wondering now if your use case would work with this approach. There would be a configuration for `Trainer` and a `FineTuner` now. So would the following work as expected?...

Interesting use case. The configuration for `Car` cannot be instantiated because there is no default value provided for `Car`. ```python >>> instantiate(builds(Car, populate_full_signature=True)) ... MissingMandatoryValue: Missing mandatory value: driver full_key:...

Oh, just playing around with this idea. I'm guessing you'd want swapable configs for `Driver` (and `Car`). So you might want to do this instead: ```python if __name__ == "__main__":...

Sorry, my comments were mostly me thinking out loud on the issue. I was definitely intrigued and frustrated that there didn't seem to be an approach to provide configuration information...

@tchaton I'm working through some issues with `teardown`. It appears there are issues with the process hanging on `torch.distributed.destroy_process_group()`, but if I do it in my task function outside of...

Just a heads up to all, I've been on vacation for the past week. I'll look at all the changes and try to respond to the comments this week.

So I'm attempting to revive this PR. I got everything running but there are still some outstanding issues to address. The main issues are: - Hydra now comes with `version_base`,...