ker2xu

Thanks for your reply! I will check those. From: ashleve ***@***.***> Sent: 2022年8月17日 20:27 To: ashleve/lightning-hydra-template ***@***.***> Cc: ker2xu ***@***.***>; Author ***@***.***> Subject: Re: [ashleve/lightning-hydra-template] Multi-GPUs without DDP for hparams_search...

> i don't remember 100% but i believe i based my implementation on the scipy and/or tensorflow-probability convention... I remembered that PyTorch's implementation of distributions module was based on tensorflow....

@kashif I fully understand your concerns and this can be break many models. Honestly speaking, the original documentation and the implemented code even do not match well and are quite...

> so note that we are in the pytorch distributions tests, testing against the scipy implementation for all the methods like log-prob etc. and they match (after changing the params...

> Did you use an automatic linter for all markdown files? Because I see a lot of spurious changes around spaces etc. Maybe raise a smaller PR with just the...

If there is a method that can recover those spaces back easily without bringing typos back, I would like to adopt it. But those trimmed lines have already been saved...

> Hi @ker2xu, > > Thank you for your PR. I want to apologize and take responsibility for many of these errors that were introduced by me during documentation development....

@anistark Conflicts resolved. Thanks for your suggestion and it is definitely better to correct typos and formatting issues in separate PRs. However, I am not available recently and separating these...