Adrian Wälchli comments

Results 399 comments of


                                            Adrian Wälchli

Fix device placement when `.cuda()` called without specifying index

@nithinraok Yes, if merged, this would go into 1.7.2 next week (~Tuesday).

Update changelog after 1.7.1 release

Gatekeeper merge

RFC: Make subprocess DDP the default when selecting multiple devices

> ddp-spawn works virtually everywhere (except for interactive envs), right? Yes, and so does ddp afaik because all it requires is that the script can launch itself as `python train.py...

Align ddp and ddp-spawn strategies in setting up the environment

The status is that I'm currently working on this and once #11163 is merged this will pretty much be ready.

Align ddp and ddp-spawn strategies in setting up the environment

@carmocca I would say quite important for #14707. Yes, I can integrate this into the two PRs. The PR was ready long before starting with Lite, I should have pushed...

Reload new data after one epoch doesnt work.

Hi @KralaBenjamin You are passing dataloaders directly to fit, but then wish to exchange them later on with a datamodule. This is not supported and not recommended. I can sketch...

Update tests that mock cuda availability and extend it for mps

Thanks @JustinGoheen. Let me know if you have any questions.

return support for torch 1.8.2 (LTS) in PL 1.7.*

@paantya Are you currently seeing errors being raised when using PyTorch 1.8 with Lightning 1.7?

[App] Moved app.py to root dir for `lightning init app <app_name>` template

Adding an entry in the changelog would be nice if you can :)

Feature request: Log optimizer and LR Scheduler hyperparams when set by LightningCLI

> A big limitation of save_hyperparameters is that it only has what goes in __init__ I wouldn't call this a limitation. You wouldn't say a big limitation of a car...