Nerogar / OneTrainer

OneTrainer is a one-stop solution for all your stable diffusion training needs.
GNU Affero General Public License v3.0
1.81k stars 154 forks source link

Provide wrapper scripts to start tensorboard manually #575

Closed clayne closed 1 week ago

clayne commented 1 week ago
Nerogar commented 1 week ago

Hard coding the workspace directory isn't very useful. That's a parameter that can be changed, so it would only work with default settings.

It could be possible to load the #.json config file and extract the workspace directory. But even that won't be very useful. It only works if the user is using the UI, in which case it makes more sense to just add a button (or modify the existing button).

Another option would be to create a script similar to scripts/train.py that takes a config file as a parameter. But at that point you could just call tensorboard directly. I'm not really sure if there is a good solution that's easy to use and still useful.

clayne commented 1 week ago

Yeah, I was kinda iffy on that directory being assumed like that. My justification for the script was getting tired of having to run it manually just to figure out what my last training results were after a run - as they aren't available when the training finishes. I'm totally fine with not merging this as it's pretty much just a Nice To Have. Maybe a better fix would be to allow tensorboard to persist after training completes?

clayne commented 1 week ago

Closing because of the hardcoded workspace directory, which I agree isn't ready for prime time.