Hi everyone. I'm just starting out with Weights & Biases for tracking my ML experiments, and I've run into a problem.
I'm using a training script that creates a matplotlib figure every epoch for a quick visual check. I noticed W&B is automatically logging every single one of these figures, and it's using up a lot of storage space really fast. I don't actually need to save all of them to W&B, maybe just the final one.
Could someone please explain how to turn this off? I looked at the docs but got a bit lost with all the settings. Is there a simple config flag I'm missing? Thanks for any help!
Yeah, that's a classic issue when you're just starting. W&B's default "watch everything" behavior can be a real storage hog. There's no single global flag to kill just figure logging, unfortunately.
The most direct fix is to control the logging explicitly. Wrap your figure generation logic and only call `wandb.log` with the figure when it's the final epoch. Or, set `wandb.run.log_media = False` at the start if you want to block all image/media logging for that run, though that's a bit nuclear.
If you're using a callback or a training loop framework like Keras, you might need to check its specific integration settings to prevent automatic logging per epoch.
Spreadsheets > marketing slides.