I’ve had a few jobs die without any information on my wandb log of what happened. I wonder if there is some logging level that is implicitly set that doesn’t allow me to see why it ended at X epochs randomly rather than the end.
Is there a way to set it to print everything?
perhaps there is a way to debug this with the extra system stuff wandb already logs?
Hey @brando, apologies about the lack of updates on this. At the moment there is no logging level that does what you want. I have filed a ticket for the engineering team to consider adding the functionality.