tginart

Results 22 comments of tginart

Hi @joecummings Current use-case is just faster fine-tuning over larger datasets. I do have access to multi-node SLURM already, and have trained using other frameworks. For various reasons I've used...

Just wondering if this PR will be merged in soon? @calvinpelletier @felipemello1 @joecummings This is much needed functionality! Thank you!!! Btw, quick question, in terms of the ```max_steps_per_epoch``` hack --->...