geronimi73

> Thanks @geronimi73 the changes look great ! can you just run the styling checks? `make precommit` , then we can merge what about the `print` statements I mentioned? leave...

> you will need to import `logging` at the top of the file and import the logger properly i tried. it works but not sure if this is the way...

> Hi @geronimi73 sorry for all the iteration ! can you re-run the styling checks again? 🙏 sure! but i'm still wondering whether it works correctly. Problem is, I never...

> > Hi @geronimi73 sorry for all the iteration ! can you re-run the styling checks again? 🙏 > > sure! but i'm still wondering whether it works. I never...

sorry to keep bothering you @younesbelkada but I think the problem ist that we are using logging from `transformers.utils` and obtaining a logger with `__name__` (=`trl.trainer.sft_trainer`). This means that setting...

Hello everyone! I'm also experimenting with the Galore optimizer and the loss curves look great! But I don't see any benefit in memory usage, Galore even uses a bit more....

+1

sorry to comment on this closed issue but I still have issues with FA2 1. loss is different with FA2 compared to without 2. loss is different even between two...

> We want to investigate it further more and attempt to find the root of the problem. We are doing a line-by-line comparison between the new model's code and the...