1 Comment

Good luck with the Substack adventure. I hope it goes great!

Fwiw, we tried Lion to do some RLHF (because it's a little more memory efficient, could be useful to push the limits of non-distributed RLHF. Docs here: https://huggingface.co/docs/trl/customization#use-lion-optimizer

"TSMC is the Iron Bank" agree...

Expand full comment