30b version

#1
by ehartford - opened

30b version?
I could help train it, I have compute

Owner

Hi @ehartford !

Apologies for the late response. This got buried in my emails and notifications.

This model was actually finetuned by the Anole (GAIR) team which I just converted to be compatible with Transformers. Here is the original model repo: https://huggingface.co/GAIR/Anole-7b-v0.1 . You can also track my progress on improve native support of Chameleon's image generation & interleaved generation modes here: https://github.com/huggingface/transformers/pull/32013

As for the 30B version, we can absolutely collaborate on it! I'm also working on a loss function that speeds up finetuning by up to 20%. It's not groundbreaking, but it's paper-worthy already--I think. Let's continue thru mail here: franzlouiscesista@gmail.com

Sign up or log in to comment