Getting model with their configs pushed to the hub

#1
by Leyo - opened

Hello,
The models seem very promising and I would love to use them! However, the checkpoints are not paired with their configs, and not individually pushed to the hub at the moment, which makes it harder to get access to them.
Let me know if I can help you to have them pushed on the hub or if they are already somewhere I'm not aware of. I think it would facilitate their use by the community!

Hi Leyo. Thank you for your interest in our work. I'll appreciate it if you could help me with that.

Hi QuanSun,
Sorry for the late reply. I'm currently working on converting the biggest models. They are simpler to deal with because they don't use rope nor swiglu. This makes them closer to the CLIP architecture and therefore easier to adapte to an hf model. I'll send you the code as soon as it's ready. Then if you want to include the smaller models as well, you will just need to make a few adaptations.

QuanSun changed discussion status to closed

Sign up or log in to comment