Upcoming model?

[SinanAkkoyun]

We are working on a more capable model that is pretrained with the data at a larger scale. Stay tuned!

Hello again! :)
What parameter size will this model be? And will it be based on > v0 vicuna?

Thank you very much for this awesome work!!!

[haotian-liu]

Hi @SinanAkkoyun, thank you for your interest in our work.

This is WIP, and we haven't make every decision choices yet. As of today, we build future LLaVA models based on Vicuna-v1 checkpoints. We are also exploring to improve the model's performance without largely increasing the model parameter size. We may explore other larger backbones in the future. Thanks!