Yep, was just reading up on it. Wow, what a cryptic language that is! I wonder if it was used to build a screen recognition model in iOS? If it was, this LLM would be a very abstract version. I may be totally misreading / misunderstanding the paper though. I wish I had enough time to train the model and see what sort of input / output you’d get out of it.

@vick21 I don't think it was used for iOS screen recognition because IOS screen recognition was out much earlier than this whole LLM trend took off. Also Llava is based on Vicuna which is based on Llama.

I was sort of alluding to this in that farret is some sort of abstraction or perhaps has been inspired by some of the earlier work. But yeah, I am totally speculating, and perhaps even hallucinating, here! :)