2 ms·

>This is by far the best open source vlm model LLaVA 1.5 is very good, at least at describing images. http://llava.hliu.cc/ http://llava.hliu.cc/

• axiom92 3y ago
  Right, but no separate image encoder + half the size could be very helpful for many applications.
  • GaggiX 3y ago
    The 7B LLaVa model is smaller, even considering the image encoder (CLIP-L).