Can you add llava next support ?
#4
by
Yifanye
- opened
It seems that a new llama 8B multimodal model based on "Llava Next" exists on the HuggingFace platform. At the same time, it seems that the principle has changed very little, but has effectively increased the resolution of the inputs that can be recognized. Is it possible to create mmproj for these models? Thank you very much if you can make them and share them!
Also, for a visual layer, Q4 quantization might be too harsh, fp16 might be better, since it is rather small.