Vision support in kimi k2.5-gguf

#7
by Johnasson - opened

So, a big beautiful person just merged support for vision in llama.cpp

https://github.com/ggml-org/llama.cpp/pull/19170

do current models in repo need to be re-quanted to be able to support vision?

If you want to use the Vision feature, might be better to use the author's repo. https://huggingface.co/AesSedai/Kimi-K2.5-GGUF

The PR was merged a couple of days ago. I tried just using the mmproj from AesSedai's repo with the unsloth quant but no joy. Will the unsloth ggufs support vision at some point?

Sign up or log in to comment