cmp-nct

/

llava-1.6-gguf

like 73

I uploaded one

Thanks so much! will this work alongside one of your mmproj models even though they have different quantizations?

It seems the 4bit models are dramatically quicker on Windows by orders of magnitude than 5 or 6 bits.

Sure, the CLIP model already is quite small I'd use the 7 bit (q6_k) variant for all purposes at the moment.
You can combine as you like

I tried that combination out, it works great! Thanks for the help!