Comment on Understanding, using, and finetuning GemmaparentComments−d-z-m2y32 isn't quite enough to run a decent quant of mixtral(on a Macbook). You could try a Q3_K_M, but not sure how lobotomized it would be.−CuriouslyC2yThat's not true, the GGUF quants aren't great but there are exl2 4bit quantizations floating around that are pretty sweet.−brucethemoose22yexl2 is Nvidia/AMD only.But GGUF Mixtral should fit in 32GB... just not with the full 32K context. Long context is very memory intense in llama.cpp, at least until they fully implement flash attention and a quantized cache.−d-z-m2yfair enough, yeah I'm talking about GGUF quants only.
Comments
32 isn't quite enough to run a decent quant of mixtral(on a Macbook). You could try a Q3_K_M, but not sure how lobotomized it would be.
That's not true, the GGUF quants aren't great but there are exl2 4bit quantizations floating around that are pretty sweet.
exl2 is Nvidia/AMD only.
But GGUF Mixtral should fit in 32GB... just not with the full 32K context. Long context is very memory intense in llama.cpp, at least until they fully implement flash attention and a quantized cache.
fair enough, yeah I'm talking about GGUF quants only.