/moe/

i just build llama cpp with cmake myself so i can get the latest model updates but i guess the AUR package will do that for you too

bleh this model takes too much RAM

if i can't make it fix i'll go down to Q4

ok 8-bit quantized context makes it barely fit

>>>/watch?v=kMGbGGllQoE

sis*