Thats barely enough for LLaMA 33b. Q3_K_S 33b fits on my 16GB laptop + 6GB GPU (so 22GB minus some overhead from the split + linux), but its extremely tight an

Thats barely enough for LLaMA 33b. Q3_K_S 33b fits on my 16GB laptop + 6GB GPU (so 22GB minus some overhead from the split + linux), but its extremely tight and only works up to ~800 context.