llama.cpp

Adding --direct-io flag for model loading

#18166

Merged