llama.cpp

4d719a6d - cuda : check if this fixes Pascal card regression (#3882)