llama.cpp

39d8bc71 - CUDA: fixed tensor cores not being used on RDNA3 (#4697)