llama.cpp

7a6e91ad - CUDA: replace GGML_CUDA_F16 with CUDA arch checks (#15433)