llama.cpp

b284197d - CUDA: fix compilation with GGML_CUDA_F16 (#14837)