llama.cpp

ggml-cuda: add stricter checking for fusion

#17568

Merged