llama.cpp
CUDA: fuse FFN gate/up matmuls and GLU in MMQ with interleaved gate/up rows
#29948
Merged

Loading