llama.cpp

dflash: pass missing NVFP4 scales to attention operations

#28000

Merged