llama.cpp

[CANN] Optimize RMS_NORM using cache

#15419

Merged