llama.cpp

model: M3: Move MSA into a new memory implementation

#26338

Merged