llama.cpp

bug fix: handle saving/loading null layers in recurrent memory

#14675

Merged