llama.cpp

f28493c7 - models : appropriately flag noscan ssm_a tensors (#28121)