llama.cpp

llama : use n_swa + n_ubatch cells for SWA cache

#13833

Merged