llama.cpp

llama : respect no_alloc in all memory caches

#31087

Merged