llama.cpp

563dec81 - llama : allocate indexer cache only in "full" indexer layers (#26474)