llama.cpp

0c898640 - retrieval : use at most n_seq_max chunks (#18400)