llama.cpp

230ea9d2 - llama-batch: add n_keep_tail in split_equal for recurrent models (#25278)