llama.cpp

9d817213 - model : load hparams.n_layer_nextn before n_layer() calls (#28159)