llama.cpp

model : load hparams.n_layer_nextn before n_layer() calls

#28159

Merged