llama.cpp

fit : account for skipped NextN layers when setting n_gpu_layers

#31569

Merged