whisper.cpp

75dc800d - talk-llama : fix n_gpu_layers usage again (#1442)