llama.cpp

603300b0 - context : fix off-by-one comparisons to n_gpu_layers (#24208)