ParentFull threadDkuku·In llama.cpp You can offload some of the layers to gpu with -ngl X. Where x is the number of layersView on HN