llama.cpp

cb1c0727 - HellaSwag: split token evaluation into batches if needed (#2681)