llama.cpp

HellaSwag: split token evaluation into batches if needed

#2681

Merged