llama.cpp

f8249026 - YaRN : correction to GPT-NeoX implementation