llama.cpp

2dd5d2c9 - convert-llama-h5-to-gguf.py : add 70b gqa support