This repository contains the BF16 and Q8_0 GGUF files for use in stock llama.cpp converted via the default convert_hf_to_gguf.py script from llama.cpp.
Usage:
./llama-server
# ... all your other command-line arguments
--spec-type draft-mtp
--spec-draft-model ./PATH_TO_ASSISTANT_GGUF
--spec-draft-n-max 2
This is an unofficial repository not endorsed by or connected to Google LLC.
