Comment by summarity

llama.cpp is just a frontend for the GGML tensor library, which required models be converted (optionally quantized) to GGML format.

Of course people are working on that for Dolly as well: https://huggingface.co/snphs/dolly-v2-12b-q4