A small tool that downloads models from the Huggingface Hub and converts them into GGML for use with llama.cpp.
- Download and compile
llama.cpp
. - Set up a virtualenv using the requirements from
llama.cpp
. - Install the requirements from this repo in that virtualenv.
- Run e.g.
python ggify.py databricks/dolly-v2-12b
(nb.: I haven't tried with that particular repo) - You'll end up with GGML models under
models/...
.
You can set --llama-cpp-dir
(or the LLAMA_CPP_DIR
environment variable) to point to the directory where you've compiled llama.cpp
.