cyllama - a cython wrapper of llama.cpp

This project provides a cython wrapper for @ggerganov's llama.cpp which is likely the most active open-source compiled LLM inference engine. It was spun-off from my earlier, now frozen, llama.cpp wrapper project, llamalib which provided early stage, but functional, wrappers using cython, pybind11, and nanobind. Further development of cyllama, the cython wrapper from llamalib, will continue in this project.

Development goals are to:

• Stay up-to-date with bleeding-edge llama.cpp (last stable build with llama.cpp b4381)

• Produce a minimal, performant, compiled, thin python wrapper around the core llama-cli feature-set of llama.cpp.

• Integrate and wrap llava-cli features.

• Integrate and wrap features from related projects such as whisper.cpp and stable-diffusion.cpp

• Learn about the internals of this popular C++/C LLM inference engine along the way. For me at least, this is definitely the most efficient way to learn about the underlying technologies.

Given that there is a fairly mature, well-maintained and performant ctypes-based wrapper provided by @abetlen's llama-cpp-python project and that LLM inference is gpu-driven rather than cpu-driven, this all may see quite redundant. Nonetheless, we anticipate some benefits to using a compiled cython-based wrapper instead of ctypes:

• Cython functions and extension classes can enforce strong type checking.

• Packaging benefits with respect to self-contained statically compiled extension modules, which include simpler compilation and reduced package size.

• There may be some performance improvements in the use of compiled wrappers over the use of ctypes.

• It may be possible to incorporate external optimizations more readily into compiled wrappers.

• It may be useful in case one wants to de-couple the python frontend and wrapper backends to existing frameworks: for example, to just replace the ctypes wrapper part in llama-cpp-python with compiled cython wrappers and contribute it back as a PR.

Status

Development is done only on macOS to keep things simple, with intermittent testing to ensure it works on Linux.

The following table provide an overview of the current wrapping/dev status:

status	cyllama
wrapper-type	cython
wrap llama.h + other headers	yes
wrap high-level simple-cli	yes
wrap low-level simple-cli	yes
wrap low-level llama-cli	WIP

The initial milestone entailed creating a high-level wrapper of the simple.cpp llama.cpp example, followed by a low-level one. The next objective is to fully wrap the functionality of llama-cli which is ongoing (see: cyllama.__init__.py).

It goes without saying that any help / collaboration / contributions to accelerate the above would be welcome!

Wrapping Guidelines

As the intent is to provide a very thin wrapping layer and play to the strengths of the original c++ library as well as python, the approach to wrapping intentionally adopts the following guidelines:

• In general, key structs are implemented as cython extension classses with related functions implemented as methods of said classes.

• Be as consistent as possible with llama.cpp's naming of its api elements, except when it makes sense to shorten functions names which are used as methods.

• Minimize non-wrapper python code.

Setup

To build cyllama:

1. A recent version of python3 (testing on python 3.12)

2. Git clone the latest version of cyllama:

git clone https://github.com/shakfu/cyllama.git
cd cyllama

3. Install dependencies of cython, setuptools, and pytest for testing:

pip install -r requirements.txt

4. Type make in the terminal.

This will:

1. Download and build llama.cpp
2. Install it into bin, include, and lib in the cloned cyllama folder
3. Build cyllama

Testing

The tests directory in this repo provides extensive examples of using cyllama.

However, as a first step, you should download a smallish llm in the .gguf model from huggingface. A good model to start and which is assumed by tests is Llama-3.2-1B-Instruct-Q8_0.gguf. cyllama expects models to be stored in a models folder in the cloned cyllama directory. So to create the models directory if doesn't exist and download this model, you can just type:

make download

This basically just does:

cd cyllama
mkdir models && cd models
wget https://huggingface.co/bartowski/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q8_0.gguf 

Now you can test it using llama-cli or llama-simple:

bin/llama-cli -c 512 -n 32 -m models/Llama-3.2-1B-Instruct-Q8_0.gguf \
 -p "Is mathematics discovered or invented?"

You can also run the test suite with pytest by typing pytest or:

make test

If all tests pass, you can type python3 -i scripts/start.py or ipython -i scripts/start.py and explore the cyllama library with a pre-configured repl:

>>> from cyllama import Llama
>>> llm = Llama(model_path='models/Llama-3.2-1B-Instruct-Q8_0.gguf')
>>> llm.ask("what is the age of the universe?")
'estimated age of the universe\nThe estimated age of the universe is around 13.8 billion years'

TODO

• wrap llama-simple

• wrap llama-cli (WIP: see: cyllama.__init__)

• wrap llama-llava-cli

• wrap whisper.cpp

• wrap stable-diffusion.cpp