Purpose

Some ggml playground of mine. Never touched c++ or anything related to it in my life, might do dumb things here :)

Ggml library lets you run deep learning models efficiently and on a huge variety of devices, including phones and microcontrollers.

I want to try to build some models and get them running to see how well and fast it works. Learning experience FTW.

Setting up

Clone the repo and get the submodules

git clone https://github.com/jezzarax/xornet.git
git submodule update --init --recursive

Python part

no venv or dependencies definition yet, might come soon. The usual deep learning stack is expected, pytorch, numpy, huggingface
Also python -m pip install ggml is needed for model conversion

C++ part

cmake and make are needed to build the runner

running

get train_xornet.py and conver_xornet.py to run to the end
build ggml model runner

mkdir build && cd build
cmake ..
make
./xornet ../models/xornet.gguf

Models

XorNet

Implementation of the classic XOR problem. The model is a simple 2 layer neural network with 2 input neurons, 2 hidden neurons and 1 output neuron. The model is trained using pytorch and then converted to ggml format.

jezzarax / xornet