urela / orolo Goto Github PK

This a machine learning library that achieves Model accuracy of about 95% on the MNIST dataset, its built on top of numpy

Heavily inspired by https://github.com/karpathy/micrograd and https://github.com/geohot/tinygrad

This is an educational project. I wish to learn how machine learning libraries work under the hood

You can train the model by executing the following command:

python mnist.py

• Add support for GPUs (Triton??)
• Build some model on top of this
  • [.] DQN on cartpole
  • VAE on on MNIST
• Move away from numpy
• Add tests