Axon: is a lightweight Python library for creating and manipulating multi-dimensional arrays, inspired by libraries such as NumPy. It's written in python only, for now.
Axon.micro: You have seen Micrograd by Karpathy, this is the upgraded version of micrograd written in c/c++ & has more functions & operational support. A light weight scalar-level autograd engine written in c/c++ & python
- Element-wise operations (addition, multiplication, etc.)
- Matrix multiplication
- Broadcasting
- Activation functions (ReLU, tanh, sigmoid, GELU)
- Reshape, transpose, flatten
- Data type conversion
- Micrograd support(Scalar level autograd engine)
Clone the repository:
git clone https://github.com/shivendrra/axon.git
cd axon
or
Install via pip:
pip install axon-pypi
You can use this similar to micrograd to build a simple neural network or do scalar level backprop.
import axon
from axon import array
# Create two 2D arrays
a = array([[1, 2], [3, 4]], dtype=axon.int32)
b = array([[5, 6], [7, 8]], dtype=axon.int32)
# Addition
c = a + b
print("Addition:\n", c)
# Multiplication
d = a * b
print("Multiplication:\n", d)
# Matrix Multiplication
e = a @ b
print("Matrix Multiplication:\n", e)
Addition:
array([6, 8], [10, 12], dtype=int32)
Multiplication:
array([5, 12], [21, 32], dtype=int32)
Matrix Multiplication:
array([19, 22], [43, 50], dtype=int32)
anyway, prefer documentation for detailed usage guide:
- axon.md: for development purpose
- usage.md: for using it like numpy
- axon_micro.md: for axon.micro i.e. scalar autograd engine
from axon.micro import scalar
a = scalar(2)
b = scalar(3)
c = a + b
d = a * b
e = c.relu()
f = d ** 2.0
f.backward()
print(a)
print(b)
print(c)
print(d)
print(e)
print(f)
you can even checkout example neural networks to run them on your system, or build your own :-D.
If you would like to contribute to this project, you can start by forking the repository:
- Click the "Fork" button at the top right of this page.
- Clone your forked repository to your local machine:
git clone https://github.com/shivendrra/axon.git
- Create a new branch:
git checkout -b my-feature-branch
- Make your changes.
- Commit and push your changes:
git add .
git commit -m "Add my feature"
git push origin my-feature-branch
- Create a pull request on the original repository.
To run the unit tests you will have to install PyTorch & Numpy, which the tests use as a reference for verifying the correctness of the calculated gradients & calculated values. Then simply run each file according to your prefrence:
python -m tests/test_array.py # for testing the axon functions with numpy
python -m tests/test_micro.py # for testing the axon.micro functions with pytorch
We welcome contributions! Please follow these steps to contribute:
- Fork the repository.
- Create a new branch for your feature or bugfix.
- Make your changes.
- Ensure all tests pass.
- Submit a pull request with a clear description of your changes.
This project is licensed under the MIT License. See the LICENSE file for more details.