
pranftw/neograd
🏗️ Frameworkpranftw
A lightweight, from-scratch deep learning framework built using Python and NumPy.
Neograd serves as a minimalist deep learning library that mirrors the functionality of larger frameworks like PyTorch, but is implemented entirely from scratch. At its core, it features a robust autograd engine that tracks operations on tensors to compute gradients automatically. The framework includes essential building blocks for deep learning, such as linear layers, activation functions, and loss functions, all constructed using NumPy arrays. By avoiding heavy dependencies, Neograd provides a transparent environment for developers to experiment with gradient-based optimization, computational graph construction, and the mathematical implementation of neural network layers. It is an excellent resource for those looking to bridge the gap between high-level API usage and the underlying calculus of machine learning.
💡Highlights
- ├─Built from scratch with NumPy
- ├─Custom autograd engine
- └─PyTorch-like API structure
🎯For
- ├─AI Researchers
- ├─Computer Science Students
- └─Machine Learning Engineers