LeanPass

Why LeanPass?

• Minimal dependency footprint: only NumPy
• Readable, transparent implementation
• Core autodiff and model-building primitives
• Great for education, prototyping, and learning

Features

• Tensor objects with reverse-mode autodiff
• Arithmetic operators, matrix multiplication, and broadcasting
• Activation functions: relu, sigmoid, softmax
• Neural network layers: Linear and MLP
• Optimizers: SGD and Adam
• Simple demo script and test coverage

Install

pip install leanpass

Quick start

from leanpass import Tensor, nn

x = Tensor([[1.0, 2.0]], requires_grad=False)
model = nn.MLP([2, 16, 3])
logits = model(x)
print(logits)

Run the demo

python demo.py

Repository layout

• leanpass/ — package source code
• tests/ — regression and gradient-check tests
• demo.py — end-to-end example
• pyproject.toml — package metadata

Contribution

See the repository files for guidelines:

• CONTRIBUTING.md
• SECURITY.md
• LICENSE