leanpass 0.1.0

A lightweight, transparent alternative to PyTorch/TensorFlow built on NumPy.

leanpass

A minimal, transparent NumPy-based autograd library for small models.

Features

• Tensor with reverse-mode autodiff
• Operators: +, -, *, /, **, @
• Activations: relu, sigmoid, softmax
• Linear and MLP modules
• SGD and Adam optimizers
• Graph visualization via Tensor.visualize_dot()

Install

pip install .

Quick start

from leanpass import Tensor, nn, optim

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

Run demo

python demo.py

Publish to PyPI

1. Update version in pyproject.toml.
2. Install build tools:

python -m pip install --upgrade build twine

3. Build distribution files:

python -m build

4. Upload to PyPI:

python -m twine upload dist/*