leanpass 0.1.4

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

LeanPass

LeanPass is a lightweight NumPy-based autodiff library for building small neural network models and understanding automatic differentiation in a simple, readable way.

Public API

Tensor

The core data structure is Tensor, which wraps a NumPy array and supports automatic differentiation.

from leanpass import Tensor

x = Tensor([[1.0, 2.0]], requires_grad=False)
y = Tensor([[3.0, 4.0]])
z = x + y

Supported operations:

• Tensor + Tensor
• Tensor - Tensor
• Tensor * Tensor
• Tensor / Tensor
• Tensor ** Tensor
• Tensor @ Tensor
• Tensor.sum()
• Tensor.mean()
• Tensor.relu()
• Tensor.sigmoid()
• Tensor.tanh()
• Tensor.leaky_relu()
• Tensor.gelu()
• Tensor.softmax()
• Tensor.backward()

Neural network layers

from leanpass import nn

layer = nn.Linear(4, 8)
mlp = nn.MLP([4, 16, 8])

Available components:

• nn.Linear(in_features, out_features) creates a linear layer with weights and bias.
• nn.MLP(layer_sizes) creates a multilayer perceptron with ReLU activations between layers.
• nn.mse_loss(predictions, targets) computes mean squared error.
• nn.cross_entropy_loss(predictions, targets) computes categorical cross-entropy for multi-class targets.
• nn.binary_cross_entropy_loss(predictions, targets) computes binary cross-entropy for binary classification.

Optimizers

from leanpass import optim

optimizer = optim.SGD(model.parameters(), lr=0.01)
# or
optimizer = optim.Adam(model.parameters(), lr=0.001)

Available optimizers:

• optim.SGD(parameters, lr=...) performs simple gradient descent.
• optim.Adam(parameters, lr=...) performs Adam optimization with bias correction.

Example

from leanpass import Tensor, nn, optim

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

Notes

This package is intended for clarity and educational use, with a compact implementation that makes the autodiff process easier to inspect.