neuro-scribe 0.2.0.post3

NeuroScribe - a lightweight deep learning framework.

NeuroScribe v0.2.0

Installation

> pip install neuro-scribe

[!Note]

• If you want CUDA support for GPU acceleration, please ensure that you have CUDA Toolkit version 12 or higher installed on your device.
  You can download CUDA Toolkit from the official NVIDIA website: CUDA Toolkit Downloads
• If you're installing NeuroScribe on an Apple Silicon Mac, NeuroScribe will automatically detect the architecture and install additional dependencies for Metal Performance Shaders (MPS) support.

Quick Start Example

This example demonstrates how to use Neuroscribe in a way that closely resembles PyTorch syntax and conventions. By following PyTorch's familiar patterns, users can easily integrate Neuroscribe into their existing deep learning workflows without significant adjustments.

import neuroscribe as ns
import neuroscribe.nn as nn


class MyModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.l1 = nn.Linear(784, 128)
        self.relu = nn.ReLU()
        self.l2 = nn.Linear(128, 10)

    def forward(self, x):
        x = self.l1(x)
        x = self.relu(x)
        x = self.l2(x)
        return x


def main():
    model = MyModel()
    model.eval()
    model.to('cuda')
    input = ns.randn(32, 784, dtype='float32', requires_grad=False, device='cuda')
    output = model(input)
    print(output)


if __name__ == '__main__':
    main()

Available Accelerators

Neuroscribe currently supports the following accelerators:

• CPU (default)
• CUDA
• MPS (Metal Performance Shaders)

We plan to add support for additional accelerators to further enhance performance capabilities.

Notes

• Functions suffixed with '_' in the Tensor class perform in-place modifications to the Tensor object and do not create a new Tensor as output.