Skip to content

Latest commit

 

History

History
64 lines (57 loc) · 1.56 KB

README.md

File metadata and controls

64 lines (57 loc) · 1.56 KB
Raw

Table of contents

Title Desc
Version Current version and features
Purpose What and why of this repo
Structure File structure of various modules
Upcoming features Features that im working on
Setup How to setup locally

Version info

Current version - 0.1.2
New features - new Flatten layer added inside PureTorch.layers

Runs on CPU only. Might add GPU support later.

Purpose

Raw implementation of PyTorch using NumPy. The structure and essence of torch remains the same, but its fully implemented using NumPy.

Structure

Current file structure:

  • PureTorch
    • activations
      • ReLU
      • Softmax
      • Tanh
    • layers
      • Sequential
      • Linear
      • Flatten
    • loss
      • CrossEntropyLoss
      • SparseCrossEntropyLoss

Will be adding other layers, activations, losses, optimizers.
As of now, the default optimizer is SGD.
To check the "development code" check the "dev" branch

Upcoming features

These are the features that im working on, and will soon be a part of PureTorch.

  • Convolutional layer(s)
    • 2D first, then 1D and multi-dim

Setup

Install git before running this command in your env, then run:

pip install "git+https://github.com/Dristro/PureTorch"

Run: (to verify the installation in python)

import PureTorch
print(PureTorch.__version__)

or: (to verify the installation on the terminal)

python3 -c "import PureTorch; print(PureTorch.__version__)"