TorchSharp - A Sharp Knife for PyTorch

Recently I code several projects with PyTorch, and find it really a light-weight and easy-to-use deep learning framework. After coding thousands of lines, a thought emerges in my mind about what we can do for improving code re-use and accelerating programming. That’s why I try to introduce torchsharp, a sharp knife for PyTorch.

Note that this package is still under early development, and I’ll add features continously in future. Issues ans Pull Requests are extremely welcomed.


torchsharp is a framework for PyTorch which provides a set of sharp utilities aiming at speeding up programming and encouraging code re-use. The repository consists of:

  • : Useful stuff about data operation such as dummy datasets and image tranforms for data argumentation, etc.
  • torchsharp.model : Helpful fucntions about model training process such as initializer and metrics, etc.
  • torchsharp.utils : Other tools like logger and timer, etc.

Apart from torchsharp, there’re also two auxiliary libraries for PyTorch - torchzoo and pytorch-starter-kit.


torchzoo is a zoo of models and datasets for PyTorch. Most of them are used frequently in my research life but not provided by torchvision.

This repository consists of:

PyTorch Starter Kit

pytorch-starter-kit is a demo project and quick starter kit for PyTorch.