EasyTorch is an open source neural network framework based on PyTorch, which encapsulates common functions in PyTorch projects to help users quickly build deep learning projects.
- 💻 Minimum Code. EasyTorch encapsulates the general neural network training pipeline. Users only need to implement key codes such as
Dataset,Model, and training/inference to build deep learning projects. - 🔧 Everything Based on Config. Users control the training mode and hyperparameters through the config file. EasyTorch automatically generates a unique result storage directory according to the MD5 of the config file content, which help users to adjust hyperparameters more conveniently.
- 🔦 Support All Devices. EasyTorch supports CPU, GPU and GPU distributed training (single node multiple GPUs and multiple nodes). Users can use it by setting parameters without modifying any code.
- 📃 Save Training Log. Support
logginglog system andTensorboard, and encapsulate it as a unified interface, users can save customized training logs by calling simple interfaces.
Ubuntu 16.04 and later systems are recommended.
python >= 3.6 (recommended >= 3.9)
Miniconda or Anaconda are recommended.
pytorch >= 1.4 (recommended >= 1.9). To use CUDA, please install the PyTorch package compiled with the corresponding CUDA version.
Note: To use Ampere GPU, PyTorch version >= 1.7 and CUDA version >= 11.0.
pip install easy-torchTODO
More examples are on the way
It is recommended to refer to the excellent open source project BasicTS.
If EasyTorch helps your research or work, please consider citing EasyTorch.
The BibTex reference item is as follows(requires the url LaTeX package).
@misc{wang2020easytorch,
author = {Yuhao Wang},
title = {{EasyTorch}: Simple and powerful pytorch framework.},
howpublished = {\url{https://github.com/cnstark/easytorch}},
year = {2020}
}If your project is using EasyTorch, please consider put the EasyTorch badge 
[](https://github.com/cnstark/easytorch)
(Full documentation is coming soon)