/DeepMuon

DeepMuon is initially built for PandaX-4T III and TRIDENT-Hailing Plan. Up tp now it provides a easy-using platform for interdisciplinary deep-learning research.

Primary LanguagePythonOtherNOASSERTION

Introduction

DeepMuon is a easy-using deep learning platform initially built for dark matter searching experiments. Up to now it has been a interdisciplinary deep learning platform. We are eager to provide advanced model training framework and excellent project management assistance.

Here we list out some available features of DeepMuon:

  • Single GPU training, Distributed Data Parallel training and Fully Sharded Distributed Parallel training.
  • Neural Network Hyperparameter Searching (NNHS)
  • Gradient accumulation
  • Gradient clipping
  • Mixed precision training
  • Double precision training
  • Customize models
  • Customize datasets
  • Customize loss functions
  • Tidy logging system
  • Model interpretation
  • Simple and direct tutorials

More details please refer to the home page of DeepMuon.

Installation (From source recommended)

git clone https://github.com/Airscker/DeepMuon.git
cd DeepMuon
pip install -v -e ./ --user

CopyRight

GNU GENERAL PUBLIC LICENSE Version 3

Project: DeepMuon Description: Interdisciplinary Deep Learning Platform Author: Airscker(Yufeng Wang) Contributors: Yufeng Wang, Yu Zhang, Shendong Su Institutions:

University of Science of Technology of China TsingHua University Stony Brook University University of Florida

If you want to publish thesis using DeepMuon, please add bibliography:

@misc{deepmuon,
  author       = {Yufeng Wang, Yu Zhang, Shendong Su},
  title        = {DeepMuon: Interdisciplinary deep-learning platform},
  year         = {2022},
  publisher    = {GitHub},
  journal      = {GitHub repository},
  howpublished = {\url{https://airscker.github.io/DeepMuon}},
}

Copyright (C) 2024 by Airscker(Yufeng), All Rights Reserved.