Introduction

Pytorch Boiler is a minimalistic boiler plate code for training pytorch models.

Quick Start

• Clone this repository

      git clone https://github.com/nmakes/pytorch_boiler
      cd pytorch_boiler

• Run sample experiments

  40-line MNIST/CIFAR classification:

      PYTHONPATH=$PYTHONPATH:./ python3 -m example_projects.image_classifier.train

  50-line MNIST/CIFAR autoencoder:

      PYTHONPATH=$PYTHONPATH:./ python3 -m example_projects.image_autoencoder.train

Installation

Basic Requirements:

• numpy
• pytorch
• torchvision

Other Requirements:

• nvidia-apex [install] (for mixed-precision training)

Supported Functionalities

• Customizable Train / Inference engine with forward and infer modes
• Tracking multiple training / validation losses and metrics
• Loading / Saving model, optimizer and trackers based on validation loss
• Training MNIST / CIFAR in 40-lines (see train.py)
• Supports Apex Amp for mixed precision training

TODO

• Support for multiple loss optimization using multiple optimizers
• Support for tensorboard plots
• Support for torchscript
• Documentation: using metrics and losses beyond the example script

Cite

If you found Pytorch Boiler useful in your project, please cite the following:

@software{Venkat_Pytorch_Boiler,
    author = {Venkat, Naveen},
    title = {{Pytorch Boiler}},
    url = {https://github.com/nmakes/pytorch_boiler},
    year = {2022}
}

Thanks!