PyTorch Project Template

Yet another template for PyTorch projects Highly customizable, modularized code

Requirements

• Python 3
• PyTorch
• pyyaml, addict (for configuration)
• wandb (for visualizing training runs)

Usage

Run cd src && python train.py for sample training. For custom use, modify the following files:

• src/dataset.py ... for custom Dataset and collate function for DataLoader
• src/model.py ... for custom Module
• src/optimization.py ... for custom loss function and optimizer
• src/evaluator.py ... for custom evaluator (used for validation)

Then, put in configurations (hyperparameters, ...) into cfg/hogehoge.yml Sample configuration file is in cfg/sample.yml

Use your own configuration file to train model

train.py --config path/to/configuration/file.yml

Resume training with configuration

train.py --config path/to/configuration/file.yml --resume