Pipelight: The Easiest Pipeline for Training, Validation and Test Based on PyTorch Lightning

Pipelight is a pipeline that is driven by configuration files for model running based on PyTorch Lightning.

Key Features

• Reproducible Pipelines: Built-in experiment tracking and deterministic training
• Configuration-Driven Workflows: Define experiments through YAML configs with PyTorch Lightning integration

Installation

pip install pipelight

Quick Start

Configuration-Based Execution

Copy pipelight/configs as a template to the project folder, and enter this folder. All YAML files can be modified to meet the task requirements.

For training and validation, just run:

python -m pipelight.run --train configs/train.yaml --val configs/val.yaml -m {model config} -r configs/running/runner.yaml -n experiment_01

After that, run the following to test:

python -m pipelight.run --test configs/test.yaml -m {model config} -r configs/running/runner.yaml -n experiment_01

Make sure the name of the experiment should be kept the same.