Are you tired of navigating through messy scripts and forgetting crucial parameters in your experiments? We certainly are! Lighter is here to organize your research code and experiments so you can focus on the science.
What is Lighter?
Lighter is a “configuration” first framework designed to organize, standardize, and make your deep learning experiments reproducible. This tool is not just a piece of software; it’s a convention that helps you take a step towards greater reproducibility and transparency in machine learning research. Once you start using configurations, argparse will seem like a stone age tool.
Here are some key features
Structured Workflow: Lighter brings much-needed structure to your experiments as we provide a “standard” yet flexible way to define your experiments in YAML files. This ensures they are organized and reproducible. No more lost or forgotten parameters!
Boilerplate-Free Design: Say goodbye to copying (and hopefully not messing up) over training logic. Lighter covers everything from classification to self-supervised learning, allowing you to focus on what’s important.
Readability: Ever wondered about the learning rate or architecture used in an old experiment? With Lighter, all you need to do is glance at the config.
Modularity: Overriding parameters and adding custom modules is a breeze. Leave behind the days of argparse and hardcoding. When you write custom logic, it is always clear and easy to share.
Get onboard Lighter with us and leave behind your worries, making you hopefully lighter.