Scale your models, not your boilerplate with PyTorch Lightning! PyTorch Lightning is the ultimate PyTorch research framework that allows you to focus on the research while it takes care of everything else. It's designed to decouple the science from the engineering in your PyTorch code, simplifying complex network coding and giving you maximum flexibility. PyTorch Lightning can be used for just about any type of research, and was built for the fast inference needed in AI research and production. When you need to scale up things like BERT and self-supervised learning, Lightning responds accordingly by automatically exporting to ONNX or TorchScript.
PyTorch Lightning can easily be applied for any use case. With just a quick refactor you can run your code on any hardware, run distributed training, perform logging, metrics, visualization and so much more!
Features
- Scale models to run on any hardware (CPU, GPUs, TPUs) without changing your model
- Produce more readable code by decoupling the research code from the engineering
- Easier reproduction
- Automates most of the training loop and tricky engineering
- Maintains flexibility while removing plenty of boilerplate
- Out-of-the-box integration with numerous logging/visualizing frameworks
- Tested rigorously with every new PR
- Minimal running speed overhead
Project Samples
