PyTorch Lightning is an open-source deep-learning framework that organizes PyTorch training code and automates repetitive engineering work. Developers keep control over model architecture and research logic while Lightning handles common infrastructure such as training loops, backpropagation, mixed precision, logging, checkpointing, GPUs, and distributed execution.

The central abstraction, LightningModule, groups a model with its training, validation, testing, and optimizer logic in a structured form. Training code can then scale from a local CPU or single GPU to multi-GPU and multi-node environments with relatively small configuration changes. For users who want lower-level control, Lightning Fabric provides infrastructure primitives without imposing the full Lightning training structure.

PyTorch Lightning is aimed at researchers and ML engineers who want to reduce boilerplate while keeping the flexibility of native PyTorch. It is useful for projects ranging from small experiments to large model pretraining and fine-tuning, and the broader Lightning ecosystem also includes tooling for cloud training and model serving.