A PyTorch-based deep learning framework that automates training loops, multi-GPU scaling and engineering boilerplate while keeping full control of model code.
PyTorch Lightning is a framework for pretraining and finetuning PyTorch models. You define the model logic in a LightningModule, and the Trainer handles backpropagation, mixed precision, device placement, and distributed training across CPUs, GPUs, TPUs, and multiple nodes. The same code can run on different hardware by changing a few Trainer arguments.
The project also includes Lightning Fabric, a lower-level layer for users who want to keep their own training loop while still getting device handling and distributed strategies such as DDP, FSDP, and DeepSpeed. Built-in features include checkpointing, early stopping, logger integrations with tools like TensorBoard, Weights & Biases, and MLflow, and export to TorchScript and ONNX.
The repository provides a quick start example, a set of task-based examples covering classification, segmentation, detection, LLM finetuning, and diffusion, and documentation for both packages.
Install Lightning with pip
pip install lightningInstall torchvision for the example
pip install torchvisionSave the toy example as main.py and run it
python main.pySummary generated by AI from the project's README. GitHub stats refresh daily.
A PyTorch-based distributed training and inference system for large AI models, offering parallelism, memory optimization and heterogeneous training.
A deep learning optimization library for making distributed training and inference of large models easier and more efficient.
An optimized Mixture-of-Experts library from Microsoft for training and serving MoE models in PyTorch, with low-precision inference for models such as DeepSeek, Qwen and GLM.
A lightweight, type-checked functional deep learning library for composing models and wrapping PyTorch, TensorFlow and MXNet layers.
CLI and Python toolkit for packaging and deploying ML models to Baseten or your own infrastructure as production model servers.
Ray is a unified framework for scaling Python and AI applications from a laptop to a cluster, with a distributed runtime and AI libraries.