The research-first deep learning framework — tensors, models, and training
PyTorch tensors, operations, GPU computing, and the autograd engine
Create neural networks by subclassing nn.Module — layers, parameters, and model composition
Datasets, DataLoaders, the training loop pattern, schedulers, and mixed precision
Image classification, transfer learning, audio processing, and data augmentation pipelines
Reduce boilerplate with Lightning, scale to multiple GPUs, and export models with ONNX