PyTorch

Founded by Soumith Chintala in 2016

Build and train deep learning models using flexible Python code

Pricing model

Free

Rating

★ People love it

Time to value

Quick Setup (< 1 hour)

You can use PyTorch to build, train, and deploy deep learning and neural network models. It lets you write Python code to define architectures and experiment with eager execution, making debugging simple. You can transition models into production using TorchScript and serve them using TorchServe. PyTorch supports scalable distributed training across clusters and hardware accelerators like NVIDIA and AMD GPUs. It also offers specialized domain libraries for computer vision, natural language processing, and graph neural networks.

What PyTorch does

Train deep learning models on GPU hardwareExport models to TorchScript for production inferenceDistribute model training across multi-node clustersCompute automatic gradients using autogradProcess audio datasets with torchaudioPreprocess image collections with torchvisionDeploy inference services using TorchServeDynamic computational graph support for eager executionTorchScript for converting research models to production environmentsDistributed training across multi-node and multi-GPU configurations with torch.distributedTorchServe for model serving and production inferenceHardware acceleration with CUDA, ROCm, and specialized chipsExtensible domain libraries like torchvision and torchaudioModel interpretability tools through Captum