ONNX Runtime natively supports models exported from PyTorch via the ONNX format. PyTorch models can be converted to ONNX using torch.onnx.export() and then optimized and deployed through ONNX Runtime for production inference across diverse hardware targets.