Torch-TensorRT is an integration of the PyTorch deep learning framework and the TensorRT inference acceleration framework. With this toolkit, users can generate an optimized TensorRT engine from a trained PyTorch model with a single line of code. The tutorial notebooks included here illustrate the inference optimization procedure (and benchmark the results) for the following networks and applications:

• Hugging Face BERT: Masked Language Modeling
• EfficientNet: Image Classification
• Resnet50: Image Classification
• SSD: Object Detection