Additional Resources

Table 23 Additional Resources

Resource	Description
Layer builder API documentation for manual TensorRT engine construction: Creating a Network Definition in C++ and Creating a Network Definition in Python	The manual layer builder API is useful when you need the maximum flexibility possible in building a TensorRT engine. The Layer Builder API lets you construct a network by hand in TensorRT and gives you tools to load in weights from your model. When using the layer builder API, your goal is to build an identical network to your training model using TensorRT layer by layer and then load in the weights from your model.
ONNX-TensorRT GitHub	The ONNX-TensorRT integration is a simple high-level interface for ONNX conversion with a Python runtime. It is useful for early prototyping of TensorRT workflows using the ONNX path.
Torch-TensorRT GitHub	Torch-TensorRT brings the power of TensorRT to PyTorch. In just one line of code, you can accelerate inference latency by up to 5x compared to eager execution.
TensorRT is integrated with NVIDIA’s profiling tool, NVIDIA Nsight Systems.	This is a great next step for further optimizing and debugging models you are working on productionizing.
TensorRT product documentation	Product documentation page for the ONNX, layer builder, C++, and legacy APIs.
TensorRT OSS GitHub	Contains OSS TensorRT components, sample applications, and plugin examples.
TensorRT developer page	Contains downloads, posts, and quick reference code samples.
Polygraphy API	Toolkit for running and debugging deep learning models across frameworks. Useful for comparing TensorRT outputs against ONNX Runtime and isolating accuracy issues.
Migrating from TensorRT 8.x to 10.x	API changes and code migration examples for upgrading from TensorRT 8.x to 10.x.
Troubleshooting	FAQs, common error messages, log interpretation, and guidance for reporting issues.
Optimizing Performance	Best practices for benchmarking, measurement methodology, and inference optimization.