trt-strong-typing-migration

Migrating TensorRT builds from weak to strong typing involves automating the transformation of weakly-typed Python network definitions into strongly-typed networks. This ensures compatibility with TensorRT 10.12 and later versions while validating new builds against baseline outputs.

ONNX AutoCast is a functionality that provides AutoCast support for ONNX files to enable mixed precision. It is used during the migration to strongly-typed networks to streamline TensorRT builds and ensure engines are rebuilt correctly.

TensorRT 10.12 and later versions require strongly-typed network definitions for compatibility. Migrating from weak to strong typing ensures your Python network definitions function correctly and allows for proper engine rebuilding and validation.

To validate TensorRT engines after migrating to strong typing, the migration process automatically validates the newly rebuilt engines against baseline results. This ensures the strongly-typed networks produce correct outputs before deployment.

Yes, you can convert an existing weakly-typed Python network definition to a strongly-typed network. The migration automates this conversion process, handling the network transformation and rebuilding to support the latest TensorRT versions.