[Relax][TensorRT] Fix YOLO BYOC offload and partitioning gaps#19998
[Relax][TensorRT] Fix YOLO BYOC offload and partitioning gaps#19998tlopex wants to merge 4 commits into
Conversation
There was a problem hiding this comment.
Code Review
This pull request introduces support for silu and image.resize2d operators in the TensorRT BYOC backend, including their respective pattern registrations, runtime converters, and unit tests. It also fixes a dependency tracking issue in MergeCompositeFunctions by explicitly handling TupleNode and TupleGetItemNode to prevent cyclic merges, and updates the arange operator to properly handle raw data types. Additionally, error reporting for unsupported TensorRT operators has been improved. No review comments were provided, so there is no feedback to address.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
Fixes #19887.
This PR fixes several gaps in the Relax TensorRT BYOC pipeline exposed by YOLO-style models.
The reported failures occurred at three different stages:
relax.permute_dimsandrelax.image.resize2dwere not offloaded consistently because TensorRT composite naming and runtime converter coverage were incomplete.MergeCompositeFunctionscould incorrectly merge branches involvingsplitandTupleGetItem, resulting in a cyclic group dependency.Rangeconversion could pass aPrimTypedirectly torelax.arange, while its FFI expects aDataType.