Zero Zeng

I guess the reason is related to some cuda kernel under cuda 11.6. will investigate further

This issue is fixed in 8.5.1 with FASTER_DYNAMIC_SHAPES_0805 preview feature. it should come with our TRT 8.5 release, ``` [I] PASSED | Output Validation [V] Loaded Module: sys [I] PASSED...

@azhurkevich ^ ^

Can you elaborate on your question?

It should be supported, but I'm not an expert on python API, @pranavm-nvidia might know more about it.

something I know that might be helpful to you: https://docs.nvidia.com/deeplearning/tensorrt/api/python_api/infer/Core/EngineInspector.html https://docs.nvidia.com/deeplearning/tensorrt/api/python_api/infer/Core/Engine.html#tensorrt.ICudaEngine.create_engine_inspector https://docs.nvidia.com/deeplearning/tensorrt/api/python_api/infer/Core/BuilderConfig.html#tensorrt.ProfilingVerbosity https://docs.nvidia.com/deeplearning/tensorrt/api/python_api/infer/Core/ExecutionContext.html#tensorrt.IExecutionContext.report_to_profiler

Can you provide a reproduce? btw can you try the latest TRT 8.4?

@kevinch-nv ^ ^

Maybe a bug, what does your input dimension looks like? 

error node: