TensorRT Support for Multiple Profiles

[adjhawar]

I was going through the documentation to use TensorRT EP with onnxruntime but did not find any mention of TensorRT profiles there. In the header files however, there were APIs that hint that multiple profiles may be supported. Multiple profiles can be useful when the input shapes are dynamic like in most language modelling usecases.

So just wanted to know if there a plan to support multiple TensorRT profiles in the same inference session? In case, the feature is already supported, could you let me know on how to use it?