TensorRT versioning question between the SDKs and external SDKs


The version 0.8.1 of NVAR ships with nvinfer which exposes a tensorrt_version= This version isn’t found on the release pages for TensorRT. We depend on TensorRT with a lot of other tools (Kinect Azure, Stereolabs ZED), so we want to try to manage the TensorRT version between all of these packages sanely. We would prefer to use the publicly released This allows us to have a full set of headers/lib as part of our build system, and presumably allows other tools to work well if they are shipped against
However, if we try to use with NVAR 0.8.1, it fails. Is this expected? Is there potentially cached data that was generated using (since we first tried with that), that we need to clear out?
We should be able to upgrade the TensorRT version without issue, no?