ZchiPitt commented 1 year ago

Summary of Changes

Add tensor name to the model metadata such that when creating a request, we can ensure the input tensors have the same name as of the model tensors.

This is to fix the problem that test on vck5000 can not be run due to tensor name mismatches. Now we can query the model metadata and give the input tensors correct and expected names.

Closes https://github.com/Xilinx/inference-server/issues/204

Motivation

We need add populate the model tensor name into the metadata such that users can query for that.

Implementation

Modified xmodel backend source code to include this metadata.

Notes

All xmodel tests in the future can now query the modelmetadata to get the tensor names. The metadata in amdinfer.ImageInferenceRequest() will be required in the future.