Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Invalid argument: unsupported datatype 'TYPE_BYTES' on 20.08

See original GitHub issue

Description Using nvcr.io/nvidia/tritonserver:20.08-py3 to load a Tensorflow 1.15 savedmodel, and get the error log as follows

triton_1  | E0904 09:38:33.860808 1 model_repository_manager.cc:928] failed to load 'test_model' version 1: Invalid argument: unsupported datatype 'TYPE_BYTES' for tensor 'feature1' for model 'test_model'

but the same model can be successful loaded with nvcr.io/nvidia/tritonserver:20.07-py3

I also try set --backend-config=tensorflow,version=1 to the new version, but still shows that error.

In addition, I can make sure this tensor in my model is DT_STRING type, not the BYTES type.

It seems this is a compatible issue between tf 1.x and 2.x

Triton Information Release 2.2.0 corresponding to NGC container 20.08

Are you using the Triton container or did you build it yourself? use official image nvcr.io/nvidia/tritonserver:20.08-py3

To Reproduce maybe load a tf 1.x model with DT_STRING tensor will reproduce this error

Expected behavior tf 1.15 savedmoel can be loaded the same as old version.