Fastest way to install TF Serving in RHEL 8

[shantanu-deepquanty]

I'd like to know the fastest way to install Tensorflow Serving in an RHEL 8 system(with GPU support) which has a CUDA version 11.0

The official documentation in the tensorflow website only contains a Model Server binary for Ubuntu (using APT) but not for RHEL
I had attempted to build a container with a Dockerfile according to the steps given in https://www.tensorflow.org/tfx/serving/building_with_docker#dockerfiles
and the docker file "Dockerfile.devel-gpu" to be specific and the build was not complete and gave a 404 Error at step 37

If not this method, is there a faster method to install in RHEL 8?

[singhniraj08]

@shantanu-deepquanty,

Kindly try to install using steps given in this docker file.

You can also refer to below links for installing TF serving using GPU.

• Serving with Docker using your GPU
• Using apt and pip to install TensorFlow Serving

[shantanu-deepquanty]

@singhniraj08
Thank you very much. The dockerfile that you had linked to (Dockerfile.gpu) worked like a charm. The other dockerfile I had used was erroneous (Dockerfile.devel-gpu)

[singhniraj08]

@shantanu-deepquanty,

Since this issue is resolved, closing this issue.
Feel free to reopen and post your comments(if you still have queries on this). Thank you!