[Feature Request]: Support ONNX runtime in RunInference API

[hurutoriya]

What would you like to happen?

ONNX Runtime is one of the famous frame work for ML inference in the real world.
https://onnxruntime.ai/
If RunInference API support the ONNX runtime which cover the various ML framework actually.
e.g. not only PyTorch , scikit-learn and TensorFlow but also Transformers(Hugging face) , XGBoost and so on.
https://onnx.ai/supported-tools.html

Issue Priority

Priority: 2

Issue Component

Component: sdk-py-core

[AnandInguva]

cc: @damccorm

[ziqi-ma]

.take-issue

[ziqi-ma]

A first-timer question: I see the integration tests for python sdk pull data from a gs location. How can I read/write to that location? Do I need special permission?

[AnandInguva]

I think so. To access apache-beam-testing, you need certain permission. But this shouldn't stop you from the API development. You can develop a local test and I can help you with the location in the later stages.

[ziqi-ma]

.Thanks - I wrote the API and local unit tests for it. Could you help me review it before I proceed with tests that require remote files? Should I make a draft pr? Thanks.

[AnandInguva]

Yes, that sounds good. Please refer this issue in the PR and tag me as a reviewer.

[ziqi-ma]

Seems like I cannot add reviewers to draft prs? Here is the PR: #24911

[tvalentyn]

I'm closing this since it seems to have been completed, however onnx functionally currently appears untested and potentially broken, let's follow up in #31254.