Model Serving - AutoGluonOnnx

For tabular model, or any model class, is there any easy way or plan in the coming release to integrate with ONNX for downstream serving ?

A side question is for any individual trained model, e.g.`model = predictor._trainer.load_model('xgboost')`, is this directly compatible to things like mlflow.xgboost or bentoml.frameworks.xgboost.XgboostModelArtifact? It is not hard to write a customised wrapper, however would like to hear from your plan. I think it would be super good to consider the overall ecosystem when we have a such user friendly autogluon already.

Start by reviewing the model-serving path around predictor._trainer.load_model('xgboost') and the requested ONNX, MLflow, and BentoML integrations. The issue names no files or tests; done would require a decided integration scope and a documented, tested serving path for the supported model classes.

Written by the indexing model from the issue text.