ONNX Tracking Issue

[luisquintanilla]

This issue tracks work planned for ONNX.

Work Items

• Increase export coverage #6504
• Support Execution Providers #6505
• Provide APIs to inspect ONNX input and output schemas. #6506
• Increase coverage of ONNX model repo models in ORT extensions. #6507
• Investigate result discrepancies #6508

Related issues

Increased export coverage

• TextFeaturizingEstimator.Options KeepNumbers and KeepPunctuations not exportable to ONNX? #6317
• Support saving as ONNX for ExpressionTransformer #4615
• Custom ONNX Export Code #6369

Performance improvements

• Does the OpenVINO execution provider for ONNX Runtime work in ML.NET? #6264
• DirectML support #5240
• Add TensorRT support #6212
• Make onnx model support batch inference #5928
• Slow inference on CPU  #5597

UX improvements

• API for loading ONNX model lacks access to input schema and is not consistent with regular model loading API #4335
• Pass customops.dll to MlContext.Transforms.ApplyOnnxModel to use ops from onnxruntime-extensions #5976

Result discrepancies

• ML.Net incorrect results with ONNX model #6324
• Validation of Onnx input using ApplyOnnxModel #6067
• Using Onnx trained on Keras with PreprocessInput #5830
• Predictions of converted LightGbm ONNX model completely mismatched ML.Net equivalent #5549