onnx/onnxmltools

What it solves

ONNXMLTools provides a unified way to convert machine learning models from various frameworks into the Open Neural Network Exchange (ONNX) format. This allows models trained in different toolkits to be used across different backends for inference, improving interoperability between different ML toolkits.

How it works

The library acts as a wrapper and converter for multiple ML toolkits. It converts each operator in a source model to the corresponding ONNX format individually, determining the final model's operator set (opset) version based on the maximal version of the operators used.

Who it’s for

Developers and ML engineers who need to migrate models from frameworks like TensorFlow, scikit-learn, Core ML, LightGBM, XGBoost, H2O, and CatBoost into the ONNX format for deployment or cross-platform compatibility.

Highlights

• Broad Framework Support: Supports conversion from TensorFlow, scikit-learn, Apple Core ML, Spark ML, LightGBM, libsvm, XGBoost, H2O, and CatBoost.
• Opset Version Control: Allows users to specify a target_opset to ensure compatibility with specific ONNX releases.
• Simplified API: Provides dedicated conversion functions (e.g., convert_keras, convert_coreml, convert_h2o) for a streamlined workflow.
• Interoperability: Enables models to be used with any ONNX-compatible runtime, such as ONNX Runtime.