onnx/onnxmltools

What it solves

ONNXMLTools provides a unified way to convert machine learning models from various frameworks and toolkits into the Open Neural Network Exchange (ONNX) format. This allows models trained in different environments to be used across different backends for prediction and deployment.

How it works

The library acts as a conversion layer, often wrapping other specialized converters (like tf2onnx or skl2onnx). It provides specific functions (e.g., convert_keras, convert_coreml, convert_h2o) to translate the operators of a source model into their ONNX equivalents. Users can specify a target_opset to ensure compatibility with specific ONNX releases.

Who it’s for

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

Highlights

• Broad Framework Support: Supports a wide array of toolkits including TensorFlow, scikit-learn, Core ML, LightGBM, XGBoost, H2O, and CatBoost.
• Opset Version Control: Allows users to define the target operator set version to maintain backwards compatibility.
• Simplified API: Provides a high-level interface for converting models from different sources with minimal code.
• Cros-Backend Compatibility: Enables models to be run on any backend that supports the ONNX runtime.