ONNX: serialization for machine learning models

The first workshop of the consortium

Résumé

Most of machine learning libraries are optimized to train models and not necessarily to use them for fast predictions in online web services. ONNX is a open format to represent machine learned models. First developped for deep learning models, it is now available for scikit-learn models. It achieves compact representation of models and fast computation of predictions on CPU, GPU or ARM processors. The talk introduces three tools to play with ONNX in Python.