Python Bindings for ONNX Runtime

ONNX Runtime enables high-performance evaluation of trained machine learning (ML) models while keeping resource usage low. Building on Microsoft’s dedication to the Open Neural Network Exchange (ONNX) community, it supports traditional ML models as well as Deep Learning algorithms in the ONNX-ML format.


The core library is implemented in C++. ONNX Runtime is available on PyPi for Linux Ubuntu 16.04, Python 3.5+ for both CPU and GPU. This example demonstrates a simple prediction for an ONNX-ML format model. The following file model.onnx is taken from github onnx…test_sigmoid.

import onnxruntime as rt
sess = rt.InferenceSession("model.onnx")
input_name = sess.get_inputs()[0].name
X = numpy.random.random((3,4,5)).astype(numpy.float32)
res =[output_name], {input_name: x})
pred_onnx =, {input_name: X})