Tutorial

This package is mostly used to validate custom implementations of a specific onnx operator or kernel. The same code is used to either implement a custom kernel for the reference implementation from onnx package or a custom kernel for onnxruntime. The last section describe how to build the package and to add a new implementation depending the technology it relies on (CPU, openmp, CUDA, eigen, …). The last section is a sorted index of the examples.

Kernels

• CReferenceEvaluator
• Cython Binding of onnxruntime
• Custom Kernels for onnxruntime
• Focus on operators optimization
• Many Tools to help investigating issues
• Build from source

Deprecated

• Experiments about parallelization