Tutorials

ONNX ecosystem

Following tutorials introduce the ONNX ecosystem. It walk the user through the ONNX specficiations, how to execute an ONNX graph, how to create an ONNX graph, how to convert a model from scikit-learn, and how to train them with onnxruntime-training.

• Introduction to ONNX
  • ONNX Concepts
  • ONNX with Python
  • Challenges
  • ONNX operators and function
• Introduction to onnxruntime
  • OrtValue
  • Inference with onnxruntime in Python
  • Training with onnxruntime
  • Extensions
  • Quantization with onnxruntime
• scikit-learn to ONNX Tutorial
  • The easy case
  • Using converter from other libraries
  • A custom converter for a custom model
  • Extend ONNX, extend runtime
  • Complex Scenarios and discepancies
• Training Tutorial
  • Full Training with OrtGradientOptimizer
  • Partial Training with OrtGradientForwardBackwardOptimizer
• Benchmarking and profiling Tutorial
  • Study behavior of one operator
  • Inference
  • Profiling
  • Training
• Parallelization
  • Multithreading with onnxruntime
  • Multithreading with onnxruntime and big models

Readings

• Add AI to mobile applications with Xamarin and ONNX Runtime

• Announcing ONNX Runtime Availability in the NVIDIA Jetson Zoo for High Performance Inferencing (8/2021)

• Speeding Up Deep Learning Inference Using TensorFlow, ONNX, and NVIDIA TensorRT (7/2021)

• Journey to optimize large scale transformer model inference with ONNX Runtime (6/2021)

• Accelerating Model Training with the ONNX Runtime (5/2020)

• Accelerate and simplify Scikit-learn model inference with ONNX Runtime (12/2020)

• Model Persistence scikit-learn and ONNX, short talk at scikit-learn foundation (2019)

Current documention of ONNX and onnxruntime

Most of the documentation related on onnx and onnxruntime is written on markdown. The following section is an attempt to render it and make it searchable.

Build

Some useful pages.

• Build onnxruntime on WSL (Windows Linux Subsystem) (2021).