#10 Open Source Forum 2019 Shintarou Okada
summary
ONNX is described as a concrete model format for moving trained neural networks between frameworks using converters and a growing model zoo, with multiple runtimes and device backends already deployed in products. The project adds quantization and training support (including gradient hooks) and includes a compiler layer that emits backend code, while governance relies on an operator/versioning working group and faces challenges like operator proliferation, C/Python bindings, and conversion fidelity.
AI-written summary · about
chapters (13)
- 0:00ONNX introduction
- 1:15Vectors and matrices in ONNX
- 3:31ONNX benefits and use cases
- 5:19ONNX openness and compatibility
- 7:18ONNX community and support
- 8:46Getting ONNX models
- 10:14ONNX technical decisions
- 11:20Running ONNX models
- 13:29Practical ONNX applications
- 15:14ONNX success stories
- 17:24ONNX scalability and future
- 18:52ONNX challenges and next steps
- 20:18Summary
related talks
05mONNX Runtime
54mOne Universal Language for All Machine Learning Models
13mKeynote: To Nemotron and Beyond: Building the Open Source AI Ecosystem
05mThe future of frameworks: Navigate the OSS landscape
41mOpen standards for machine learning deployment
33mOpen Standards for Machine Learning Deployment
14mOpen, connected and compliant. How Oniro builds seamless global smart ecosystems
26mBuilding the Next Generation of Open Source AI Tooling