Data Skeptic Onnx
Jul 22, 2019
Prasanth Pulavarthi, Product Management Lead at Microsoft for AI frameworks, dives into the transformative ONNX format for deep learning models. He discusses how ONNX promotes model interoperability across various frameworks like TensorFlow and PyTorch, making tech accessible for all. Prasanth highlights the challenges of deploying models like BERT and the efficiencies of Protocol Buffers. He also shares the benefits of using ONNX Runtime for optimizing performance, containerization with Docker, and enhancing deployment flexibility.
AI Snips
Chapters
Transcript
Episode notes
ONNX: A Universal Format
- ONNX is an open standard format for machine learning models, enabling interoperability.
- It allows models trained in various frameworks (like TensorFlow, PyTorch) to be converted for portability.
Optimize Tools with ONNX
- Use ONNX to choose the best tools for training and deployment.
- Prioritize flexibility during training, and focus on performance during deployment.
ONNX at Microsoft
- Microsoft uses ONNX across various products (Bing, Office, Windows).
- This addresses the challenge of deploying models built with different frameworks efficiently.