PyTorch Mobile: Past, present and 2.x future

on YouTube ↗09m

" summary

PyTorch Mobile claims production-grade on-device ML with real results: released three years ago, it runs across Meta apps with 50 models doing 70 billion daily inferences, plus prior work on binary-size reduction and runtime performance. The 2.x roadmap centers on four principles: preserve PyTorch authoring semantics, enable user extensibility via narrow, verifiable entry points, provide a portable small runtime targeting phones to microcontrollers, and ship compiler backends plus an integrated developer SDK for profiling and debugging. Expected rollout late summer next year.

AI-written summary · about

" chapters (6)

1. 0:00Introduction to on-device ML
2. 0:53Benefits of edge AI
3. 2:08PyTorch mobile success
4. 2:49Challenges in edge computing
5. 4:28The future of PyTorch on edge
6. 5:54New software stack principles

" related talks

• 08mPyTorch MobileDavid Reiss · PyTorch Developer Day 2020
• 09mPyTorch MobileDavid Reiss · PyTorch Developer Conference 2019
• 03mRecapPyTorch Developer Conference 2019
• 12mWhat’s new in PyTorch 1.3Lin Qiao · PyTorch Developer Conference 2019
• 23mBringing ExecuTorch To the Next Frontiers of Edge AIMergen Nachin · PyTorch Conference Europe 2026
• 10mMachine Learning for mobileKemal El Moujahid · Google's Machine Learning Virtual Community Day 2021
• 06mPyTorch Mobile and Android Neural Networks APIOli Gaymond · PyTorch Developer Day 2020
• 54mKeynote: PyTorch 2.1 Technical Deep DiveJoe Spisak, Peng Wu, Mergen Nachin, Mark Saroufim, Yanan Cao, Will Constable, Mario Lezcano · PyTorch Conference 2023