Cool, there's consensus between APIs on how to expose "tensor cores" now? Very exciting! Although I think that relaxing memory limitations and providing more vi

Cool, there's consensus between APIs on how to expose "tensor cores" now? Very exciting! Although I think that relaxing memory limitations and providing more visibility and control there is even more important for running ML on the web right now. And harder to make progress on because there isn't a single team that clearly owns "all memory management".