List view
TorchDynamo should compose with other things in the PyTorch ecosystem.
No due dateMake TorchDynamo's guards: - more correct: by making sure we guard on the needed things - faster: by guarding on fewer undeeded things and optimizing the implementation
No due dateUse TorchDynamo to build a tool to make training faster.
No due date•15/15 issues closedTorchDynamo doesn't support all of Python, and will fall back to eager mode when it hits something it can't handle. This (ongoing) milestone tracks issues related to capturing more things.
No due date•7/7 issues closedUse TorchDynamo to build a user-facing inference acceleration tool. We are close to this, mainly requiring polish, testing, and documentation.
Overdue by 3 year(s)•Due by May 1, 2022•11/11 issues closed