Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds support for forward-mode and higher-order differentiation in PyTorch autograd engine. Additionally, it enables integration with
torch.func
, which implements JAX-style composable functional transforms in PyTorch, allowing for a flexible and elegant use of different types of automatic differentiation (+vectorizing map functionality also implemented in this PR).Integration with
torch.func
breaks compatibility with PyTorch < 2.0, however, so we need to estimate if it's safe to drop support for it at this moment.To do: