Skip to content

Conversation

raayandhar
Copy link

More details are in the issue: #4339

Since split.Tensor and split_with_sizes decompose to as_strided (see issue above why as_strided is problematic), we remove them from the decomp tables.

cc @zjgarvey

@zjgarvey
Copy link
Collaborator

You may need to add the recompose pass to the very beginning of lib/Dialect/Torch/Transforms/Passes.cpp::mlir::torch::Torch::createTorchDynamoExportToTorchBackendPipeline

@zjgarvey
Copy link
Collaborator

Hey, can you open a separate PR for the pytorch cxx abi changes? They look great, but make that a standalone patch. We can sync this branch with main when done.

@raayandhar
Copy link
Author

Hey, can you open a separate PR for the pytorch cxx abi changes? They look great, but make that a standalone patch. We can sync this branch with main when done.

Sure, I don't even know if they work though. I can't reproduce locally.

@raayandhar
Copy link
Author

@zjgarvey opened a PR #4345 and reset this back to a better state

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants