Implement conversion for aten.expand by using ttnn.repeat #21
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
* Add retention of shape metadata information for Placeholder nodes since transform() removes them * Add missing metadata for conversions done under ReplaceMoreTtManually * Add logic to insert layout changes before and/or after ttnn.repeat
| # Replace more patterns with torch.fx.Transformer | ||
| gm = ReplaceMoreTt(gm).transform() | ||
|
|
||
| # Restore metadata for Placeholder nodes |
There was a problem hiding this comment.
From my investigations so far, it could be an oversight. I found some very recent discussion https://discuss.pytorch.org/t/understanding-torch-fx-traceback-preserve-node-meta/205319 with the same issue but unfortunately no response yet. I could create an issue to track this actually. Could be something we optimize away at a later time if we find a better solution.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Problem description
Transform aten.expand by using ttnn.repeat.
What's changed
The interpretation of the
shapeargument between the two ops differ. For aten.expand theshapeargument is the desired output shape. For ttnn.repeat theshapeargument is a multiplier of the input shape instead. Therefore the shape information of the input tensor is needed.In addition, ttnn.repeat acts differently between TILE and ROW_MAJOR layouts. In most cases, we would need to change the layout of a node to ROW_MAJOR before passing that to ttnn.repeat. Similarly, we would need to check if a layout change back to TILE is needed if any following node that uses ttnn.repeat is a tt compute node.