Support Empty Input Tensors and > 5 Cat Inputs #7855
Merged
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7855
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit f05849d with merge base 5cbfcdc ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D68523312 |
digantdesai
approved these changes
Jan 23, 2025
mcr229
force-pushed
the
export-D68523312
branch
from
20a2133 to
2da5785
Compare
mcr229
added a commit
to mcr229/executorch
that referenced
this pull request
Jan 24, 2025
Summary: PyTorch's cat.default operator can take in arbitrarily large number of inputs. This is because the input is a Tensor List. XNNPACK however supports largest of 5 input tensors at a time. It is common for > 5 input tensors to be concatenated together, so we should still support cat's with this operation. We can do so by adding a pass which decomposes the Cat operator. The first 5 operators can be concatenated together, and then we recursively inject more concatenate nodes to concatenate the result of the last operation with the next 4 input tensors. Another common design pattern is for Concatenates to start with an empty tensor and then concatenat tensors together into it. This results in some empty tensors as inputs to concatenate. Previously we don't partition inputs with empty tensors. I don't remember what the case was with empty tensors, but it seems to work now, so disabling that partitioner check for now. Perhaps CI will pick up an error if this is indeed erroronous Reviewed By: digantdesai Differential Revision: D68523312
|
This pull request was exported from Phabricator. Differential Revision: D68523312 |
Summary: PyTorch's cat.default operator can take in arbitrarily large number of inputs. This is because the input is a Tensor List. XNNPACK however supports largest of 5 input tensors at a time. It is common for > 5 input tensors to be concatenated together, so we should still support cat's with this operation. We can do so by adding a pass which decomposes the Cat operator. The first 5 operators can be concatenated together, and then we recursively inject more concatenate nodes to concatenate the result of the last operation with the next 4 input tensors. Another common design pattern is for Concatenates to start with an empty tensor and then concatenat tensors together into it. This results in some empty tensors as inputs to concatenate. Previously we don't partition inputs with empty tensors. I don't remember what the case was with empty tensors, but it seems to work now, so disabling that partitioner check for now. Perhaps CI will pick up an error if this is indeed erroronous Reviewed By: digantdesai Differential Revision: D68523312
mcr229
force-pushed
the
export-D68523312
branch
from
2da5785 to
24c593b
Compare
|
This pull request was exported from Phabricator. Differential Revision: D68523312 |
|
@mcr229 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
YIWENX14
pushed a commit
that referenced
this pull request
Jan 28, 2025
Differential Revision: D68523312 Pull Request resolved: #7855
zonglinpeng
pushed a commit
to zonglinpeng/executorch
that referenced
this pull request
Jan 30, 2025
Differential Revision: D68523312 Pull Request resolved: pytorch#7855
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.
Summary:
PyTorch's cat.default operator can take in arbitrarily large number of inputs. This is because the input is a Tensor List. XNNPACK however supports largest of 5 input tensors at a time. It is common for > 5 input tensors to be concatenated together, so we should still support cat's with this operation. We can do so by adding a pass which decomposes the Cat operator. The first 5 operators can be concatenated together, and then we recursively inject more concatenate nodes to concatenate the result of the last operation with the next 4 input tensors.
Another common design pattern is for Concatenates to start with an empty tensor and then concatenat tensors together into it. This results in some empty tensors as inputs to concatenate.
Previously we don't partition inputs with empty tensors. I don't remember what the case was with empty tensors, but it seems to work now, so disabling that partitioner check for now. Perhaps CI will pick up an error if this is indeed erroronous
Differential Revision: D68523312