[DTensor][1/N] add forward layer norm support #113105
Closed
Conversation
This file contains hidden or 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
[ghstack-poisoned]
XilunWu
requested review from
H-Huang,
awgu,
d4l3k,
fduwjj,
fegin,
kiukchung,
kwen2501,
mrshenli,
rohan-varma,
wanchaol,
wz337 and
zhaojuanmao
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/113105
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 400ef8f with merge base 56e514a ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
wanchaol
reviewed
Nov 7, 2023
| # must be in form of OpStrategy | ||
| assert isinstance(input_schema, OpStrategy) | ||
| assert isinstance(normalized_shape, (int, Sequence, torch.Size)) | ||
| normalized_size = ( |
There was a problem hiding this comment.
I thought you should call normalize_to_torch_size that's why we refactored that?
There was a problem hiding this comment.
oops, the code was override in rebase...
There was a problem hiding this comment.
note: forward fix normalize_to_torch_size in #113244 to handle the size argument of type int, Sequence, or torch.Size. This fix also enables 3 dtensor op tests.
**Summary**: This PR adds DTensor implementation for ATen op `native_layer_norm`. **Test**: `pytest test/distributed/_tensor/test_dtensor_ops.py -s -k layer_norm` [ghstack-poisoned]
wanchaol
reviewed
Nov 7, 2023
| @@ -1,5 +1,5 @@ | |||
| # Copyright (c) Meta Platforms, Inc. and affiliates | |||
| from typing import cast, List, Optional, Tuple | |||
| from typing import cast, List, Optional, Sequence, Tuple | |||
There was a problem hiding this comment.
In addition to the op level tests, I think we should add a test in test_math_ops.py to use distribute_module to partition a nn.LayerNorm, and mark input as sharded in prepare_input_fn, to make sure things work for the forward case when we apply this to nn.Module, this can be a follow up PR
**Summary**: This PR adds DTensor implementation for ATen op `native_layer_norm`. **Test**: `pytest test/distributed/_tensor/test_dtensor_ops.py -s -k layer_norm` [ghstack-poisoned]
XilunWu
added
ciflow/trunk
pytorchmergebot
pushed a commit
that referenced
this pull request
Nov 8, 2023
…ize to accept single int size (#113244) **Summary**: In #113105 I used the util function `normalize_to_torch_size` to unify the `size` argument that may be in multiple formats. However the use of that function would only handle input of `Sequence[int]` therefore I submit this forward fix to have `normalize_to_torch_size` also able to handle the size argument of type `int` and `torch.Size`. A side product of this fix is it also enables 3 dtensor op tests (check `test_dtensor_ops.py`). **Test**: `pytest test/distributed/_tensor/test_dtensor_ops.py` Pull Request resolved: #113244 Approved by: https://github.com/wanchaol ghstack dependencies: #113105
Skylion007
pushed a commit
to Skylion007/pytorch
that referenced
this pull request
Nov 14, 2023
**Summary**: This PR adds DTensor implementation for ATen op `native_layer_norm`. **Test**: `pytest test/distributed/_tensor/test_dtensor_ops.py -s -k layer_norm` Pull Request resolved: pytorch#113105 Approved by: https://github.com/wanchaol
Skylion007
pushed a commit
to Skylion007/pytorch
that referenced
this pull request
Nov 14, 2023
…ize to accept single int size (pytorch#113244) **Summary**: In pytorch#113105 I used the util function `normalize_to_torch_size` to unify the `size` argument that may be in multiple formats. However the use of that function would only handle input of `Sequence[int]` therefore I submit this forward fix to have `normalize_to_torch_size` also able to handle the size argument of type `int` and `torch.Size`. A side product of this fix is it also enables 3 dtensor op tests (check `test_dtensor_ops.py`). **Test**: `pytest test/distributed/_tensor/test_dtensor_ops.py` Pull Request resolved: pytorch#113244 Approved by: https://github.com/wanchaol ghstack dependencies: pytorch#113105
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Stack from ghstack (oldest at bottom):
Summary:
This PR adds DTensor implementation for ATen op
native_layer_norm.Test:
pytest test/distributed/_tensor/test_dtensor_ops.py -s -k layer_normcc @wanchaol