[dtensor] full_tensor to return synchronously #113322
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
full_tensor API should return synchronously instead of AsyncCollectiveTensor and if the return is that, we do the wait directly, this makes the full_tensor API be more percise [ghstack-poisoned]
wanchaol
requested review from
H-Huang,
LucasLLC,
awgu,
d4l3k,
fduwjj,
fegin,
kiukchung,
kwen2501,
mrshenli,
rohan-varma,
wz337 and
zhaojuanmao
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/113322
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit e098488 with merge base 84d64d7 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
wanchaol
added
ciflow/trunk
full_tensor API should return synchronously instead of AsyncCollectiveTensor and if the return is that, we do the wait directly, this makes the full_tensor API be more percise [ghstack-poisoned]
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Skylion007
pushed a commit
to Skylion007/pytorch
that referenced
this pull request
Nov 14, 2023
full_tensor API should return synchronously instead of AsyncCollectiveTensor and if the return is that, we do the wait directly, this makes the full_tensor API be more percise Pull Request resolved: pytorch#113322 Approved by: https://github.com/wz337
fduwjj
reviewed
Nov 16, 2023
Comment on lines
+63
to
+75
| def forward( # type: ignore[override] | ||
| ctx, | ||
| input: "DTensor", | ||
| grad_placements: Optional[Sequence[Placement]], | ||
| async_output: bool, | ||
| ): | ||
| ctx.dtensor_spec = input._spec | ||
| ctx.grad_placements = grad_placements | ||
| local_tensor = input._local_tensor | ||
| if not async_output and isinstance(local_tensor, funcol.AsyncCollectiveTensor): | ||
| # synchronously wait for any pending collectives to get the result tensor | ||
| local_tensor = local_tensor.trigger_wait() | ||
| local_tensor = local_tensor.elem # type: ignore[attr-defined] |
There was a problem hiding this comment.
I don't have a strong opinion on this, but this might bring some perf regression because for every .to_local we are waiting. But if we think this is the defined or expected behavior, then that's fine.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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):
full_tensor API should return synchronously instead of
AsyncCollectiveTensor and if the return is that, we do the wait
directly, this makes the full_tensor API be more percise