[DTensor] ignore fresh unbacked symbols in shard prop#166989
Closed
pianpwk wants to merge 2 commits intogh/pianpwk/29/basefrom
Closed
[DTensor] ignore fresh unbacked symbols in shard prop#166989pianpwk wants to merge 2 commits intogh/pianpwk/29/basefrom
pianpwk wants to merge 2 commits intogh/pianpwk/29/basefrom
Conversation
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/166989
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 5cdeade with merge base 82fa2aa ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
ciflow/inductor
oncall: distributed
| from torch.fx.experimental.proxy_tensor import disable_proxy_modes_tracing | ||
|
|
||
| with FakeTensorMode(), disable_proxy_modes_tracing(): | ||
| fake_mode = detect_fake_mode() or FakeTensorMode() |
Contributor
There was a problem hiding this comment.
should we also initialize dummy shape env here as well? When there is no fake mode from tracing context, below lines will fail with errors like "NoneType doesn't have create_unbacked_symint". It seems to me that even in eager, you would do this fake tensor prop thing right?
pianpwk
changed the title
pianpwk
added
release notes: distributed (dtensor)
pianpwk
changed the title
ezyang
reviewed
Nov 6, 2025
| y = torch.randint(1, (10,)).bool() | ||
| x_dt = distribute_tensor(x, device_mesh, placements=[Replicate()]) | ||
| y_dt = distribute_tensor(y, device_mesh, placements=[Replicate()]) | ||
| _dynamo_graph_capture_for_export(Foo())(x_dt, y_dt) |
ezyang
approved these changes
Nov 6, 2025
Contributor
|
@laithsakka ptal |
laithsakka
reviewed
Nov 6, 2025
|
|
||
| with FakeTensorMode(), disable_proxy_modes_tracing(): | ||
| fake_mode = detect_fake_mode() or FakeTensorMode() | ||
| suppress_fresh_symbols_ctx = ( |
Contributor
There was a problem hiding this comment.
- can you add a comment explaining why ignore_fresh_unbacked_symbols() is safe here.
This fixes 2 issues with the DTensor data-dependent test case: 1) ShapeEnv not found when doing shard prop on data-dependent ops - fix was to detect the outer tracing fake mode. Maybe ShardingPropagator should just own a FakeMode & ShapeEnv for these purposes? The previous behavior was to initialize a new fake mode on every call. 2) Pending unbacked symbols not found. This happens because DTensor dispatch runs fake prop twice, once while figuring out the output sharding: https://github.com/pytorch/pytorch/blob/2bba37309bc8996fc6a190592e5ad9aac53761c9/torch/distributed/tensor/_sharding_prop.py#L175 and again to actually get the resulting local tensor: https://github.com/pytorch/pytorch/blob/2bba37309bc8996fc6a190592e5ad9aac53761c9/torch/distributed/tensor/_dispatch.py#L254-L255 With data-dependent ops, both calls will produce an unbacked symbol, but symbols in the first invocation are never surfaced, producing this error, so we ignore pending symbols from this site. cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
Contributor
Author
|
@pytorchbot merge |
Collaborator
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 |
Khanaksahu
pushed a commit
to Khanaksahu/pytorch
that referenced
this pull request
Nov 17, 2025
ghstack-source-id: 0f80220 Pull Request resolved: pytorch/pytorch#166989
Silv3S
pushed a commit
to Silv3S/pytorch
that referenced
this pull request
Nov 18, 2025
This fixes 2 issues with the DTensor data-dependent test case: 1) ShapeEnv not found when doing shard prop on data-dependent ops - fix was to detect the outer tracing fake mode. Maybe ShardingPropagator should just own a FakeMode & ShapeEnv for these purposes? The previous behavior was to initialize a new fake mode on every call. 2) Pending unbacked symbols not found. This happens because DTensor dispatch runs fake prop twice, once while figuring out the output sharding: https://github.com/pytorch/pytorch/blob/2bba37309bc8996fc6a190592e5ad9aac53761c9/torch/distributed/tensor/_sharding_prop.py#L175 and again to actually get the resulting local tensor: https://github.com/pytorch/pytorch/blob/2bba37309bc8996fc6a190592e5ad9aac53761c9/torch/distributed/tensor/_dispatch.py#L254-L255 With data-dependent ops, both calls will produce an unbacked symbol, but symbols in the first invocation are never surfaced, producing this error, so we ignore pending symbols from this site. Pull Request resolved: pytorch#166989 Approved by: https://github.com/ezyang
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 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.
This fixes 2 issues with the DTensor data-dependent test case:
ShapeEnv not found when doing shard prop on data-dependent ops - fix was to detect the outer tracing fake mode. Maybe ShardingPropagator should just own a FakeMode & ShapeEnv for these purposes? The previous behavior was to initialize a new fake mode on every call.
Pending unbacked symbols not found. This happens because DTensor dispatch runs fake prop twice, once while figuring out the output sharding:
pytorch/torch/distributed/tensor/_sharding_prop.py
Line 175 in 2bba373
pytorch/torch/distributed/tensor/_dispatch.py
Lines 254 to 255 in 2bba373
Stack from ghstack (oldest at bottom):
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @pragupta @msaroufim @dcci