[coor-targets] Accept ProcessGroup directly in functional collective op schemas by aorenste · Pull Request #172795 · pytorch/pytorch

aorenste · 2026-01-19T22:56:35Z

Change _c10d_functional and _dtensor op schemas from str group_name
to Any group so callers can pass either a string name (resolved via
the group registry) or a ProcessGroup object directly, avoiding an
unnecessary registry round-trip.

Each C++ collective gets a ProcessGroup-typed overload (the real
implementation) and a string-typed overload (thin wrapper that resolves
and delegates). New dispatch functions handle the runtime Any by
checking whether the IValue is a string or capsule.

On the Python side, _resolve_group() replaces _resolve_group_name()
and returns ProcessGroup objects when possible. The AOTInductor proxy
executor gains AnyType argument handling.

This is intended to be used by compile-on-one-rank to avoid turning
ProcessGroups into constant strings.

Stack from ghstack (oldest at bottom):

…ring or ProcessGroup inputs [ghstack-poisoned]

pytorch-bot · 2026-01-19T22:56:39Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/172795

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki

Note: Links to docs will display an error until the docs builds have been completed.

⏳ 1 Pending, 1 Unrelated Failure

As of commit 2080c58 with merge base 79546f5 ():

FLAKY - The following job failed but was likely due to flakiness present on trunk:

inductor / unit-test / inductor-test / test (inductor, 1, 2, linux.g5.4xlarge.nvidia.gpu) (gh) (similar failure)
test/inductor/test_torchinductor_opinfo.py::TestInductorOpInfoCUDA::test_comprehensive_pca_lowrank_cuda_float32

This comment was automatically generated by Dr. CI and updates every 15 minutes.

…ring or ProcessGroup inputs ghstack-source-id: 24ef88b Pull Request resolved: #172795

…ake both string or ProcessGroup inputs" [ghstack-poisoned]

…collective op schemas" Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. [ghstack-poisoned]

…op schemas Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. Authored with Claude. ghstack-source-id: 527fedc Pull Request resolved: #172795

…collective op schemas" Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. [ghstack-poisoned]

…op schemas Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. Authored with Claude. ghstack-source-id: 04a2d91 Pull Request resolved: #172795

…collective op schemas" Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. [ghstack-poisoned]

…op schemas Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. Authored with Claude. ghstack-source-id: b906563 Pull Request resolved: #172795

…collective op schemas" Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. [ghstack-poisoned]

…op schemas Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. Authored with Claude. ghstack-source-id: 05d9dbf Pull Request resolved: #172795

…ring or ProcessGroup inputs ghstack-source-id: ae449aa Pull Request resolved: pytorch/pytorch#172795

…collective op schemas" Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. [ghstack-poisoned]

…op schemas Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. Authored with Claude. ghstack-source-id: 94fd9b1 Pull Request resolved: #172795

pytorch-auto-revert · 2026-03-18T03:43:49Z

@pytorchbot revert -m "Reverted automatically by pytorch's autorevert, to avoid this behaviour add the tag autorevert: disable" -c autorevert

This PR is attributed to have caused regression in:

slow: dynamo/test_fake_distributed.py::test_compiled_p2p_interleave_graph (hud), dynamo/test_fake_distributed.py::test_compiled_isend_graph (hud), dynamo/test_fake_distributed.py::test_compiled_fire_and_forget_isend_graph (hud), dynamo/test_fake_distributed.py::test_compiled_fire_and_forget_isend_graph (hud), dynamo/test_fake_distributed.py::test_compiled_batch_isend_irecv_mixed_graph (hud), dynamo/test_fake_distributed.py::test_compiled_batch_isend_irecv_mixed_graph (hud), dynamo/test_fake_distributed.py::test_compiled_isend_graph (hud), dynamo/test_fake_distributed.py::test_compiled_isend_graph (hud), dynamo/test_fake_distributed.py::test_compiled_batch_isend_irecv_mixed_graph (hud), dynamo/test_fake_distributed.py::test_compiled_irecv_graph (hud), dynamo/test_fake_distributed.py::test_compiled_p2p_interleave_graph (hud), dynamo/test_fake_distributed.py::test_compiled_fire_and_forget_isend_graph (hud), dynamo/test_fake_distributed.py::test_compiled_irecv_graph (hud), dynamo/test_fake_distributed.py::test_compiled_p2p_interleave_graph (hud), dynamo/test_fake_distributed.py::test_compiled_irecv_graph (hud)

Please investigate and fix the issues.
@claude Can you please read this revert comment, follow the links and read the errors, to then give a brief diagnostics on the cause of the error? If you judge the error to be legitimate reason for a revert, please provide brief guidance on how the author could fix it.

claude · 2026-03-18T03:44:15Z

Claude finished @pytorch-auto-revert[bot]'s task in 11m 10s —— View job

Diagnosis of Revert for PR #172795

Read failing test code and changed files
Analyze the code paths affected by the PR
Diagnose root cause
Provide fix guidance

Failing Tests

All failures are in test/dynamo/test_fake_distributed.py, class TestFakeDistributedP2P:

test_compiled_isend_graph
test_compiled_fire_and_forget_isend_graph
test_compiled_irecv_graph
test_compiled_batch_isend_irecv_mixed_graph
test_compiled_p2p_interleave_graph

Root Cause

These tests use assertExpectedInline to compare the FX graph output against hardcoded expected strings. All five tests assert that the group argument in the traced graph is the string '0', e.g.:

# From test_compiled_isend_graph (line 213):
tensor: "f32[0]" = torch.ops._c10d_functional.isend(l_tensor_, 1, 0, '0')

The PR changed op schemas from str group_name to Any group across _c10d_functional ops. While the PR description mentions this was done for all functional collectives, the P2P ops (isend, irecv, batch_p2p_ops) were likely also affected.

The critical code path is:

Dynamo intercepts dist.isend()/dist.irecv()/dist.batch_isend_irecv() via CollectiveFunctionRewriteVariable in torch/_dynamo/variables/functions.py:2571-2644
Remapped functions (isend_inplace, irecv_inplace, batch_p2p_ops_inplace) are called during tracing
These functions call _resolve_group_name(group) (line 1777, 1803, 1819) which always returns a string (e.g., '0')
The string is passed to torch.ops._c10d_functional.isend(..., group_name)

The PR introduced _resolve_group() (line 1195) which returns either a ProcessGroup or a string, and the _group_or_group_name() function (line 1831) that also returns either type.

The most likely issue is that the PR changed something in how the P2P compilation path passes the group argument — either:

The P2P ops were switched from _resolve_group_name() to _resolve_group(), which now returns a ProcessGroup object instead of a string '0'. When Dynamo traces this, the FX graph captures the ProcessGroup object differently than a string constant, causing all the assertExpectedInline checks to fail.
The schema change from str to Any caused Dynamo's FX graph printer to represent the group argument differently (e.g., as an IValue rather than a quoted string).

Note: I was unable to access the actual CI logs (the jobs show conclusion: null, likely still running or logs not yet downloadable) or the full PR diff (the PR is closed and fetch commands require additional permissions). However, the diagnosis is high confidence based on code analysis.

Is the Revert Legitimate?

Yes — this is a legitimate revert. The PR broke 5 existing P2P compilation tests that were not updated to match the new graph output format. These tests verify the compiled graph structure for distributed P2P operations, and graph correctness is critical.

How to Fix

The author should:

Update the expected graph strings in the 5 failing tests in test/dynamo/test_fake_distributed.py (lines 206-388) to match the new representation of the group argument after the schema change. If the group argument now appears as e.g. a ProcessGroup capsule reference instead of '0', the expected strings need to reflect that.
Alternatively, if P2P ops should still pass string group names (as suggested by the review comments — the author mentioned P2P changes would come in a follow-up PR), ensure the P2P code paths (isend_inplace, irecv_inplace, batch_p2p_ops_inplace) continue to use _resolve_group_name() which always returns a string, rather than _resolve_group().
Run the tests locally before re-landing:
```
python test/dynamo/test_fake_distributed.py -k "TestFakeDistributedP2P"
```
This will show the actual vs expected graph output, making it straightforward to update the expected strings.

The cleanest fix is likely option (2): keep P2P ops using _resolve_group_name() → string in this PR, and defer the P2P Any group changes to the follow-up PR as originally planned. This would avoid changing the traced graph representation for P2P ops.

pytorchmergebot · 2026-03-18T03:45:38Z

@pytorchbot successfully started a revert job. Check the current status here.
Questions? Feedback? Please reach out to the PyTorch DevX Team

pytorchmergebot · 2026-03-18T03:45:47Z

@aorenste your PR has been successfully reverted.

…lective op schemas (#172795)" This reverts commit 71352fa. Reverted #172795 on behalf of https://github.com/pytorch-auto-revert due to Reverted automatically by pytorch's autorevert, to avoid this behaviour add the tag autorevert: disable ([comment](#172795 (comment)))

…collective op schemas" Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. [ghstack-poisoned]

aorenste · 2026-03-19T00:21:35Z

@pytorchbot merge

pytorchmergebot · 2026-03-19T00:23:45Z

Merge started

Your 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

Advanced Debugging

Check the merge workflow status
here

Pull Request resolved: #177446 Approved by: https://github.com/Skylion007 ghstack dependencies: #172795

…op schemas (pytorch#172795) Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. Pull Request resolved: pytorch#172795 Approved by: https://github.com/angelayi

Pull Request resolved: pytorch#177446 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#172795

…op schemas (pytorch#172795) Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. Pull Request resolved: pytorch#172795 Approved by: https://github.com/angelayi

…lective op schemas (pytorch#172795)" This reverts commit 71352fa. Reverted pytorch#172795 on behalf of https://github.com/pytorch-auto-revert due to Reverted automatically by pytorch's autorevert, to avoid this behaviour add the tag autorevert: disable ([comment](pytorch#172795 (comment)))

…op schemas (pytorch#172795) Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. Pull Request resolved: pytorch#172795 Approved by: https://github.com/angelayi

Pull Request resolved: pytorch#177446 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#172795

…op schemas (pytorch#172795) Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. Pull Request resolved: pytorch#172795 Approved by: https://github.com/angelayi

…lective op schemas (pytorch#172795)" This reverts commit 71352fa. Reverted pytorch#172795 on behalf of https://github.com/pytorch-auto-revert due to Reverted automatically by pytorch's autorevert, to avoid this behaviour add the tag autorevert: disable ([comment](pytorch#172795 (comment)))

…op schemas (pytorch#172795) Change _c10d_functional and _dtensor op schemas from `str group_name` to `Any group` so callers can pass either a string name (resolved via the group registry) or a ProcessGroup object directly, avoiding an unnecessary registry round-trip. Each C++ collective gets a ProcessGroup-typed overload (the real implementation) and a string-typed overload (thin wrapper that resolves and delegates). New dispatch functions handle the runtime `Any` by checking whether the IValue is a string or capsule. On the Python side, `_resolve_group()` replaces `_resolve_group_name()` and returns ProcessGroup objects when possible. The AOTInductor proxy executor gains AnyType argument handling. This is intended to be used by compile-on-one-rank to avoid turning ProcessGroups into constant strings. Pull Request resolved: pytorch#172795 Approved by: https://github.com/angelayi

Pull Request resolved: pytorch#177446 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#172795

[coor-targets] Change functional collectives so they can take both st…

de8a9da

…ring or ProcessGroup inputs [ghstack-poisoned]

aorenste mentioned this pull request Jan 19, 2026

[inductor] Support opaque objects as graph inputs via OpaqueObjectState #172793

Closed

aorenste mentioned this pull request Jan 19, 2026

[coor-targets] Enable ProcessGroup round-trip through JIT via CapsuleType #172794

Closed

pytorch-bot bot added ciflow/inductor release notes: distributed (c10d) release notes category release notes: distributed (dtensor) release notes category labels Jan 19, 2026

aorenste added a commit that referenced this pull request Jan 19, 2026

[coor-targets] Change functional collectives so they can take both st…

41e7b84

…ring or ProcessGroup inputs ghstack-source-id: 24ef88b Pull Request resolved: #172795

Update on "[coor-targets] Change functional collectives so they can t…

92c6789

…ake both string or ProcessGroup inputs" [ghstack-poisoned]

aorenste mentioned this pull request Jan 21, 2026

[coor-targets] Extract ProcessGroups from DTensors as graph inputs for compile_on_one_rank #172959

Closed

Update on "[coor-targets] Change functional collectives so they can t…

0069619

…ake both string or ProcessGroup inputs" [ghstack-poisoned]

aorenste added the suppress-bc-linter Suppresses the failures of API backward-compatibility linter (Lint/bc_linter) label Jan 22, 2026

This was referenced Jan 22, 2026

[coor] patches #173056

Closed

working man #173233

Closed

non strict trace working man #173236

Closed

non strict man #173253

Closed

aorenste changed the title ~~[coor-targets] Change functional collectives so they can take both string or ProcessGroup inputs~~ [coor-targets] Accept ProcessGroup directly in functional collective op schemas Mar 10, 2026

pytorch-bot bot added the ciflow/torchtitan Run TorchTitan integration tests label Mar 10, 2026

pytorch-bot bot added the release notes: inductor (aoti) label Mar 10, 2026

sandy-gags pushed a commit to sandy-gags/pytorch that referenced this pull request Mar 12, 2026

[coor-targets] Change functional collectives so they can take both st…

21f2c89

…ring or ProcessGroup inputs ghstack-source-id: ae449aa Pull Request resolved: pytorch/pytorch#172795

pytorchmergebot added Reverted ci-no-td Do not run TD on this PR labels Mar 18, 2026

pytorchmergebot reopened this Mar 18, 2026

This was referenced Mar 18, 2026

Fix type annotation for _sym_get_coordinate #177446

Closed

Make run_dtensor_rng_op compatible with compile_on_one_rank #177447

Closed

aorenste added 2 commits March 18, 2026 07:55

pytorchmergebot added the merging label Mar 19, 2026

pytorchmergebot closed this in 12bfd2c Mar 19, 2026

pytorchmergebot removed the merging label Mar 19, 2026

pytorchmergebot pushed a commit that referenced this pull request Mar 19, 2026

Fix type annotation for _sym_get_coordinate (#177446)

b8d53c6

Pull Request resolved: #177446 Approved by: https://github.com/Skylion007 ghstack dependencies: #172795

ryanzhang22 pushed a commit to ryanzhang22/pytorch that referenced this pull request Mar 19, 2026

Fix type annotation for _sym_get_coordinate (pytorch#177446)

0b1c3df

Pull Request resolved: pytorch#177446 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#172795

EmanueleCoradin pushed a commit to EmanueleCoradin/pytorch that referenced this pull request Mar 30, 2026

Fix type annotation for _sym_get_coordinate (pytorch#177446)

ced06b9

Pull Request resolved: pytorch#177446 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#172795

AaronWang04 pushed a commit to AaronWang04/pytorch that referenced this pull request Mar 31, 2026

Fix type annotation for _sym_get_coordinate (pytorch#177446)

66fedb1

Pull Request resolved: pytorch#177446 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#172795

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[coor-targets] Accept ProcessGroup directly in functional collective op schemas#172795

[coor-targets] Accept ProcessGroup directly in functional collective op schemas#172795
aorenste wants to merge 13 commits intogh/aorenste/184/basefrom
gh/aorenste/184/head

aorenste commented Jan 19, 2026 •

edited

Loading

Uh oh!

pytorch-bot bot commented Jan 19, 2026 •

edited

Loading

Uh oh!

pytorch-auto-revert bot commented Mar 18, 2026

Uh oh!

claude bot commented Mar 18, 2026 •

edited

Loading

Uh oh!

pytorchmergebot commented Mar 18, 2026

Uh oh!

pytorchmergebot commented Mar 18, 2026

Uh oh!

aorenste commented Mar 19, 2026

Uh oh!

pytorchmergebot commented Mar 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

aorenste commented Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/172795

⏳ 1 Pending, 1 Unrelated Failure

Uh oh!

pytorch-auto-revert bot commented Mar 18, 2026

Uh oh!

claude bot commented Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Diagnosis of Revert for PR #172795

Failing Tests

Root Cause

Is the Revert Legitimate?

How to Fix

Uh oh!

pytorchmergebot commented Mar 18, 2026

Uh oh!

pytorchmergebot commented Mar 18, 2026

Uh oh!

aorenste commented Mar 19, 2026

Uh oh!

pytorchmergebot commented Mar 19, 2026

Merge started

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

aorenste commented Jan 19, 2026 •

edited

Loading

pytorch-bot bot commented Jan 19, 2026 •

edited

Loading

claude bot commented Mar 18, 2026 •

edited

Loading