AllToAll implementation #5705

Priya2698 · 2025-12-18T00:39:56Z

No description provided.

github-actions · 2026-01-07T20:26:57Z

Review updated until commit 27fa2d0

Description

Implement AllToAll communication primitive for multi-device operations
Add lowerToAllToAll function to handle AllToAll communication lowering
Add postAllToAll function using alltoall_base backend operation
Add test coverage for expert parallelism AllToAll patterns (dispatch/combine)

Changes walkthrough

Relevant files

Enhancement

lower_to_communication.cpp `Implement AllToAll communication lowering` csrc/host_ir/lower_to_communication.cpp Added lowerToAllToAll function for AllToAll communication lowering Modified getCommunicationInfo to handle AllToAll when logical IDs differ Updated getCommunicationLayout to include AllToAll in non-reordering types Added AllToAll case to convertSingleOpToCommunication	+49/-3
communication.cpp `Implement AllToAll communication posting` csrc/multidevice/communication.cpp Added AllToAll case to CommunicationType operator<< Updated hasRoot to return false for AllToAll (no root process) Updated isReduction to return false for AllToAll Implemented postAllToAll function using alltoall_base Added AllToAll case to postSingleCommunication	+47/-4
c10d_mock.h `Add AllToAll backend interface` csrc/multidevice/c10d_mock.h Added AllToAllOptions struct definition Added alltoall_base method to Backend class mock	+11/-0
communication.h `Add AllToAll to communication types` csrc/multidevice/communication.h Added AllToAll to CommunicationType enum	+2/-1

Tests

test_communication.py `Add AllToAll test coverage` tests/python/multidevice/test_communication.py Added test_alltoall function for expert parallelism patterns Tests both dispatch (axis 0→1) and combine (axis 1→0) patterns Validates tensor sharding and communication correctness	+31/-0

PR Reviewer Guide

Here are some key observations to aid the review process:

🧪 PR contains tests
⚡ Recommended focus areas for review
Missing SendRecv enum The code had switch cases handling SendRecv communication type, but SendRecv was not defined in the CommunicationType enum. This PR fixes this by adding SendRecv to the enum, but this suggests there may have been existing code paths that were unreachable or causing issues. case BinaryOpType::FMin: case BinaryOpType::Min: AllToAll implementation assumptions The postAllToAll function assumes even splits of input/output tensors and uses tensor_split with team_size. This may not handle cases where tensor dimensions are not evenly divisible by team size. Consider adding validation or handling for uneven splits. auto input_splits = at::tensor_split( flattened_input_tensor, communication->team_size(), /dim=/0); auto output_splits = at::tensor_split( flattened_output_tensor, communication->team_size(), /dim=/0); AllToAll layout constraints The comment indicates AllToAll doesn't support reordering and only works when input/output don't require reordering. This limitation should be clearly documented and potentially enforced with runtime checks to prevent silent incorrect behavior. // Note: We do not yet reorder for AllToAll and only support cases where the // input and output do not require any reordering.

Priya2698 · 2026-01-10T02:13:31Z

!test

Priya2698 · 2026-01-13T02:09:11Z

csrc/host_ir/lower_to_communication.cpp

  // For the following communication types, the sharded_id does not have to be
  // outermost in allocation domain. Nonetheless, `tv` still needs to be
  // contiguous and therefore .contiguous() at the beginning of this function.
+  // TODO(prmishra): Fix the layout for AllToAll.


Depending on where we relayout the input/output to be compliant with alltoall requirements, this function and potentially ReorderShardedAxisPass will be affected. I will do it in a following PR once the current PR has been reviewed and we agree on the approach for reordering input/output of alltoall

greptile-apps · 2026-01-13T02:16:51Z

Greptile Summary

Implements AllToAll collective communication for expert parallelism patterns where sharding dimensions change between producer and consumer tensors. The implementation detects AllToAll when both producer and consumer have sharded dimensions but the logical dimension mapping differs (dimension transpose). The runtime operation uses alltoall_base with flattened tensors split evenly across the team.

Added CommunicationType::AllToAll enum value and corresponding handling across communication pipeline
Detection logic in getCommunicationInfo: identifies AllToAll when p_sharded && c_sharded && c_logical_id != p2c_map[p_logical_id]
lowerToAllToAll validates sender/receiver meshes are 1D and identical, creates Communication IR node
postAllToAll flattens tensors, splits by team size, and calls c10d::alltoall_base
Test validates dispatch/combine patterns with parameterized axis configurations
Mock implementation added to support testing without actual distributed backend

Confidence Score: 4/5

This PR is safe to merge with normal review attention
Implementation follows established patterns for other communication primitives and includes test coverage. The detection logic correctly identifies dimension transposition cases. AllToAll can only be triggered when both meshes have size > 1 based on the p_sharded && c_sharded guard, so edge cases around single-device teams cannot occur. The comment at line 484 notes layout handling limitations but indicates this is acceptable for the current use case
No files require special attention - implementation is consistent with existing communication patterns

Important Files Changed

Filename	Overview
csrc/multidevice/communication.h	Added `AllToAll` enum value to `CommunicationType` - straightforward enum extension
csrc/multidevice/c10d_mock.h	Added `AllToAllOptions` struct and `alltoall_base` mock method - standard mock implementation matching c10d API
tests/python/multidevice/test_communication.py	Added parameterized test for AllToAll dispatch/combine patterns in expert parallelism - validates correctness of implementation
csrc/host_ir/lower_to_communication.cpp	Added AllToAll detection logic and lowering function - detects dimension transpose in sharding, validates 1D meshes match
csrc/multidevice/communication.cpp	Implemented `postAllToAll` function using `alltoall_base` with flattened tensors and even splits - validates contiguity, splits tensors evenly

Sequence Diagram

sequenceDiagram
    participant User as User Code
    participant FD as FusionDefinition
    participant Detect as getCommunicationInfo
    participant Lower as lowerToAllToAll
    participant Post as postAllToAll
    participant Backend as c10d::Backend

    User->>FD: define tensor with mesh_x sharding
    User->>FD: add output with different axis sharded
    FD->>Detect: analyze producer/consumer sharding
    
    Note over Detect: Check p_sharded && c_sharded
    Note over Detect: c_logical_id != p2c_map[p_logical_id]
    Detect->>Detect: Identify as AllToAll
    
    Detect->>Lower: lowerToAllToAll(input_tv, output_tv)
    Lower->>Lower: Validate sender/receiver meshes are 1D and equal
    Lower->>Lower: Create Communication(AllToAll, team=mesh.vector())
    
    Note over Post: At execution time
    Post->>Post: Validate input/output contiguity
    Post->>Post: viewAsCompact(input_tensor)
    Post->>Post: viewAsCompact(output_tensor)
    Post->>Post: tensor_split by team_size()
    Post->>Post: assertBuffersHaveSameSize
    Post->>Backend: alltoall_base(output, input, empty, empty)
    Backend-->>Post: Work handle
    Post-->>User: Result tensor

greptile-apps

_{4 files reviewed, 4 comments}

_{Edit Code Review Agent Settings | Greptile}

csrc/multidevice/communication.cpp

csrc/host_ir/lower_to_communication.cpp

tests/python/multidevice/test_multidevice.py

greptile-apps

_{4 files reviewed, 4 comments}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-01-13T02:30:11Z

csrc/multidevice/communication.cpp

+  // input_tv = [DIDx(d), n/d, m, ...]
+  // output_tv = [n, DIDx(d), m/d, ...]
+  // `n`: gathered dimension
+  // `m`: scattered dimension
+  // For alltoall correctness, we split `m` and reorder as [DIDx(d), d, n/d,
+  // m/d, ...] such that alltoall_base splits across the `d` dimension.


The comment describing the input/output tensor shapes appears to be inconsistent with the actual implementation below.

The comment states:

input_tv = [DIDx(d), n/d, m, ...]

m: scattered dimension (at position 2)

But the code at lines 644-651 operates on dimension 1 (input_sizes.at(1)), treating it as the scattered dimension. Based on the test case in test_alltoall, the actual runtime input tensor shape is [n/d, m, k, ...] (after permutation), where:

Position 0: n/d (gathered dimension)

Position 1: m (scattered dimension)

Position 2+: other dimensions

The comment should be updated to reflect the actual runtime tensor layout rather than the logical sharding representation, or clarify that it's describing logical sharding rather than physical memory layout.

greptile-apps · 2026-01-13T02:30:12Z

csrc/host_ir/lower_to_communication.cpp

+        if (c_logical_id == p2c_map.at(p_logical_id)) {
+          fill_communication_info(
+              CommunicationType::SendRecv, p_logical_id, c_logical_id);
+        } else {
+          fill_communication_info(
+              CommunicationType::AllToAll, nullptr, nullptr);
+        }


Setting both p_sharded_id and c_sharded_id to nullptr for AllToAll communication may cause issues in other parts of the codebase that expect valid IterDomain pointers.

For example, getCommunicationLayout() (lines 476-521) is called with sharded_id as a parameter, and at line 493 it checks posInDomain(layout.allocation_domain(), sharded_id), which could fail if sharded_id is nullptr.

The current workaround is that AllToAll is explicitly handled in getCommunicationLayout() at line 489 to return early, but this creates a tight coupling and fragile dependency. Consider either:

Storing the actual producer and consumer sharded IDs for AllToAll (even though they map to different logical dimensions)

Adding explicit nullptr checks in all functions that consume CommunicationInfo

Documenting this assumption clearly in the CommunicationInfo struct definition

csrc/host_ir/lower_to_communication.cpp

greptile-apps · 2026-01-13T02:30:14Z

csrc/multidevice/communication.cpp

+  NVF_CHECK(
+      input_sizes.at(1) % d == 0,
+      "Scattered dimension must be divisible by the team size");


The error message says "Scattered dimension must be divisible by the team size" but this check only validates one dimension.

For a complete AllToAll operation, both the scattered dimension (at position 1 in the runtime tensor) and the gathered dimension (at position 0) should be validated:

The scattered dimension input_sizes.at(1) must be divisible by d (currently checked ✓)

The gathered dimension should result in an output where the gathered size equals d * input_sizes.at(0)

Consider adding validation for the output tensor dimensions as well to catch shape mismatches early, rather than relying on assertBuffersHaveSameSize at line 668 which only provides a generic error.

tests/python/multidevice/test_multidevice.py

csrc/multidevice/communication.cpp

greptile-apps

_{4 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-01-16T00:39:54Z

csrc/multidevice/communication.cpp

+c10::intrusive_ptr<c10d::Work> postAllToAll(
+    Communication* communication,
+    DeviceIdxType my_device_index,
+    c10d::Backend* backend,
+    at::Tensor input_tensor,
+    at::Tensor output_tensor) {
+  NVF_ERROR(
+      isTvContiguous(communication->in()),
+      "Input tensor is not contiguous: ",
+      communication->in(),
+      " contiguity: ",
+      communication->in()->domain()->getContiguityString());
+  NVF_ERROR(
+      isTvContiguous(communication->out()),
+      "Output tensor is not contiguous: ",
+      communication->out(),
+      " contiguity: ",
+      communication->out()->domain()->getContiguityString());
+
+  auto flattened_input_tensor = viewAsCompact(input_tensor);
+  auto flattened_output_tensor = viewAsCompact(output_tensor);
+
+  // alltoall_base requires even splits of the input and output tensors.
+  auto input_splits = at::tensor_split(
+      flattened_input_tensor, communication->team_size(), /*dim=*/0);
+  auto output_splits = at::tensor_split(
+      flattened_output_tensor, communication->team_size(), /*dim=*/0);
+  assertBuffersHaveSameSize(input_splits, output_splits);
+
+  std::vector<int64_t> empty_split_sizes;
+  return backend->alltoall_base(
+      flattened_output_tensor,
+      flattened_input_tensor,
+      empty_split_sizes,
+      empty_split_sizes,
+      /*options=*/{});
+}


logic: Missing team_size == 1 edge case handling. Other communication functions like postBroadcast (line 347) and postSendRecv (line 582) handle single-device teams by either returning early or doing local copies. AllToAll should handle this case to avoid unnecessary communication overhead and potential runtime errors.

Suggested change

c10::intrusive_ptr<c10d::Work> postAllToAll(

Communication* communication,

DeviceIdxType my_device_index,

c10d::Backend* backend,

at::Tensor input_tensor,

at::Tensor output_tensor) {

NVF_ERROR(

isTvContiguous(communication->in()),

"Input tensor is not contiguous: ",

communication->in(),

" contiguity: ",

communication->in()->domain()->getContiguityString());

NVF_ERROR(

isTvContiguous(communication->out()),

"Output tensor is not contiguous: ",

communication->out(),

" contiguity: ",

communication->out()->domain()->getContiguityString());

auto flattened_input_tensor = viewAsCompact(input_tensor);

auto flattened_output_tensor = viewAsCompact(output_tensor);

// alltoall_base requires even splits of the input and output tensors.

auto input_splits = at::tensor_split(

flattened_input_tensor, communication->team_size(), /*dim=*/0);

auto output_splits = at::tensor_split(

flattened_output_tensor, communication->team_size(), /*dim=*/0);

assertBuffersHaveSameSize(input_splits, output_splits);

std::vector<int64_t> empty_split_sizes;

return backend->alltoall_base(

flattened_output_tensor,

flattened_input_tensor,

empty_split_sizes,

empty_split_sizes,

/*options=*/{});

}

c10::intrusive_ptr<c10d::Work> postAllToAll(

Communication* communication,

DeviceIdxType my_device_index,

c10d::Backend* backend,

at::Tensor input_tensor,

at::Tensor output_tensor) {

if (communication->team().size() == 1) {

doLocalCopy(output_tensor, input_tensor);

return nullptr;

}

NVF_ERROR(

isTvContiguous(communication->in()),

"Input tensor is not contiguous: ",

communication->in(),

" contiguity: ",

communication->in()->domain()->getContiguityString());

NVF_ERROR(

isTvContiguous(communication->out()),

"Output tensor is not contiguous: ",

communication->out(),

" contiguity: ",

communication->out()->domain()->getContiguityString());

auto flattened_input_tensor = viewAsCompact(input_tensor);

auto flattened_output_tensor = viewAsCompact(output_tensor);

// alltoall_base requires even splits of the input and output tensors.

auto input_splits = at::tensor_split(

flattened_input_tensor, communication->team_size(), /*dim=*/0);

auto output_splits = at::tensor_split(

flattened_output_tensor, communication->team_size(), /*dim=*/0);

assertBuffersHaveSameSize(input_splits, output_splits);

std::vector<int64_t> empty_split_sizes;

return backend->alltoall_base(

flattened_output_tensor,

flattened_input_tensor,

empty_split_sizes,

empty_split_sizes,

/*options=*/{});

}

wujingyue

LGTM otherwise

wujingyue · 2026-01-16T00:40:52Z

csrc/host_ir/lower_to_communication.cpp

-        fill_communication_info(
-            CommunicationType::SendRecv, p_logical_id, c_logical_id);
+
+        if (c_logical_id == p2c_map.at(p_logical_id)) {


As long as tests pass, I'm fine with this condition. But can you tell me the math behind it? If the two DIDs are mapped, isn't that a non-communication?

The implementation of send recv assumes that the input and output are sharded on the same dimension, but have different meshes.

There is isResharding check at the beginning.

Got it -- we'll have to think about how to represent CollectivePermute for ring-based communication.

csrc/host_ir/lower_to_communication.cpp

Priya2698 · 2026-01-16T01:09:27Z

!test

greptile-apps

_{4 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-01-16T01:11:45Z

csrc/multidevice/communication.cpp

+c10::intrusive_ptr<c10d::Work> postAllToAll(
+    Communication* communication,
+    DeviceIdxType my_device_index,
+    c10d::Backend* backend,
+    at::Tensor input_tensor,
+    at::Tensor output_tensor) {
+  NVF_ERROR(
+      isTvContiguous(communication->in()),
+      "Input tensor is not contiguous: ",
+      communication->in(),
+      " contiguity: ",
+      communication->in()->domain()->getContiguityString());
+  NVF_ERROR(
+      isTvContiguous(communication->out()),
+      "Output tensor is not contiguous: ",
+      communication->out(),
+      " contiguity: ",
+      communication->out()->domain()->getContiguityString());
+
+  auto flattened_input_tensor = viewAsCompact(input_tensor);
+  auto flattened_output_tensor = viewAsCompact(output_tensor);
+
+  // alltoall_base requires even splits of the input and output tensors.
+  auto input_splits = at::tensor_split(
+      flattened_input_tensor, communication->team_size(), /*dim=*/0);
+  auto output_splits = at::tensor_split(
+      flattened_output_tensor, communication->team_size(), /*dim=*/0);
+  assertBuffersHaveSameSize(input_splits, output_splits);
+
+  std::vector<int64_t> empty_split_sizes;
+  return backend->alltoall_base(
+      flattened_output_tensor,
+      flattened_input_tensor,
+      empty_split_sizes,
+      empty_split_sizes,
+      /*options=*/{});
+}


logic: missing team_size == 1 edge case - when team size is 1, no communication needed, just local copy (see postBroadcast at line 347 or postSendRecv at line 582)

Suggested change

c10::intrusive_ptr<c10d::Work> postAllToAll(

Communication* communication,

DeviceIdxType my_device_index,

c10d::Backend* backend,

at::Tensor input_tensor,

at::Tensor output_tensor) {

NVF_ERROR(

isTvContiguous(communication->in()),

"Input tensor is not contiguous: ",

communication->in(),

" contiguity: ",

communication->in()->domain()->getContiguityString());

NVF_ERROR(

isTvContiguous(communication->out()),

"Output tensor is not contiguous: ",

communication->out(),

" contiguity: ",

communication->out()->domain()->getContiguityString());

auto flattened_input_tensor = viewAsCompact(input_tensor);

auto flattened_output_tensor = viewAsCompact(output_tensor);

// alltoall_base requires even splits of the input and output tensors.

auto input_splits = at::tensor_split(

flattened_input_tensor, communication->team_size(), /*dim=*/0);

auto output_splits = at::tensor_split(

flattened_output_tensor, communication->team_size(), /*dim=*/0);

assertBuffersHaveSameSize(input_splits, output_splits);

std::vector<int64_t> empty_split_sizes;

return backend->alltoall_base(

flattened_output_tensor,

flattened_input_tensor,

empty_split_sizes,

empty_split_sizes,

/*options=*/{});

}

c10::intrusive_ptr<c10d::Work> postAllToAll(

Communication* communication,

DeviceIdxType my_device_index,

c10d::Backend* backend,

at::Tensor input_tensor,

at::Tensor output_tensor) {

if (communication->team_size() == 1) {

doLocalCopy(output_tensor, input_tensor);

return nullptr;

}

NVF_ERROR(

isTvContiguous(communication->in()),

"Input tensor is not contiguous: ",

communication->in(),

" contiguity: ",

communication->in()->domain()->getContiguityString());

NVF_ERROR(

isTvContiguous(communication->out()),

"Output tensor is not contiguous: ",

communication->out(),

" contiguity: ",

communication->out()->domain()->getContiguityString());

auto flattened_input_tensor = viewAsCompact(input_tensor);

auto flattened_output_tensor = viewAsCompact(output_tensor);

// alltoall_base requires even splits of the input and output tensors.

auto input_splits = at::tensor_split(

flattened_input_tensor, communication->team_size(), /*dim=*/0);

auto output_splits = at::tensor_split(

flattened_output_tensor, communication->team_size(), /*dim=*/0);

assertBuffersHaveSameSize(input_splits, output_splits);

std::vector<int64_t> empty_split_sizes;

return backend->alltoall_base(

flattened_output_tensor,

flattened_input_tensor,

empty_split_sizes,

empty_split_sizes,

/*options=*/{});

}

wujingyue · 2026-01-16T01:47:38Z

csrc/multidevice/communication.cpp

    default:
      NVF_THROW("unrecognized CommunicationType: ", type);


Suggested change

default:

NVF_THROW("unrecognized CommunicationType: ", type);

trust -Wswitch

wujingyue · 2026-01-16T01:48:21Z

csrc/multidevice/communication.cpp

    default:
      NVF_THROW("unrecognized CommunicationType: ", type);


Suggested change

default:

NVF_THROW("unrecognized CommunicationType: ", type);

You may need an std::unreachable() after the switch.

wujingyue · 2026-01-16T01:48:25Z

csrc/multidevice/communication.cpp

+    case CommunicationType::AllToAll:
      return false;
    default:
      NVF_THROW("unrecognized CommunicationType: ", type);


Priya2698 · 2026-01-16T03:01:53Z

!test

greptile-apps

_{4 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-01-16T03:04:05Z

csrc/multidevice/communication.cpp

  }
 }
+
+c10::intrusive_ptr<c10d::Work> postAllToAll(


logic: missing team_size == 1 edge case - when only one device, no communication needed, should return early after local copy (like postBroadcast at line 344)

wujingyue · 2026-01-16T06:30:21Z

For the igpu failure, update csrc/multidevice/c10d_mock.h

Priya2698 · 2026-01-16T07:27:59Z

For the igpu failure, update csrc/multidevice/c10d_mock.h

Thanks!

greptile-apps · 2026-01-16T07:30:54Z

Greptile's behavior is changing!

From now on, if a review finishes with no comments, we will not post an additional "statistics" comment to confirm that our review found nothing to comment on. However, you can confirm that we reviewed your changes in the status check section.

_{This feature can be toggled off in your Code Review Settings by deselecting "Create a status check for each PR".}

Priya2698 · 2026-01-16T08:03:41Z

!build

Priya2698 added 5 commits December 10, 2025 14:53

all2all wip impl

caefdd1

Merge remote-tracking branch 'origin/main' into pm/all2all

f6c1684

attempt 3d -- not working

12cc822

working alltoall example

0370b89

Merge remote-tracking branch 'origin/main' into pm/all2all

06ea908

Priya2698 added 2 commits January 9, 2026 11:45

working all2all with permute

f0763e5

add comment

2bb6353

Priya2698 added 3 commits January 12, 2026 18:00

Merge remote-tracking branch 'origin/main' into pm/all2all

a3c6a47

Merge remote-tracking branch 'origin/main' into pm/all2all

f0626d8

use tv to shard

07b51ad

Priya2698 changed the title ~~AllToAll Draft~~ AllToAll lowering Jan 13, 2026

Priya2698 commented Jan 13, 2026

View reviewed changes

Priya2698 changed the title ~~AllToAll lowering~~ AllToAll implementation Jan 13, 2026

Priya2698 marked this pull request as ready for review January 13, 2026 02:09

greptile-apps bot reviewed Jan 13, 2026

View reviewed changes

csrc/multidevice/communication.cpp Show resolved Hide resolved

csrc/host_ir/lower_to_communication.cpp Show resolved Hide resolved

csrc/host_ir/lower_to_communication.cpp Show resolved Hide resolved

tests/python/multidevice/test_multidevice.py Outdated Show resolved Hide resolved

remove intermediate outputs

62109c2

greptile-apps bot reviewed Jan 13, 2026

View reviewed changes

Priya2698 requested a review from wujingyue January 13, 2026 02:36

wujingyue reviewed Jan 14, 2026

View reviewed changes

remove copy, add exper-parallel patterns, check

90a2174

Priya2698 requested a review from wujingyue January 16, 2026 00:38

greptile-apps bot reviewed Jan 16, 2026

View reviewed changes

wujingyue reviewed Jan 16, 2026

View reviewed changes

set p_sharded and c_sharded_id

347061b

greptile-apps bot reviewed Jan 16, 2026

View reviewed changes

wujingyue approved these changes Jan 16, 2026

View reviewed changes

review

3166c61

greptile-apps bot reviewed Jan 16, 2026

View reviewed changes

add alltoall in c10d_mock

31da19a

define AllToAll options for c10d_mock

27fa2d0

wujingyue merged commit 36c6cfa into main Jan 16, 2026
18 checks passed

wujingyue deleted the pm/all2all branch January 16, 2026 16:16

		default:
		NVF_THROW("unrecognized CommunicationType: ", type);

AllToAll implementation #5705

AllToAll implementation #5705

Conversation

Priya2698 commented Dec 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Jan 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Changes walkthrough

PR Reviewer Guide

Uh oh!

Priya2698 commented Jan 10, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot commented Jan 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 4/5

Important Files Changed

Sequence Diagram

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Jan 13, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Jan 13, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

greptile-apps bot Jan 13, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

wujingyue left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Priya2698 commented Jan 16, 2026

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Priya2698 commented Jan 16, 2026

Priya2698 commented Dec 18, 2025 •

edited

Loading

github-actions bot commented Jan 7, 2026 •

edited

Loading

greptile-apps bot commented Jan 13, 2026 •

edited

Loading