[https://nvbugs/5636916][fix] Fix accuracy issue of TWOSHOT AllReduce kernel#10654

hyukn · 2026-01-14T04:59:38Z

Summary by CodeRabbit

Refactor
- Simplified barrier synchronization implementation in communication kernels by consolidating architecture-specific code paths into a unified approach, reducing code complexity while maintaining existing behavior.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

Description

Test Coverage

PR Checklist

Please review the following before submitting your PR:

PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
Update tava architecture diagram if there is a significant design change in PR.
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...

Provide a user friendly way for developers to interact with a Jenkins server.

Run /bot [-h|--help] to print this help message.

See details below for each supported subcommand.

Details

run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]

Launch build/test pipelines. All previously running jobs will be killed.

--reuse-test (optional)pipeline-id (OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.

--disable-reuse-test (OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.

--disable-fail-fast (OPTIONAL) : Disable fail fast on build/tests/infra failures.

--skip-test (OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.

--stage-list "A10-PyTorch-1, xxx" (OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT update GitHub check status.

--gpu-type "A30, H100_PCIe" (OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.

--test-backend "pytorch, cpp" (OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline status.

--only-multi-gpu-test (OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.

--disable-multi-gpu-test (OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.

--add-multi-gpu-test (OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.

--post-merge (OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.

--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" (OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".

--detailed-log (OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.

--debug (OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in the stage-list parameter to access the appropriate container environment. Note: Does NOT update GitHub check status.

For guidance on mapping tests to stage names, see docs/source/reference/ci-overview.md
and the scripts/test_to_stage_mapping.py helper.

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip testing for latest commit on pull request. --comment "Reason for skipping build/test" is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

reuse-pipeline

reuse-pipeline

Reuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

hyukn · 2026-01-14T05:00:07Z

/bot run --disable-fail-fast

coderabbitai · 2026-01-14T05:01:32Z

📝 Walkthrough

Walkthrough

The change simplifies barrier synchronization logic in a GPU kernel by removing architecture-specific inline assembly code paths and replacing them with a uniform store operation (st_flag) applicable across all CUDA architectures. A release fence operation on architectures >= 900 was also removed. Control flow remains unchanged.

Changes

Cohort / File(s)	Summary
GPU Kernel Simplification `cpp/tensorrt_llm/kernels/communicationKernels/allReduceFusionKernels.cu`	Removed architecture-specific conditional paths for barrier flag updates; replaced with uniform st_flag store for all architectures. Eliminated optional release fence synchronization previously used on CUDA >= 900. Net: -9 lines.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The PR description is largely empty with only the template structure filled out and no actual content in the Description, Test Coverage, or PR Checklist sections.	Provide a detailed description explaining the accuracy issue being fixed, the root cause, the solution implemented, relevant test cases that verify the fix, and confirm all PR checklist items have been reviewed.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly identifies the issue (accuracy issue of TWOSHOT AllReduce kernel), references the bug tracker (NVBugs ID), and specifies the type of change (fix), directly relating to the code modification.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

📝 Generate docstrings

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

tensorrt-cicd · 2026-01-14T05:06:02Z

PR_Github #31890 [ run ] triggered by Bot. Commit: 70f813e

tensorrt-cicd · 2026-01-14T08:36:12Z

PR_Github #31890 [ run ] completed with state SUCCESS. Commit: 70f813e
/LLM/main/L0_MergeRequest_PR pipeline #24694 completed with status: 'FAILURE'

⚠️ Action Required:

Please check the failed tests and fix your PR
If you cannot view the failures, ask the CI triggerer to share details
Once fixed, request an NVIDIA team member to trigger CI again

hyukn · 2026-01-14T08:43:21Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-01-14T08:48:47Z

PR_Github #31936 [ run ] triggered by Bot. Commit: 70f813e

tensorrt-cicd · 2026-01-14T11:57:44Z

PR_Github #31936 [ run ] completed with state SUCCESS. Commit: 70f813e
/LLM/main/L0_MergeRequest_PR pipeline #24735 completed with status: 'FAILURE'

⚠️ Action Required:

Please check the failed tests and fix your PR
If you cannot view the failures, ask the CI triggerer to share details
Once fixed, request an NVIDIA team member to trigger CI again

hyukn · 2026-01-15T02:17:09Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-01-15T02:26:18Z

PR_Github #32046 [ run ] triggered by Bot. Commit: 70f813e

tensorrt-cicd · 2026-01-15T05:50:25Z

PR_Github #32046 [ run ] completed with state SUCCESS. Commit: 70f813e
/LLM/main/L0_MergeRequest_PR pipeline #24837 completed with status: 'FAILURE'

⚠️ Action Required:

Please check the failed tests and fix your PR
If you cannot view the failures, ask the CI triggerer to share details
Once fixed, request an NVIDIA team member to trigger CI again

hyukn · 2026-01-15T06:41:21Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-01-15T06:46:41Z

PR_Github #32084 [ run ] triggered by Bot. Commit: 70f813e

tensorrt-cicd · 2026-01-15T08:31:11Z

PR_Github #32084 [ run ] completed with state SUCCESS. Commit: 70f813e
/LLM/main/L0_MergeRequest_PR pipeline #24865 completed with status: 'FAILURE'

⚠️ Action Required:

Please check the failed tests and fix your PR
If you cannot view the failures, ask the CI triggerer to share details
Once fixed, request an NVIDIA team member to trigger CI again

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

hyukn · 2026-01-16T13:34:03Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-01-16T13:40:51Z

PR_Github #32304 [ run ] triggered by Bot. Commit: 192a3cf

tensorrt-cicd · 2026-01-16T19:48:37Z

PR_Github #32304 [ run ] completed with state SUCCESS. Commit: 192a3cf
/LLM/release-1.2/L0_MergeRequest_PR pipeline #25 completed with status: 'FAILURE'

⚠️ Action Required:

Please check the failed tests and fix your PR
If you cannot view the failures, ask the CI triggerer to share details
Once fixed, request an NVIDIA team member to trigger CI again

hyukn · 2026-01-17T05:49:25Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-01-17T05:56:07Z

PR_Github #32402 [ run ] triggered by Bot. Commit: 192a3cf

tensorrt-cicd · 2026-01-17T07:37:34Z

PR_Github #32402 [ run ] completed with state SUCCESS. Commit: 192a3cf
/LLM/release-1.2/L0_MergeRequest_PR pipeline #35 completed with status: 'FAILURE'

⚠️ Action Required:

Please check the failed tests and fix your PR
If you cannot view the failures, ask the CI triggerer to share details
Once fixed, request an NVIDIA team member to trigger CI again

hyukn · 2026-01-19T00:55:13Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-01-19T01:01:12Z

PR_Github #32485 [ run ] triggered by Bot. Commit: 192a3cf

tensorrt-cicd · 2026-01-19T02:40:29Z

PR_Github #32485 [ run ] completed with state SUCCESS. Commit: 192a3cf
/LLM/release-1.2/L0_MergeRequest_PR pipeline #41 completed with status: 'FAILURE'

⚠️ Action Required:

Please check the failed tests and fix your PR
If you cannot view the failures, ask the CI triggerer to share details
Once fixed, request an NVIDIA team member to trigger CI again

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

hyukn · 2026-01-19T07:05:36Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-01-19T07:11:38Z

PR_Github #32535 [ run ] triggered by Bot. Commit: ff528ae

tensorrt-cicd · 2026-01-19T07:12:07Z

PR_Github #32536 Bot args parsing error: usage: /bot [-h]
{run,kill,skip,submit,reviewers,reuse-pipeline,reuse-review} ...
/bot: error: unrecognized arguments: run

tensorrt-cicd · 2026-01-19T12:43:28Z

PR_Github #32535 [ run ] completed with state SUCCESS. Commit: ff528ae
/LLM/release-1.2/L0_MergeRequest_PR pipeline #50 completed with status: 'FAILURE'

⚠️ Action Required:

Please check the failed tests and fix your PR
If you cannot view the failures, ask the CI triggerer to share details
Once fixed, request an NVIDIA team member to trigger CI again

hyukn · 2026-01-20T00:43:21Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-01-20T00:50:01Z

PR_Github #32619 [ run ] triggered by Bot. Commit: ff528ae

tensorrt-cicd · 2026-01-20T04:21:31Z

PR_Github #32619 [ run ] completed with state SUCCESS. Commit: ff528ae
/LLM/release-1.2/L0_MergeRequest_PR pipeline #62 completed with status: 'FAILURE'

⚠️ Action Required:

Please check the failed tests and fix your PR
If you cannot view the failures, ask the CI triggerer to share details
Once fixed, request an NVIDIA team member to trigger CI again

hyukn · 2026-01-20T04:25:56Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-01-20T04:36:17Z

PR_Github #32657 [ run ] triggered by Bot. Commit: ff528ae

tensorrt-cicd · 2026-01-20T06:00:13Z

PR_Github #32657 [ run ] completed with state SUCCESS. Commit: ff528ae
/LLM/release-1.2/L0_MergeRequest_PR pipeline #71 completed with status: 'SUCCESS'

… kernel (NVIDIA#10654) Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

…of TWO-SHOT AllReduce kernel (#10841) Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

… kernel (NVIDIA#10654) Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

… kernel (NVIDIA#10654) Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com> Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>

hyukn requested review from mikeiovine and yilin-void January 14, 2026 04:59

yilin-void approved these changes Jan 14, 2026

View reviewed changes

mikeiovine approved these changes Jan 14, 2026

View reviewed changes

mikeiovine mentioned this pull request Jan 15, 2026

[Bug]: Eagle3 acc len decreases with bigger batch size #9208

Open

5 tasks

hyukn changed the base branch from main to release/1.2 January 16, 2026 06:48

hyukn requested a review from a team as a code owner January 16, 2026 06:48

hyukn added 2 commits January 16, 2026 08:08

[https://nvbugs/5636916][fix] Fix TWOSHOT AllReduce accuracy issue.

f9c24fa

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

Revise LUT after twoshot kernel is updated.

192a3cf

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

hyukn force-pushed the fix/5636916 branch from 70f813e to 192a3cf Compare January 16, 2026 13:23

Fixing the ray ut for allreduce

ff528ae

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

litaotju approved these changes Jan 20, 2026

View reviewed changes

hyukn merged commit ed6df35 into NVIDIA:release/1.2 Jan 20, 2026
5 of 6 checks passed

hyukn added a commit to hyukn/TensorRT-LLM that referenced this pull request Jan 20, 2026

[https://nvbugs/5636916][fix] Fix accuracy issue of TWOSHOT AllReduce…

dd95c44

… kernel (NVIDIA#10654) Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

pengbowang-nv mentioned this pull request Jan 21, 2026

[https://nvbugs/5779536][fix] Unwaive Llama 3.3 related multi GPU tests #10855

Merged

1 task

hyukn added a commit that referenced this pull request Jan 21, 2026

[https://nvbugs/5636916][fix] Cherry-pick #10654: Fix accuracy issue …

bf7303c

…of TWO-SHOT AllReduce kernel (#10841) Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

dominicshanshan pushed a commit to dominicshanshan/TensorRT-LLM that referenced this pull request Jan 21, 2026

[https://nvbugs/5636916][fix] Fix accuracy issue of TWOSHOT AllReduce…

7475a52

… kernel (NVIDIA#10654) Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

dominicshanshan mentioned this pull request Jan 22, 2026

[None][chroe] Mass integration of release/1.2 #10888

Merged

1 task

Comments

Conversation

hyukn commented Jan 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Description

Test Coverage

PR Checklist

GitHub Bot Help

kill

skip

reuse-pipeline

Uh oh!

hyukn commented Jan 14, 2026

Uh oh!

coderabbitai bot commented Jan 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Estimated code review effort

Uh oh!

tensorrt-cicd commented Jan 14, 2026

Uh oh!

tensorrt-cicd commented Jan 14, 2026

Uh oh!

hyukn commented Jan 14, 2026

Uh oh!

tensorrt-cicd commented Jan 14, 2026

Uh oh!

tensorrt-cicd commented Jan 14, 2026

Uh oh!

hyukn commented Jan 15, 2026

Uh oh!

tensorrt-cicd commented Jan 15, 2026

Uh oh!

tensorrt-cicd commented Jan 15, 2026

Uh oh!

hyukn commented Jan 15, 2026

Uh oh!

tensorrt-cicd commented Jan 15, 2026

Uh oh!

tensorrt-cicd commented Jan 15, 2026

Uh oh!

hyukn commented Jan 16, 2026

Uh oh!

tensorrt-cicd commented Jan 16, 2026

Uh oh!

tensorrt-cicd commented Jan 16, 2026

Uh oh!

hyukn commented Jan 17, 2026

Uh oh!

tensorrt-cicd commented Jan 17, 2026

Uh oh!

tensorrt-cicd commented Jan 17, 2026

Uh oh!

hyukn commented Jan 19, 2026

Uh oh!

tensorrt-cicd commented Jan 19, 2026

Uh oh!

tensorrt-cicd commented Jan 19, 2026

Uh oh!

hyukn commented Jan 19, 2026

Uh oh!

tensorrt-cicd commented Jan 19, 2026

Uh oh!

tensorrt-cicd commented Jan 19, 2026

Uh oh!

tensorrt-cicd commented Jan 19, 2026

Uh oh!

hyukn commented Jan 20, 2026

Uh oh!

tensorrt-cicd commented Jan 20, 2026

Uh oh!

tensorrt-cicd commented Jan 20, 2026

Uh oh!

hyukn commented Jan 20, 2026

Uh oh!

tensorrt-cicd commented Jan 20, 2026

Uh oh!

tensorrt-cicd commented Jan 20, 2026

hyukn commented Jan 14, 2026 •

edited

Loading

coderabbitai bot commented Jan 14, 2026 •

edited

Loading