[#8982][feat] AutoDeploy attention dp support #10728

lucaslie · 2026-01-15T22:50:33Z

Summary by CodeRabbit

New Features
- Added attention data parallelism support for distributed inference across multiple GPUs.
Improvements
- Enhanced CUDA graph optimization with cross-rank consistency checks to ensure consistent behavior in distributed deployments.
Tests
- Added test coverage for attention data parallelism across multi-GPU configurations.

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

Description

fixes #8982

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.

tensorrt_llm/_torch/pyexecutor/py_executor.py

lucaslie · 2026-01-16T22:11:05Z

/bot run --add-multi-gpu-test --disable-fail-fast

tensorrt-cicd · 2026-01-16T22:16:41Z

PR_Github #32356 [ run ] triggered by Bot. Commit: e2c9817

coderabbitai · 2026-01-16T22:18:27Z

📝 Walkthrough

Walkthrough

Attention data parallelism (attention DP) support is added to AutoDeploy through distributed-aware CUDA graph padding logic. ADEngine construction is updated to accept and propagate distributed configuration, and sharding behavior is conditionally skipped when attention DP is enabled. Tests validate the feature across multi-GPU configurations.

Changes

Cohort / File(s)	Summary
Distributed CUDA Graph Padding Logic `tensorrt_llm/_torch/auto_deploy/shim/ad_executor.py`	Replaces single-rank CUDA graph checks with cross-rank consistency checks when attention DP is enabled. Adds distributed gathering via `tp_allgather` to determine `can_run_cuda_graph_all`, `max_batch_size`, and `can_pad_all` across ranks. Recomputes CUDA graph batch size using maximum per-rank value; pads with persistent dummy request only when allowed. Adds assertions for padding validity.
ADEngine Public API Updates `tensorrt_llm/_torch/auto_deploy/shim/ad_executor.py`	`ADEngine.build_from_config` and `ADEngine.__init__` now accept optional `dist: Optional[Distributed]` parameter; `create_autodeploy_executor` builds `dist_mapping` including `enable_attention_dp`, creates `dist`, and propagates to ADEngine. Adds logging for `dist_mapping` and `dist` information during initialization.
Sharding Configuration `tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py`	Adds `enable_attention_dp: bool` flag to `ShardingTransformConfig`. Updates `_apply` flow to skip sharding when `enable_attention_dp` is True; logs skip reason and returns early.
Test Coverage `tests/integration/defs/accuracy/test_llm_api_autodeploy.py`, `tests/integration/test_lists/test-db/l0_dgx_h100.yml`	Adds parametrized `test_attention_dp` method for world sizes [2, 4] with attention DP enabled. Registers 4-GPU test entry in CI configuration.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 36.36% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check	⚠️ Warning	The PR description is minimal and lacks required sections. It references GitHub issue `#8982` (though the objectives mention `#10063`) but provides no explanation of what was changed, why, or test coverage details.	Add a Description section explaining the changes and implementation details. Add a Test Coverage section listing the test methods (test_attention_dp) that validate this feature. Update the issue reference if needed.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	The PR implements attention DP support in AutoDeploy by adding distributed-aware CUDA graph padding logic, configuration flags, and test coverage, satisfying issue `#10063`'s core requirement.
Out of Scope Changes check	✅ Passed	All code changes are directly scoped to implementing attention DP support: CUDA graph padding logic, configuration flags, test cases, and test listings.
Title check	✅ Passed	The title clearly summarizes the main change: adding attention data parallel support to AutoDeploy, which is the primary objective of this PR.

_{✏️ 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 · 2026-01-17T05:26:56Z

PR_Github #32356 [ run ] completed with state SUCCESS. Commit: e2c9817
/LLM/main/L0_MergeRequest_PR pipeline #25072 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

lucaslie · 2026-01-20T16:28:09Z

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py

lucaslie · 2026-01-20T20:02:01Z

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

tensorrt-cicd · 2026-01-20T20:08:33Z

PR_Github #32804 [ run ] triggered by Bot. Commit: 8f9b201

tensorrt-cicd · 2026-01-20T21:13:52Z

PR_Github #32804 [ run ] completed with state SUCCESS. Commit: 8f9b201
/LLM/main/L0_MergeRequest_PR pipeline #25382 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

lucaslie · 2026-01-20T22:00:46Z

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

tensorrt-cicd · 2026-01-20T22:07:26Z

PR_Github #32813 [ run ] triggered by Bot. Commit: 8f9b201

tensorrt-cicd · 2026-01-20T23:16:18Z

PR_Github #32813 [ run ] completed with state SUCCESS. Commit: 8f9b201
/LLM/main/L0_MergeRequest_PR pipeline #25389 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

lucaslie · 2026-01-21T14:30:53Z

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

tensorrt-cicd · 2026-01-21T14:37:12Z

PR_Github #32964 [ run ] triggered by Bot. Commit: d9edf96

tensorrt-cicd · 2026-01-21T15:57:30Z

PR_Github #32964 [ run ] completed with state SUCCESS. Commit: d9edf96
/LLM/main/L0_MergeRequest_PR pipeline #25491 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: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

lucaslie · 2026-01-22T16:15:54Z

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

tensorrt-cicd · 2026-01-22T16:22:54Z

PR_Github #33201 [ run ] triggered by Bot. Commit: 97b4e05

tensorrt-cicd · 2026-01-22T18:10:57Z

PR_Github #33201 [ run ] completed with state SUCCESS. Commit: 97b4e05
/LLM/main/L0_MergeRequest_PR pipeline #25650 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

lucaslie · 2026-01-22T18:52:00Z

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

tensorrt-cicd · 2026-01-22T18:57:43Z

PR_Github #33213 [ run ] triggered by Bot. Commit: 97b4e05

tensorrt-cicd · 2026-01-22T19:55:28Z

PR_Github #33213 [ run ] completed with state SUCCESS. Commit: 97b4e05
/LLM/main/L0_MergeRequest_PR pipeline #25661 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

lucaslie · 2026-01-22T19:58:52Z

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

tensorrt-cicd · 2026-01-22T20:04:26Z

PR_Github #33219 [ run ] triggered by Bot. Commit: 97b4e05

tensorrt-cicd · 2026-01-22T20:53:03Z

PR_Github #33219 [ run ] completed with state SUCCESS. Commit: 97b4e05
/LLM/main/L0_MergeRequest_PR pipeline #25666 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

lucaslie · 2026-01-22T21:27:49Z

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

tensorrt-cicd · 2026-01-22T21:33:55Z

PR_Github #33228 [ run ] triggered by Bot. Commit: 97b4e05

tensorrt-cicd · 2026-01-22T22:21:10Z

PR_Github #33228 [ run ] completed with state SUCCESS. Commit: 97b4e05
/LLM/main/L0_MergeRequest_PR pipeline #25669 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

lucaslie · 2026-01-22T23:07:43Z

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

tensorrt-cicd · 2026-01-22T23:15:37Z

PR_Github #33230 [ run ] triggered by Bot. Commit: 97b4e05

tensorrt-cicd · 2026-01-23T00:05:37Z

PR_Github #33230 [ run ] completed with state SUCCESS. Commit: 97b4e05
/LLM/main/L0_MergeRequest_PR pipeline #25672 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

lucaslie commented Jan 15, 2026

View reviewed changes

tensorrt_llm/_torch/pyexecutor/py_executor.py Show resolved Hide resolved

lucaslie force-pushed the ll/attention_dp branch 2 times, most recently from 2be1144 to e2c9817 Compare January 16, 2026 22:09

lucaslie marked this pull request as ready for review January 16, 2026 22:10

lucaslie requested a review from a team as a code owner January 16, 2026 22:10

lucaslie requested a review from bmarimuthu-nv January 16, 2026 22:10

lucaslie requested a review from suyoggupta January 16, 2026 22:11

suyoggupta approved these changes Jan 16, 2026

View reviewed changes

lucaslie enabled auto-merge (squash) January 16, 2026 23:08

lucaslie changed the title ~~[#10063][feat] AutoDeploy attention dp support~~ [#8982][feat] AutoDeploy attention dp support Jan 20, 2026

lucaslie force-pushed the ll/attention_dp branch from e2c9817 to 1fd00b7 Compare January 20, 2026 16:27

lucaslie force-pushed the ll/attention_dp branch from 1fd00b7 to 5d18a0f Compare January 20, 2026 16:35

lucaslie commented Jan 20, 2026

View reviewed changes

tensorrt_llm/_torch/auto_deploy/transform/library/sharding.py Show resolved Hide resolved

lucaslie force-pushed the ll/attention_dp branch from 5d18a0f to 8f9b201 Compare January 20, 2026 20:01

lucaslie disabled auto-merge January 20, 2026 20:01

lucaslie enabled auto-merge (squash) January 20, 2026 22:02

lucaslie force-pushed the ll/attention_dp branch from 8f9b201 to d9edf96 Compare January 21, 2026 14:25

lucaslie added 2 commits January 22, 2026 08:06

[NVIDIA#10063][feat] AutoDeploy attention dp support

abe6030

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

fix cudagraph padding logic

97b4e05

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

lucaslie force-pushed the ll/attention_dp branch from d9edf96 to 97b4e05 Compare January 22, 2026 16:15

[#8982][feat] AutoDeploy attention dp support #10728

Are you sure you want to change the base?

[#8982][feat] AutoDeploy attention dp support #10728

Conversation

lucaslie commented Jan 15, 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!

Uh oh!

lucaslie commented Jan 16, 2026

Uh oh!

tensorrt-cicd commented Jan 16, 2026

Uh oh!

coderabbitai bot commented Jan 16, 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 17, 2026

Uh oh!

lucaslie commented Jan 20, 2026

Uh oh!

Uh oh!

lucaslie commented Jan 20, 2026

Uh oh!

tensorrt-cicd commented Jan 20, 2026

Uh oh!

tensorrt-cicd commented Jan 20, 2026

Uh oh!

lucaslie commented Jan 20, 2026

Uh oh!

tensorrt-cicd commented Jan 20, 2026

Uh oh!

tensorrt-cicd commented Jan 20, 2026

Uh oh!

lucaslie commented Jan 21, 2026

Uh oh!

tensorrt-cicd commented Jan 21, 2026

Uh oh!

tensorrt-cicd commented Jan 21, 2026

Uh oh!

lucaslie commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 22, 2026

Uh oh!

lucaslie commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 22, 2026

Uh oh!

lucaslie commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 22, 2026

Uh oh!

lucaslie commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 22, 2026

Uh oh!

lucaslie commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 22, 2026

Uh oh!

tensorrt-cicd commented Jan 23, 2026

Uh oh!

Reviewers

lucaslie commented Jan 15, 2026 •

edited

Loading

coderabbitai bot commented Jan 16, 2026 •

edited

Loading