[Fix][nvbug 5401163][nvbug 5404726][Qwen3] Fix bug of MoE on tp > 1 with trtllm moe backend #6235

byshiue · 2025-07-21T22:04:13Z

Unwaive following tests

accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm]
accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_trtllm]
accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm]

Summary by CodeRabbit

New Features
- Added a new attribute to improve module preloading behavior for certain model configurations.
Documentation
- Updated comments to clarify supported models for a specific parameter.
Tests
- Re-enabled several previously skipped tests for Qwen3 models, ensuring broader test coverage.
- Added a new test for the Qwen3-30B-A3B model using the Eagle decoding algorithm.
- Included a new accuracy reference entry for the Eagle decoding algorithm on Qwen3-30B-A3B.

byshiue · 2025-07-21T22:04:20Z

/bot run

coderabbitai · 2025-07-21T22:04:20Z

Walkthrough

The changes introduce a new attribute, preload_weight_modules, to the Qwen3MoEModel and propagate it to Qwen3MoeForCausalLM. A comment is updated to include "Qwen3" alongside "llama4" for a workaround in _load_weights_impl_v2. Three test skip entries related to Qwen3 nvfp4 tests are removed. Additionally, a new Eagle decoding algorithm entry and a corresponding test are added for Qwen3-30B-A3B, and a conditional skip is removed from a Qwen3-235B-A22B test.

Changes

File(s)	Change Summary
tensorrt_llm/_torch/models/modeling_qwen3_moe.py	Added `preload_weight_modules` attribute to `Qwen3MoEModel` and propagated it to `Qwen3MoeForCausalLM`.
tensorrt_llm/_torch/models/modeling_utils.py	Updated comment in `_load_weights_impl_v2` to include "Qwen3" as supported for a workaround.
tests/integration/test_lists/waives.txt	Removed three skip entries for Qwen3 nvfp4-related tests.
tests/integration/defs/accuracy/gsm8k.yaml	Added Eagle decoding algorithm entry for Qwen3-30B-A3B with accuracy 83.43.
tests/integration/defs/accuracy/test_llm_api_pytorch.py	Added `test_eagle3` method for Qwen3-30B-A3B; removed conditional skip in `test_nvfp4` for Qwen3-235B-A22B.

Estimated code review effort

2 (~15 minutes)

Possibly related PRs

[Infra] - Waive failed tests in post-merge #6176: Removed skip entries for Qwen3 nvfp4 tests in waivers file, related contextually to this PR's waiver changes.

Suggested labels

Community want to contribute

Suggested reviewers

nv-guomingz
litaotju

Poem

A tweak for Qwen3, so neat and precise,
With modules to preload, and comments concise.
Skipped tests now run, the waivers are gone—
The code hops ahead, like a rabbit at dawn!
🐇✨

📜 Recent review details

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 60714eb and 1966e99.

📒 Files selected for processing (5)

tensorrt_llm/_torch/models/modeling_qwen3_moe.py (2 hunks)
tensorrt_llm/_torch/models/modeling_utils.py (1 hunks)
tests/integration/defs/accuracy/references/gsm8k.yaml (1 hunks)
tests/integration/defs/accuracy/test_llm_api_pytorch.py (1 hunks)
tests/integration/test_lists/waives.txt (0 hunks)

💤 Files with no reviewable changes (1)

tests/integration/test_lists/waives.txt

✅ Files skipped from review due to trivial changes (2)

tests/integration/defs/accuracy/references/gsm8k.yaml
tensorrt_llm/_torch/models/modeling_utils.py

🚧 Files skipped from review as they are similar to previous changes (1)

tensorrt_llm/_torch/models/modeling_qwen3_moe.py

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)

GitHub Check: Pre-commit Check

🔇 Additional comments (1)

tests/integration/defs/accuracy/test_llm_api_pytorch.py (1)

1759-1782: LGTM! Well-structured Eagle3 decoding test.

The test method is properly configured with appropriate settings for Eagle3 speculative decoding:

Correctly disables overlap scheduler and block reuse for speculative decoding

Uses multiple batch sizes in CUDA graph configuration

The eagle3_one_model=True parameter and draft length of 1 are appropriate for Eagle3

Follows established patterns from other Eagle tests in the codebase

✨ Finishing Touches

📝 Generate Docstrings

🪧 Tips

Chat

There are 3 ways to chat with CodeRabbit:

Review comments: Directly reply to a review comment made by CodeRabbit. Example:
- I pushed a fix in commit <commit_id>, please review it.
- Explain this complex logic.
- Open a follow-up GitHub issue for this discussion.
Files and specific lines of code (under the "Files changed" tab): Tag @coderabbitai in a new review comment at the desired location with your query. Examples:
- @coderabbitai explain this code block.
- @coderabbitai modularize this function.
PR comments: Tag @coderabbitai in a new PR comment to ask questions about the PR branch. For the best results, please provide a very specific query, as very limited context is provided in this mode. Examples:
- @coderabbitai gather interesting stats about this repository and render them as a table. Additionally, render a pie chart showing the language distribution in the codebase.
- @coderabbitai read src/utils.ts and explain its main purpose.
- @coderabbitai read the files in the src/scheduler package and generate a class diagram using mermaid and a README in the markdown format.
- @coderabbitai help me debug CodeRabbit configuration file.

Support

Need help? Create a ticket on our support page for assistance with any issues or questions.

Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments.

CodeRabbit Commands (Invoked using PR comments)

@coderabbitai pause to pause the reviews on a PR.
@coderabbitai resume to resume the paused reviews.
@coderabbitai review to trigger an incremental review. This is useful when automatic reviews are disabled for the repository.
@coderabbitai full review to do a full review from scratch and review all the files again.
@coderabbitai summary to regenerate the summary of the PR.
@coderabbitai generate docstrings to generate docstrings for this PR.
@coderabbitai generate sequence diagram to generate a sequence diagram of the changes in this PR.
@coderabbitai resolve resolve all the CodeRabbit review comments.
@coderabbitai configuration to show the current CodeRabbit configuration for the repository.
@coderabbitai help to get help.

Other keywords and placeholders

Add @coderabbitai ignore anywhere in the PR description to prevent this PR from being reviewed.
Add @coderabbitai summary to generate the high-level summary at a specific location in the PR description.
Add @coderabbitai anywhere in the PR title to generate the title automatically.

Documentation and Community

Visit our Documentation for detailed information on how to use CodeRabbit.
Join our Discord Community to get help, request features, and share feedback.
Follow us on X/Twitter for updates and announcements.

tensorrt-cicd · 2025-07-21T22:27:23Z

PR_Github #12480 [ run ] triggered by Bot

tensorrt-cicd · 2025-07-22T01:02:32Z

PR_Github #12480 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9285 completed with status: 'FAILURE'

byshiue · 2025-07-22T16:23:13Z

/bot run

tensorrt-cicd · 2025-07-22T16:29:19Z

PR_Github #12592 [ run ] triggered by Bot

…g on tp > 1 with trtllm moe backend Signed-off-by: bhsueh <[email protected]>

Signed-off-by: bhsueh <[email protected]>

byshiue · 2025-07-22T20:20:42Z

/bot run --disable-fail-fast

tensorrt-cicd · 2025-07-22T20:25:51Z

PR_Github #12604 [ run ] triggered by Bot

tensorrt-cicd · 2025-07-22T20:25:53Z

PR_Github #12592 [ run ] completed with state ABORTED

tensorrt-cicd · 2025-07-23T00:30:35Z

PR_Github #12604 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9379 completed with status: 'FAILURE'

byshiue · 2025-07-23T15:52:28Z

/bot run

tensorrt-cicd · 2025-07-23T15:57:57Z

PR_Github #12722 [ run ] triggered by Bot

tensorrt-cicd · 2025-07-23T18:19:09Z

PR_Github #12722 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9469 completed with status: 'FAILURE'

tensorrt_llm/_torch/models/modeling_qwen3_moe.py

byshiue · 2025-07-23T19:50:06Z

/bot run

tensorrt-cicd · 2025-07-23T19:55:00Z

PR_Github #12744 [ run ] triggered by Bot

tensorrt-cicd · 2025-07-23T23:11:48Z

PR_Github #12744 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9488 completed with status: 'SUCCESS'

shaharmor98

LGTM

tensorrt_llm/_torch/models/modeling_qwen3_moe.py

…ith trtllm moe backend (NVIDIA#6235) Signed-off-by: bhsueh <[email protected]> Signed-off-by: Shreyas Misra <[email protected]>

…ith trtllm moe backend (NVIDIA#6235) Signed-off-by: bhsueh <[email protected]> Signed-off-by: Ransiki Zhang <[email protected]>

…ith trtllm moe backend (NVIDIA#6235) Signed-off-by: bhsueh <[email protected]> Signed-off-by: Lanyu Liao <[email protected]>

byshiue requested a review from a team as a code owner July 21, 2025 22:04

byshiue requested review from brb-nv and nv-yilinf July 21, 2025 22:04

byshiue added 3 commits July 22, 2025 20:06

add some modules into serailized loading list for Qwen3 to prevent bu…

3c496cf

…g on tp > 1 with trtllm moe backend Signed-off-by: bhsueh <[email protected]>

unwaive related CI tests

2ed4e24

Signed-off-by: bhsueh <[email protected]>

add qwen3 30b eagle3 test

1966e99

Signed-off-by: bhsueh <[email protected]>

byshiue force-pushed the nvbug_5401163 branch from 60714eb to 1966e99 Compare July 22, 2025 20:19

byshiue changed the title ~~[Fix][Nvbug 5401163] Fix bug of MoE on tp > 1 with trtllm moe backend~~ [Fix][nvbug 5401163][nvbug 5404726][Qwen3] Fix bug of MoE on tp > 1 with trtllm moe backend Jul 22, 2025

byshiue mentioned this pull request Jul 22, 2025

[Eagle3] Accuracy Issue #5791

Open

4 tasks

brb-nv reviewed Jul 23, 2025

View reviewed changes

tensorrt_llm/_torch/models/modeling_qwen3_moe.py Show resolved Hide resolved

shaharmor98 approved these changes Jul 24, 2025

View reviewed changes

tensorrt_llm/_torch/models/modeling_qwen3_moe.py Show resolved Hide resolved

byshiue merged commit 7b6aadc into NVIDIA:main Jul 24, 2025
3 checks passed

This was referenced Jul 29, 2025

[nvbug 5380101][fix] Fix nemotronNAS loading for TP>1 #6447

Merged

[None][feat] Add Qwen3 MoE support to TensorRT backend #6470

Merged

coderabbitai bot mentioned this pull request Aug 21, 2025

[https://nvbugs/5437405][fix] qwen3 235b eagle3 ci #7000

Merged

[Fix][nvbug 5401163][nvbug 5404726][Qwen3] Fix bug of MoE on tp > 1 with trtllm moe backend #6235

[Fix][nvbug 5401163][nvbug 5404726][Qwen3] Fix bug of MoE on tp > 1 with trtllm moe backend #6235

Uh oh!

Conversation

byshiue commented Jul 21, 2025 • edited by coderabbitai bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Uh oh!

byshiue commented Jul 21, 2025

Uh oh!

coderabbitai bot commented Jul 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Estimated code review effort

Possibly related PRs

Suggested labels

Suggested reviewers

Poem

Chat

Support

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

Documentation and Community

Uh oh!

tensorrt-cicd commented Jul 21, 2025

Uh oh!

tensorrt-cicd commented Jul 22, 2025

Uh oh!

byshiue commented Jul 22, 2025

Uh oh!

tensorrt-cicd commented Jul 22, 2025

Uh oh!

byshiue commented Jul 22, 2025

Uh oh!

tensorrt-cicd commented Jul 22, 2025

Uh oh!

tensorrt-cicd commented Jul 22, 2025

Uh oh!

tensorrt-cicd commented Jul 23, 2025

Uh oh!

byshiue commented Jul 23, 2025

Uh oh!

tensorrt-cicd commented Jul 23, 2025

Uh oh!

tensorrt-cicd commented Jul 23, 2025

Uh oh!

Uh oh!

byshiue commented Jul 23, 2025

Uh oh!

tensorrt-cicd commented Jul 23, 2025

Uh oh!

tensorrt-cicd commented Jul 23, 2025

Uh oh!

shaharmor98 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

byshiue commented Jul 21, 2025 •

edited by coderabbitai bot

Loading

coderabbitai bot commented Jul 21, 2025 •

edited

Loading