Change the all-reduce strategy to NCCL #6226

nzmora-nvidia · 2025-07-21T17:09:39Z

When the strategy is set to AUTO and world_size>1 we experience hangs and CUDA
memory errors.

This is the same issue as https://nvbugspro.nvidia.com/bug/5331013
Without this change test test_ad_build_small_multi.py fails (tp==2)
This is a temporary change until we understand why this hang is happening.
On dllcuster this issue does not manifest.

Re-enable test_ad_build_small_multi.py

Summary by CodeRabbit

Bug Fixes
- Improved distributed operation reliability by updating the strategy used for multi-GPU communication.
Tests
- Enabled test coverage for both single and multi-GPU configurations by removing previous test skips related to multi-GPU runs.

When the strategy is set to AUTO and world_size>1 we experience hangs and CUDA memory errors. * This is the same issue as https://nvbugspro.nvidia.com/bug/5331013 * Without this change test test_ad_build_small_multi.py fails (tp==2) * This is a temporary change until we understand why this hang is happening. * On dllcuster this issue does not manifest. Signed-off-by: Neta Zmora <[email protected]>

tests/unittest/_torch/auto_deploy/unit/multigpu/test_ad_build_small_multi.py Signed-off-by: Neta Zmora <[email protected]>

coderabbitai · 2025-07-21T17:09:46Z

Walkthrough

The changes update the all-reduce strategy in a distributed deployment module to use NCCL instead of AUTO as a workaround for a known NVIDIA bug, with a comment noting this is temporary. Additionally, a test that was previously skipped for multi-GPU scenarios is now enabled to run for all configurations.

Changes

File(s)	Change Summary
tensorrt_llm/_torch/auto_deploy/distributed/trtllm.py	Changed all-reduce strategy from AUTO to NCCL with a comment referencing an NVIDIA bug workaround.
tests/unittest/_torch/auto_deploy/unit/multigpu/test_ad_build_small_multi.py	Removed conditional skip, enabling the test to run for all world sizes, including multi-GPU scenarios.

Estimated code review effort

2 (~12 minutes)

Poem

🐇
A bug in the code, a workaround found,
NCCL now leads where AUTO was bound.
Tests once asleep now eagerly run,
Multi-GPU dreams—debugging is fun!
With every fix, we hop ahead,
Toward a world where all bugs are shed.

📜 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 d7f0b0a and ba1b6ef.

📒 Files selected for processing (2)

tensorrt_llm/_torch/auto_deploy/distributed/trtllm.py (1 hunks)
tests/unittest/_torch/auto_deploy/unit/multigpu/test_ad_build_small_multi.py (0 hunks)

💤 Files with no reviewable changes (1)

tests/unittest/_torch/auto_deploy/unit/multigpu/test_ad_build_small_multi.py

🧰 Additional context used

🧠 Learnings (2)

📓 Common learnings

Learnt from: amitz-nv
PR: NVIDIA/TensorRT-LLM#5616
File: tensorrt_llm/executor/worker.py:375-384
Timestamp: 2025-07-17T09:01:27.374Z
Learning: In tensorrt_llm/executor/worker.py, the LoRA adapter cache optimization logic that checks `is_adapter_in_cpu_cache()` and conditionally passes None for weights/config has a known race condition issue that cannot be solved with simple error handling or verification checks. This is a known limitation that requires a more comprehensive solution.

tensorrt_llm/_torch/auto_deploy/distributed/trtllm.py (1)

Learnt from: amitz-nv
PR: #5616
File: tensorrt_llm/executor/worker.py:375-384
Timestamp: 2025-07-17T09:01:27.374Z
Learning: In tensorrt_llm/executor/worker.py, the LoRA adapter cache optimization logic that checks is_adapter_in_cpu_cache() and conditionally passes None for weights/config has a known race condition issue that cannot be solved with simple error handling or verification checks. This is a known limitation that requires a more comprehensive solution.

🔇 Additional comments (1)

tensorrt_llm/_torch/auto_deploy/distributed/trtllm.py (1)

20-21: Well-documented temporary workaround for known NVIDIA bug.

The change from AllReduceStrategy.AUTO to AllReduceStrategy.NCCL with the accompanying comment clearly documents this as a temporary fix for NVIDIA bug 5331013. This addresses the hang and CUDA memory errors when world size > 1, which aligns with the PR objectives.

✨ 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.

nzmora-nvidia · 2025-07-21T17:10:19Z

/bot run

tensorrt-cicd · 2025-07-21T17:15:46Z

PR_Github #12459 [ run ] triggered by Bot

tensorrt-cicd · 2025-07-21T18:24:58Z

PR_Github #12459 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9267 completed with status: 'FAILURE'

nzmora-nvidia added 2 commits July 21, 2025 10:02

Re-enable test_ad_build_small_multi.py

ba1b6ef

tests/unittest/_torch/auto_deploy/unit/multigpu/test_ad_build_small_multi.py Signed-off-by: Neta Zmora <[email protected]>

nzmora-nvidia requested a review from a team as a code owner July 21, 2025 17:09

nzmora-nvidia requested a review from lucaslie July 21, 2025 17:09

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Change the all-reduce strategy to NCCL #6226

Change the all-reduce strategy to NCCL #6226

Uh oh!

nzmora-nvidia commented Jul 21, 2025 •

edited by coderabbitai bot

Loading

Uh oh!

coderabbitai bot commented Jul 21, 2025 •

edited

Loading

Chat

Support

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

Documentation and Community

Uh oh!

nzmora-nvidia commented Jul 21, 2025

Uh oh!

tensorrt-cicd commented Jul 21, 2025

Uh oh!

tensorrt-cicd commented Jul 21, 2025

Uh oh!

Uh oh!

Change the all-reduce strategy to NCCL #6226

Are you sure you want to change the base?

Change the all-reduce strategy to NCCL #6226

Uh oh!

Conversation

nzmora-nvidia 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!

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

Poem

Chat

Support

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

Documentation and Community

Uh oh!

nzmora-nvidia commented Jul 21, 2025

Uh oh!

tensorrt-cicd commented Jul 21, 2025

Uh oh!

tensorrt-cicd commented Jul 21, 2025

Uh oh!

Uh oh!

nzmora-nvidia commented Jul 21, 2025 •

edited by coderabbitai bot

Loading

coderabbitai bot commented Jul 21, 2025 •

edited

Loading