[None][feat] Add support for fused gate_up_proj scales for FP8 blockwise #6496

achartier · 2025-07-30T23:17:02Z

Summary by CodeRabbit

Bug Fixes
- Improved handling of different weight loading modes for expert weight scales, ensuring compatibility with multiple configurations and better error handling for unsupported modes.
Tests
- Enhanced unit tests to cover new weight loading modes, verifying correct weight structure and loading behavior across configurations.

feat: Add support for fused gate_up_proj scales for FP8 blockwise

Description

Handle FUSED_GATE_UP_PROJ MOE weight loading mode in DeepSeekFP8BlockScalesFusedMoEMethod
This is used by Llama4 models quantized with modelopt fp8_pb_wo

Test Coverage

Existing DeepSeek tests

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.

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.

coderabbitai · 2025-07-30T23:17:09Z

📝 Walkthrough

Walkthrough

The method load_expert_all_weight_scale_fp8_block_scale within the DeepSeekFP8BlockScalesFusedMoEMethod class was refactored to support multiple weight loading modes. The implementation now conditionally loads and processes weight scale tensors based on the weight_loading_mode attribute, with explicit error handling for unknown modes and changes to how scales are combined and copied. Correspondingly, the test test_fused_moe_fp8_blockwise was updated to parameterize over the new weight loading modes, adjusting weight initialization and model construction accordingly.

Changes

Cohort / File(s)	Change Summary
Fused MoE Quantization Logic `tensorrt_llm/_torch/modules/fused_moe/quantization.py`	Refactored the `load_expert_all_weight_scale_fp8_block_scale` method to support multiple weight loading modes, update control flow, and add error handling. No public API changes.
Unit Tests for Fused MoE `tests/unittest/_torch/modules/test_fused_moe.py`	Extended `test_fused_moe_fp8_blockwise` to include a new `WeightLoadingMode` parameter, modifying weight initialization and model instantiation to test both `VANILLA` and `FUSED_GATE_UP_PROJ` modes. Updated function signature accordingly.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Note

⚡️ Unit Test Generation is now available in beta!

Learn more here, or try it out under "Finishing Touches" below.

📜 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 37e3969 and 7038a2e.

📒 Files selected for processing (2)

tensorrt_llm/_torch/modules/fused_moe/quantization.py (1 hunks)
tests/unittest/_torch/modules/test_fused_moe.py (6 hunks)

🚧 Files skipped from review as they are similar to previous changes (2)

tests/unittest/_torch/modules/test_fused_moe.py
tensorrt_llm/_torch/modules/fused_moe/quantization.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

✨ Finishing Touches

📝 Generate Docstrings

🧪 Generate unit tests

Create PR with unit tests
Post copyable unit tests in a comment

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

🪧 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.
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.

Support

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

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 generate unit tests to generate unit tests for 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 or @coderabbitai title 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.

achartier · 2025-07-30T23:21:03Z

@hlu1 Should we add an unofficial fp8 blockwise version of Scout under llm-models for coverage?

hlu1 · 2025-07-31T19:53:32Z

@hlu1 Should we add an unofficial fp8 blockwise version of Scout under llm-models for coverage?

Please add unit test to test/unittest/_torch/modules/test_fused_moe.py
Unit test coverage is more important for this kind of changes

achartier · 2025-07-31T21:24:02Z

/bot run

tensorrt-cicd · 2025-07-31T21:29:46Z

PR_Github #13724 [ run ] triggered by Bot

tensorrt-cicd · 2025-08-01T02:39:20Z

PR_Github #13724 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #10312 completed with status: 'FAILURE'

achartier · 2025-08-01T19:31:51Z

/bot run

tensorrt-cicd · 2025-08-01T19:37:14Z

PR_Github #13827 [ run ] triggered by Bot

tensorrt-cicd · 2025-08-01T20:29:18Z

PR_Github #13827 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #10397 completed with status: 'FAILURE'

achartier · 2025-08-01T21:00:20Z

/bot run

tensorrt-cicd · 2025-08-01T21:05:44Z

PR_Github #13836 [ run ] triggered by Bot

tensorrt-cicd · 2025-08-02T03:14:03Z

PR_Github #13836 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #10405 completed with status: 'FAILURE'

hlu1 · 2025-08-02T19:34:13Z

Adding @Tracin as a reviewer

achartier · 2025-08-04T02:55:33Z

/bot run

tensorrt-cicd · 2025-08-04T03:04:38Z

PR_Github #13905 [ run ] triggered by Bot

tensorrt-cicd · 2025-08-04T03:39:44Z

PR_Github #13905 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #10468 completed with status: 'FAILURE'

Signed-off-by: Aurelien Chartier <[email protected]>

achartier · 2025-08-04T16:43:17Z

/bot run

tensorrt-cicd · 2025-08-04T16:48:33Z

PR_Github #14021 [ run ] triggered by Bot

tensorrt-cicd · 2025-08-04T23:03:51Z

PR_Github #14021 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #10570 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.

fix: Fix poor generation with FP8 Gemma3 1B checkpoint (NVIDIA#6499) Signed-off-by: Balaram Buddharaju <[email protected]> [None][fix] Serialize the window_size in the kv event (NVIDIA#6526) Signed-off-by: richardhuo-nv <[email protected]> [None][feat] Multi-block mode for Hopper spec dec XQA kernel (NVIDIA#4416) Signed-off-by: Jhao-Ting Chen <[email protected]> [None][feat] Add support for fused gate_up_proj scales for FP8 blockwise (NVIDIA#6496) Signed-off-by: Aurelien Chartier <[email protected]>

…ise (NVIDIA#6496) Signed-off-by: Aurelien Chartier <[email protected]> Signed-off-by: Lanyu Liao <[email protected]>

…ise (NVIDIA#6496) Signed-off-by: Aurelien Chartier <[email protected]>

achartier requested review from hlu1 and symphonylyh July 30, 2025 23:17

achartier requested a review from a team as a code owner July 30, 2025 23:17

achartier force-pushed the moe-blockwise branch from 4ded660 to 96a9388 Compare July 31, 2025 21:23

achartier force-pushed the moe-blockwise branch from 96a9388 to 27a1a80 Compare August 1, 2025 19:31

hlu1 requested a review from Tracin August 2, 2025 19:33

achartier force-pushed the moe-blockwise branch from 27a1a80 to 37e3969 Compare August 4, 2025 02:55

achartier added 2 commits August 4, 2025 09:42

feat: Add support for fused gate_up_proj scales for FP8 blockwise

90fc712

Signed-off-by: Aurelien Chartier <[email protected]>

add unit test

7038a2e

Signed-off-by: Aurelien Chartier <[email protected]>

achartier force-pushed the moe-blockwise branch from 37e3969 to 7038a2e Compare August 4, 2025 16:42

achartier changed the title ~~feat: Add support for fused gate_up_proj scales for FP8 blockwise~~ [None][feat] Add support for fused gate_up_proj scales for FP8 blockwise Aug 4, 2025

symphonylyh approved these changes Aug 5, 2025

View reviewed changes

symphonylyh merged commit 6da95f2 into NVIDIA:main Aug 5, 2025
4 of 5 checks passed

coderabbitai bot mentioned this pull request Aug 6, 2025

[TRTLLM-6898][feat] make fused_moe_cute_dsl work on blackwell #6616

Merged

jain-ria pushed a commit to jain-ria/TensorRT-LLM that referenced this pull request Aug 7, 2025

[None][feat] Add support for fused gate_up_proj scales for FP8 blockw…

eaf255f

…ise (NVIDIA#6496) Signed-off-by: Aurelien Chartier <[email protected]>

This was referenced Aug 13, 2025

Support W4A8 method of AngleSlim tool #6857

Open

[None][fix] Fix assertion errors of quantization when using online EPLB #6922

Merged

achartier deleted the moe-blockwise branch August 15, 2025 19:16

coderabbitai bot mentioned this pull request Aug 25, 2025

[None][test] Add MNNVL to unittest and speed up weight creation #7071

Closed

[None][feat] Add support for fused gate_up_proj scales for FP8 blockwise #6496

[None][feat] Add support for fused gate_up_proj scales for FP8 blockwise #6496

Uh oh!

Conversation

achartier commented Jul 30, 2025 • edited by coderabbitai bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Description

Test Coverage

GitHub Bot Help

kill

skip

reuse-pipeline

Uh oh!

coderabbitai bot commented Jul 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Estimated code review effort

Chat

Support

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

Documentation and Community

Uh oh!

achartier commented Jul 30, 2025

Uh oh!

hlu1 commented Jul 31, 2025

Uh oh!

achartier commented Jul 31, 2025

Uh oh!

tensorrt-cicd commented Jul 31, 2025

Uh oh!

tensorrt-cicd commented Aug 1, 2025

Uh oh!

achartier commented Aug 1, 2025

Uh oh!

tensorrt-cicd commented Aug 1, 2025

Uh oh!

tensorrt-cicd commented Aug 1, 2025

Uh oh!

achartier commented Aug 1, 2025

Uh oh!

tensorrt-cicd commented Aug 1, 2025

Uh oh!

tensorrt-cicd commented Aug 2, 2025

Uh oh!

hlu1 commented Aug 2, 2025

Uh oh!

achartier commented Aug 4, 2025

Uh oh!

tensorrt-cicd commented Aug 4, 2025

Uh oh!

tensorrt-cicd commented Aug 4, 2025

Uh oh!

achartier commented Aug 4, 2025

Uh oh!

tensorrt-cicd commented Aug 4, 2025

Uh oh!

tensorrt-cicd commented Aug 4, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

achartier commented Jul 30, 2025 •

edited by coderabbitai bot

Loading

coderabbitai bot commented Jul 30, 2025 •

edited

Loading