[https://nvbugs/6652876][fix] Add correct autotuner constraint for sm120 fp8 block scale moe

[DomBrown]

Dev Engineer Review

• Added the SM120-specific FP8 block-scale MoE autotuner constraint.
• The constraint pads activation dimension M to a multiple of 4.
• SM120 applies the constraint to scale tensor dimension 0.
• SM100 and other architectures retain their existing paths.
• The change is scoped and does not alter public APIs.
• Updated tests/integration/test_lists/waives.txt to remove obsolete waivers and add updated test configurations and NVBugs references.
• The implementation and test-list updates are consistent with the PR objective.

QA Engineer Review

• Modified test-list file: tests/integration/test_lists/waives.txt.
• No test-db or qa files were modified.
• The waiver changes cover multiple hardware and test categories, including MoE and RTXPro6000D FP8.
• CBTS coverage data is unavailable in the provided context.
• Verdict: needs follow-up.

Description

The FP8 blockscale MoE requires a different autotuner constraint than the one used for SM100. This PR adds the missing constraint and unwaives the failing test.

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)

• If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

• 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

To see a list of available CI bot commands, please comment /bot help.

[DomBrown]

/bot run --stage-list "RTXPro6000D-PyTorch-1"

[tensorrt-cicd]

PR_Github #68808 [ run ] triggered by Bot. Commit: 7115c2a Link to invocation

[tensorrt-cicd]

PR_Github #68808 [ run ] completed with state SUCCESS. Commit: 7115c2a
/LLM/main/L0_MergeRequest_PR pipeline #56202 (Partly Tested) completed with status: 'SUCCESS'

CI Report

Link to invocation

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 3032300d-eac6-4d4a-b0e5-9d376a4bcf23

📥 Commits

Reviewing files that changed from the base of the PR and between 7115c2a and 94c682d.

📒 Files selected for processing (1)

• tests/integration/test_lists/waives.txt

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

---

Walkthrough

The change adds an SM120-specific FP8 block-scaling GEMM constraint and updates architecture-based selection. It also revises integration test waivers across multiple hardware targets and test categories.

Changes

FP8 GEMM and integration test updates

Layer / File(s)	Summary
SM120 constraint selection tensorrt_llm/_torch/custom_ops/torch_custom_ops.py	SM120 pads the activation M dimension to a multiple of 4 and applies the constraint to scale tensor dimension 0. Other SM versions retain their existing paths.
Integration waiver updates tests/integration/test_lists/waives.txt	The waiver list adds, removes, and revises accuracy, serving, performance, visual-generation, MoE, attention, sampler, and multimodal test configurations across hardware targets.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Merge Risk: ⚪ Minimal · up to 94c68

The PR adds the SM120 FP8 block-scale MoE autotuner constraint and updates the related test waiver; no actionable merge-blocking risk remains beyond normal checks and review.

Possibly related PRs

• NVIDIA/TensorRT-LLM#15585: Updates test skip waivers in the same waiver file.

Suggested reviewers: bowenfu, junyixu-nv

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly identifies the NVBugs issue, fix type, SM120 FP8 block-scale MoE scope, and autotuner constraint change.
Description check	✅ Passed	The description explains the issue and solution and includes the repository checklist. It omits the separate Test Coverage section, but it identifies the affected failing test through the unwaiver cha…
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 0…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

Full details: Description check

Explanation

The description explains the issue and solution and includes the repository checklist. It omits the separate Test Coverage section, but it identifies the affected failing test through the unwaiver change, so the description is mostly complete.

Full details: Docstring Coverage

Explanation

No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 0 files.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[DomBrown]

/bot run

[tensorrt-cicd]

PR_Github #68831 [ run ] triggered by Bot. Commit: 7115c2a Link to invocation

[tensorrt-cicd]

PR_Github #68831 [ run ] completed with state SUCCESS. Commit: 7115c2a
/LLM/main/L0_MergeRequest_PR pipeline #56223 completed with status: 'FAILURE'

CI Report

⚠️ 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

CI Agent Failure Analysis

Link to invocation

[DomBrown]

/bot run --disable-fail-fast

[allisonlim-nv]

LGTM

[tensorrt-cicd]

PR_Github #68856 [ run ] triggered by Bot. Commit: 7115c2a Link to invocation

[tensorrt-cicd]

PR_Github #68856 [ run ] completed with state SUCCESS. Commit: 7115c2a
/LLM/main/L0_MergeRequest_PR pipeline #56242 completed with status: 'UNSTABLE'

CI Report

⚠️ Multi-GPU Label Required:
Multi-GPU tests require the ci: full pre-merge approved label on this PR. Ask a member of NVIDIA/trt-llm-ci-approvers to add the label, then re-trigger CI with the same bot command (no rebase needed).

⚠️ 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

Link to invocation

[DomBrown]

/bot run

[tensorrt-cicd]

PR_Github #68891 [ run ] triggered by Bot. Commit: 7115c2a Link to invocation

[tensorrt-cicd]

PR_Github #68891 [ run ] completed with state FAILURE. Commit: 7115c2a
/LLM/main/L0_MergeRequest_PR pipeline #56280 completed with status: 'FAILURE'

CI Report

⚠️ 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

CI Agent Failure Analysis

Link to invocation

[DomBrown]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #69082 [ run ] triggered by Bot. Commit: 94c682d Link to invocation

[tensorrt-cicd]

PR_Github #69082 [ run ] completed with state FAILURE. Commit: 94c682d
/LLM/main/L0_MergeRequest_PR pipeline #56454 completed with status: 'FAILURE'

CI Report

⚠️ 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

CI Agent Failure Analysis

Link to invocation

[DomBrown]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #69137 [ run ] triggered by Bot. Commit: 94c682d Link to invocation

[DomBrown]

/bot run

[tensorrt-cicd]

PR_Github #69425 [ run ] triggered by Bot. Commit: 94c682d Link to invocation

[tensorrt-cicd]

PR_Github #69137 [ run ] completed with state ABORTED. Commit: 94c682d

Link to invocation

[tensorrt-cicd]

PR_Github #69425 [ run ] completed with state SUCCESS. Commit: 94c682d
/LLM/main/L0_MergeRequest_PR pipeline #56757 completed with status: 'FAILURE'

CI Report

⚠️ 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

CI Agent Failure Analysis

Link to invocation

[DomBrown]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #69453 [ run ] triggered by Bot. Commit: 94c682d Link to invocation

[tensorrt-cicd]

PR_Github #69453 [ run ] completed with state SUCCESS. Commit: 94c682d
/LLM/main/L0_MergeRequest_PR pipeline #56785 completed with status: 'FAILURE'

CI Report

⚠️ 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

CI Agent Failure Analysis

Link to invocation

[DomBrown]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #69492 [ run ] triggered by Bot. Commit: 94c682d Link to invocation

[tensorrt-cicd]

PR_Github #69492 [ run ] completed with state SUCCESS. Commit: 94c682d
/LLM/main/L0_MergeRequest_PR pipeline #56819 completed with status: 'SUCCESS'

CI Report

Link to invocation