[https://nvbugs/6095421][fix] Update resolve_moe_backend

[heyuhhh]

Summary by CodeRabbit

• New Features

  • Improved MOE backend selection to account for quantization configuration, enabling optimal backend resolution when using FP8_BLOCK_SCALES quantization with specific GPU architectures.

• Tests

  • Activated a previously waived accuracy test for enhanced validation coverage.

Description

Follows PR#14182 to complete the fix

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)

• 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.

[heyuhhh]

/bot run --disable-fail-fast

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR introduces quantization-aware MOE backend resolution by extending the resolve_moe_backend method with an optional quant_config parameter and refactoring from_pretrained into a two-phase process that loads quantization configuration before finalizing backend selection. A test waiver is removed as no longer necessary.

Changes

Quantization-aware MOE Backend Selection

Layer / File(s)	Summary
Enhanced resolve_moe_backend method with quantization awareness tensorrt_llm/_torch/model_config.py	Method signature accepts optional quant_config parameter. New quantization-aware logic detects FP8_BLOCK_SCALES algorithm on SM 100f GPUs and returns "DEEPGEMM" backend. Import formatting updated for multi-line block.
Two-phase MOE backend resolution in from_pretrained tensorrt_llm/_torch/model_config.py	from_pretrained refactored to compute MOE backend hint first using architecture only, then load quantization config, then finalize backend resolution with loaded quant_config to enable quantization-aware backend selection.
Test waiver cleanup tests/integration/test_lists/waives.txt	Removed waiver for Qwen3_30B_A3B MXFP8 latency test, restoring it to active test execution.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Suggested reviewers

• yuxianq
• niukuo
• zhenhuaw-me
• jieli-matrix

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 33.33% 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 critical technical details. It references PR#14182 but provides no explanation of what is being fixed, why, or the technical changes involved.	Expand the description to explain the issue being fixed (MoE backend selection on Blackwell with FP8_BLOCK_SCALES), the solution approach, and reference relevant test coverage. Include why this follows PR#14182.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The PR title clearly and specifically describes the main change: updating the resolve_moe_backend method to be quantization-aware on specific hardware.
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.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[tensorrt-cicd]

PR_Github #49042 [ run ] triggered by Bot. Commit: b38eb84 Link to invocation

[tensorrt-cicd]

PR_Github #49042 [ run ] completed with state SUCCESS. Commit: b38eb84
/LLM/main/L0_MergeRequest_PR pipeline #38778 completed with status: 'SUCCESS'

CI Report

Link to invocation

[heyuhhh]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #49224 [ run ] triggered by Bot. Commit: 25d6fd7 Link to invocation

[tensorrt-cicd]

PR_Github #49224 [ run ] completed with state SUCCESS. Commit: 25d6fd7
/LLM/main/L0_MergeRequest_PR pipeline #38897 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.

CI Report

Link to invocation