[https://nvbugs/6193836][test] Use EP=8 + attention DP for minimax_m2.5 8-GPU perf

[ruodil]

MiniMax-M2.5 FP8 has intermediate_size=1536 and weight_block_size=128. TRT-LLM-gen / CUTLASS / DeepGEMM FP8 MoE kernels require the per-rank intermediate size to be a multiple of the block size 128. Under TP=8 each rank gets 1536/8=192, which fails the assert. Per developer guidance, route MoE through EP=8 and rely on attention DP instead of TP.

Changes:

• llm_perf_core.yml: switch the 7 minimax_m2.5_fp8 8-GPU test names from tp:8-gpus:8 to ep:8-gpus:8.
• pytorch_model_config.py: add a pattern matching exactly those 7 cases and enable attention_dp: True in the generated trtllm-bench config.

The 4-GPU tests (TP=4 -> 1536/4=384) are unaffected and not touched.

Fixes: NVBugs 6193836.

Summary by CodeRabbit

• Tests

  • Updated performance testing configurations for PyTorch model optimization evaluation across multi-GPU setups.

• Chores

  • Enhanced internal benchmarking infrastructure with new parallelism strategy configurations.

Description

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.

[ruodil]

/bot skip --comment "skip test as just modifying cases"

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR updates the minimax_m2.5_fp8 performance testing configuration to use expert parallelism (EP=8) with attention-DP enabled. A new pattern config entry is added to enable attention distributed parallelism, and test parameters are updated to switch from tensor parallelism (TP=8) to expert parallelism (EP=8).

Changes

minimax_m2.5_fp8 Attention DP Configuration

Layer / File(s)	Summary
Pattern config and test parameter updates for EP=8 + attention-DP tests/integration/defs/perf/pytorch_model_config.py, tests/integration/test_lists/qa/llm_perf_core.yml	A pattern config entry is added for minimax_m2.5_fp8 on 8-GPU setups to enable enable_attention_dp: True for multiple input_output_len/maxbs variants. Corresponding test cases are updated from tp:8 to ep:8 to activate this configuration across several test runs.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Suggested reviewers

• 2ez4bz
• yuxianq

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The PR title clearly specifies the change: using EP=8 + attention DP for minimax_m2.5 8-GPU perf tests, directly matching the file modifications.
Description check	✅ Passed	The PR description provides detailed technical context (divisibility issue, kernel requirements, intermediate size calculations), explains the solution, lists all changes, and includes the bug fix reference.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
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.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/integration/test_lists/qa/llm_perf_core.yml`:
- Around line 324-331: The QA perf list changed the minimax_m2.5_fp8 rows to use
ep:8-gpus:8 but the corresponding test-db perf YAMLs were not updated; search
for entries named minimax_m2.5_fp8 (and any minimax / m2.5 variants) under the
test-db perf lists and update their rows to match the QA values (replace
whatever EP/GPUs fields they have with ep:8-gpus:8, including the
maxbs/max_throughput and min_latency variants), or if the model is intentionally
not covered add a short YAML comment explaining why; ensure you update all
occurrences so CI mirrors QA.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 5e36aa5d-3eb9-459c-a8b9-b588a3e8e506

📥 Commits

Reviewing files that changed from the base of the PR and between 276ccd6 and c1bdd95.

📒 Files selected for processing (2)

• tests/integration/defs/perf/pytorch_model_config.py
• tests/integration/test_lists/qa/llm_perf_core.yml

[tensorrt-cicd]

PR_Github #50445 [ skip ] triggered by Bot. Commit: c1bdd95 Link to invocation

[tensorrt-cicd]

PR_Github #50445 [ skip ] completed with state SUCCESS. Commit: c1bdd95
Skipping testing for commit c1bdd95

Link to invocation