[chenfeiz0326]

Dev Engineer Review

The change adds GB200 GLM-5 aggregated cases and Qwen3-235B context-only, generation-only, and generation-only-without-context cases to the multi-node QA list. It changes the mode documentation and no configuration files. Review finding counts are unavailable.

QA Engineer Review

Modified tests/integration/test_lists/qa/llm_perf_multinode.txt. The entries cover the requested GB200 scenarios. The PR objectives report successful list validation and QA cluster runs, but those results were not independently verified here. Coverage verdict: needs follow-up. Confirm that the QA runner uses the default AGG_CONFIG_FOLDER for the aggregated cases.

Per-File QA Perspective

• tests/integration/test_lists/qa/llm_perf_multinode.txt: Adds manual-QA performance cases for GB200 GLM-5 and Qwen3-235B. Verify the aggregated cases with the QA runner’s default configuration.

Description

The GB200 perf-sanity cases that CI runs post-merge are all Qwen3-235B FP4 or GLM-5 FP4: 12 cases in the
l0_gb200_*perf_sanity* test-db lists. Only one of them,
disagg-e2e-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL, is in the QA multi-node perf
list. This PR adds the other 11 to tests/integration/test_lists/qa/llm_perf_multinode.txt, in the list's own
spelling (no _upload):

Section in the list	Model	Cases	Nodes (as in CI)
aggr (new)	GLM-5 FP4	glm5_fp4_dep8_mtp1_8k1k, glm5_fp4_tep8_mtp3_8k1k	2
ctx_only	Qwen3-235B FP4	con1, con64, con1024	1
gen_only	Qwen3-235B FP4	con1, con64 / con1024	2 / 3
gen_only_no_context (new)	Qwen3-235B FP4	con1, con64 / con1024	1 / 2

No config file is added or changed:

• The Qwen3-235B ids resolve against the QA copies of the three configs in tests/scripts/perf/disaggregated/, the
  folder QA's runs read through DISAGG_CONFIG_FOLDER (every disagg entry already in this list resolves there).
  Those copies differ from the tests/scripts/perf-sanity/disaggregated/ ones only by a disabled accuracy: block
  and an explicit transceiver_runtime: PYTHON, which is that field's default, so QA runs the same cases as CI.
• The two GLM-5 cases are aggregated. Their config, glm5_fp4_2_nodes_grace_blackwell.yaml, is in
  tests/scripts/perf-sanity/aggregated/, the default AGG_CONFIG_FOLDER; the aggregated configs of
  llm_config_database.yml resolve there too, and tests/scripts/perf/ has no aggregated folder. These are the
  first plain aggregated ids in this list, so please confirm that the QA multi-node runner leaves
  AGG_CONFIG_FOLDER at its default.

llm_perf_multinode_basic.txt is generated by the trtllm-ci selector and is not edited here; re-running the
selector lets the basic tier consider the new cases.

Test Coverage

• scripts/format_test_list.py and scripts/check_test_list.py --validate (the test lists format and
  validate-test-lists pre-commit hooks) pass.
• Rebuilt the id set the way test_perf_sanity.py generates it, with
  DISAGG_CONFIG_FOLDER=tests/scripts/perf/disaggregated and the default AGG_CONFIG_FOLDER: all 93 entries of the
  list resolve, the 11 new ones included; no duplicates; all 12 GB200 cases of the CI test-db are now listed.
• The same 12 cases run in every L0_PostMerge build; all GB200 PerfSanity stages passed in builds 3043–3050.
  The 7 multi-node ones were also run on QA's GB200 cluster with the CI wheel of 7723a3b34, 2 reps each, all
  passing with no failed requests.

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.

🤖 Generated with Claude Code

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

The multi-node QA test list adds GB200 GLM-5 and Qwen3-235B performance cases. It also updates the note about configuration YAML reuse.

Changes

GB200 multi-node performance cases

Layer / File(s)	Summary
GLM-5 aggregated-serving cases tests/integration/test_lists/qa/llm_perf_multinode.txt	The note includes gen_only_no_context among modes that reuse disaggregated configuration YAMLs. Two GB200 GLM-5 aggregated-serving cases are added.
Qwen3-235B mode coverage tests/integration/test_lists/qa/llm_perf_multinode.txt	Cases are added for context-only, generation-only, and generation-only without context at concurrency 1, 64, and 1024.

Estimated code review effort: 2 (Simple) | ~8 minutes

Change: Other

Merge Risk: Minimal · up to 4f829

No concrete merge-blocking risk was identified in the added GB200 performance cases.

Pre-merge checks: 5

Check name	Status	Explanation
Title check		The title names the two model families added (GB200 Qwen3-235B FP4 and GLM-5 FP4) and the target file (QA multi-node perf list), which matches the changeset that adds 11 perf-sanity cases to llm_perf_…
Description check		The description follows the repository template. It explains the problem (12 GB200 post-merge perf-sanity cases, only one listed) and the solution (11 added cases, with a table by section, model, and …
Docstring Coverage		No eligible functions found in the changed files. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentat…
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Full details: Title check

Explanation

The title names the two model families added (GB200 Qwen3-235B FP4 and GLM-5 FP4) and the target file (QA multi-node perf list), which matches the changeset that adds 11 perf-sanity cases to llm_perf_multinode.txt. It follows the [None][test] template with a valid ticket tag and type. It does not describe the new aggregated section or the gen_only_no_context mode, but it covers the primary change.

Full details: Description check

Explanation

The description follows the repository template. It explains the problem (12 GB200 post-merge perf-sanity cases, only one listed) and the solution (11 added cases, with a table by section, model, and node count). It states that no config file changed and why the QA config copies resolve the same cases. The Test Coverage section lists the format and validation hooks, the id-set rebuild, and the CI and QA runs. The checklist is marked. One gap: the template's Description section asks for a short issue and solution statement, and this section is long, though it still covers the required content. The description also asks reviewers to confirm the AGG_CONFIG_FOLDER default, which is a useful open question.

Full details: Docstring Coverage

Explanation

No eligible functions found in the changed files. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentation-only edits and deletions within surviving functions are included; unchanged diff context and fully deleted functions are excluded. Analyzed 0 functions across 0 files. (1 skipped: 1 unsupported.)

Finishing Touches

Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: NVIDIA/TensorRT-LLM
• Branch: main
• File: .coderabbit.yaml
• Commit: 9c0e412
• Layers: Repository YAML
• Review profile: CHILL
• Plan: Enterprise
• Run ID: e64d8af3-4bad-4e94-8cd1-b9f04670bf5d

Commits

Reviewing files that changed from the base of the PR and between 9c0e412 and 4f82958.

Files selected for processing: 1

• tests/integration/test_lists/qa/llm_perf_multinode.txt

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 10 remain after this review.