[TRTLLM-15264][test] Wire KDA disagg transfer tests into CI (cpu_only CPU stage + l0_b200 GPU)

[brnguyen2]

Description

Follow-up to review feedback on #17334, wiring tests/unittest/disaggregated/test_kda_mamba_transfer.py into CI. Before this change the file was in no test list except the dir-level unittest/disaggregated entry that only the CPU stage consumes — and it was silently dropped there too, so none of its tests ran anywhere.

Two commits:

1. Mark the CPU-runnable tests cpu_only (T3 of TRTLLM-15264).
The CPU-Generic-* stages run pytest with -m cpu_only, and conftest.py's pytest_ignore_collect skips any file in unittest/disaggregated that contains no pytest.mark.cpu_only text. The two GPU-free tests — test_kda_peer_validation_synthetic_cpu and test_kda_peer_validation_allows_pipeline_parallel_layer_split (synthetic rank-info / page-table builders + MambaPolicy.validate_peer_compatible, no CUDA manager) — now carry the marker and run on the CPU stage.

2. Wire the GPU tests into l0_b200.
The remaining tests (test_kda_transfer NIXL loopback + the manager-backed peer-validation cases) build simulated TP ranks in one process on a single device, so a single-GPU Blackwell stage suffices; added to the Kimi K3 block of l0_b200 alongside the other single-GPU K3 KDA suites. The GPU stage's -m "not cpu_only" markexpr selects exactly these and skips the two cpu_only tests.

Net: the CPU subset runs on CPU-Generic-*, the GPU subset on l0_b200, with the single cpu_only marker partitioning them.

Test Coverage

No product code and no test logic changed — marker + test-list wiring only. Not run locally (dev box has no torch/CUDA stack); CI now executes the tests it previously skipped. If the GPU cases show a B200-specific issue (cf. test_kda_prefill_op.py, which NaNs on B200 but passes on GB300 — TRTLLM-15204), the file can move to l0_gb300_multi_gpus.

PR Checklist

• PR title follows the [TRTLLM-15264][test] convention
• pre-commit run on the touched files (incl. test-list AST validation)

Dev Engineer Review

• Added pytest.mark.cpu_only to:
  • test_kda_peer_validation_synthetic_cpu
  • test_kda_peer_validation_allows_pipeline_parallel_layer_split
• Added unittest/disaggregated/test_kda_mamba_transfer.py to tests/integration/test_lists/test-db/l0_b200.yml.
• The marker matches the l0_cpu CI filter.
• Manager-backed and NIXL transfer tests remain GPU-only.
• The test path and list format are consistent.
• No product code or test logic changed.
• No correctness, performance, API, or regression concerns identified.

QA Engineer Review

• Modified test functions:
  • test_kda_peer_validation_synthetic_cpu
  • test_kda_peer_validation_allows_pipeline_parallel_layer_split
• Added unittest/disaggregated/test_kda_mamba_transfer.py to tests/integration/test_lists/test-db/l0_b200.yml.
• The GPU tests are covered by the B200 CI test list.
• The two CPU-only tests are selected by the l0_cpu CI sweep.
• Verdict: sufficient

[brnguyen2]

/bot run

[coderabbitai]

Walkthrough

Two disaggregated validation tests are marked as CPU-only. The KDA Mamba transfer test module is added to the B200 pre-merge test list.

Changes

Test execution

Layer / File(s)	Summary
CPU-only validation scheduling tests/unittest/disaggregated/test_kda_mamba_transfer.py, tests/integration/test_lists/test-db/l0_b200.yml	The synthetic peer-validation matrix test and pipeline-parallel layer-split validation test use pytest.mark.cpu_only. The test module is added to the B200 pre-merge configuration, with CPU-only cases assigned to the CPU-Generic stage.

Estimated code review effort: 1 (Trivial) | ~2 minutes

Suggested reviewers: bowenfu, schetlur-nv

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.
Title check	✅ Passed	The title clearly summarizes the main change: wiring KDA disaggregated transfer tests into CPU and B200 GPU CI stages.
Description check	✅ Passed	The description explains the purpose, test partitioning, coverage, CI behavior, and checklist items with sufficient detail.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[tensorrt-cicd]

PR_Github #65166 [ run ] triggered by Bot. Commit: 60e6720 Link to invocation

[brnguyen2]

/bot run

[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/test-db/l0_b200.yml`:
- Around line 102-104: Add the standard NVIDIA copyright header at the beginning
of l0_b200.yml, preserving the existing test-list entries and formatting.

🪄 Autofix

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: cccb855e-f124-4aed-9f56-08ca5a5b469e

📥 Commits

Reviewing files that changed from the base of the PR and between 60e6720 and ff7acc7.

📒 Files selected for processing (1)

• tests/integration/test_lists/test-db/l0_b200.yml

[tensorrt-cicd]

PR_Github #65170 [ run ] triggered by Bot. Commit: ff7acc7 Link to invocation

[tensorrt-cicd]

PR_Github #65166 [ run ] completed with state ABORTED. Commit: 60e6720

Link to invocation

[tensorrt-cicd]

PR_Github #65170 [ run ] completed with state SUCCESS. Commit: ff7acc7
/LLM/main/L0_MergeRequest_PR pipeline #52964 completed with status: 'SUCCESS'

CI Report

Link to invocation