[https://nvbugs/6185248][test] Unwaive K2.5 thinking MTP3 perf sanity test

[tianyuxbear]

Summary

Re-enables test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]. The underlying defect tracked by nvbug 6185248 is already fixed on main by PR #14379; this PR only removes the waive entry that was added in parallel.

Background

Nvbug 6185248: AttributeError: 'KimiK25ForConditionalGeneration' object has no attribute 'draft_config' at tensorrt_llm/_torch/pyexecutor/model_loader.py when MTP speculative decoding is enabled. Root cause: PR #12788 refactored KimiK25ForConditionalGeneration into a composition pattern (self.llm = DeepseekV3ForCausalLM(...)), but the loader expected draft_config directly on the top-level model.

PR #14379 fixed this by adding draft_config / draft_model / load_draft_weights property forwarders on KimiK25ForConditionalGeneration (delegating to self.llm) and threading **kwargs through KimiK25.forward so spec_metadata reaches DeepseekV3's MTP forward path.

Why a separate unwaive PR

The fix PR #14379 was merged but did not remove the waive line, due to the merge ordering between the fix PR and the waive PR:

Time (UTC+8)	Event	Commit
5/21 09:08	PR #14379 (fix) opened, base = b3fcf083 (waive does not exist yet on main)	—
5/21 18:21	PR #14405 (waive) merged → kimi waive line added to main	736d580f80
5/22 09:34	PR #14379 (fix) merged; diff only touches modeling_kimi_k25.py	eb6ee930

Verification: git merge-base --is-ancestor 736d580f80 b3fcf083 → NO. PR #14379's base predates the waive, so #14379's diff does not include an unwaive change. This PR removes the now-stale waive entry explicitly.

[tianyuxbear]

/bot run --extra-stage "GB200-4_GPUs-PyTorch-PerfSanity-Post-Merge-1"

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR removes a single test-waive entry from tests/integration/test_lists/waives.txt, un-waiving a specific perf/test_perf_sanity.py::test_e2e scenario with the kimi-k25-thinking configuration so it will no longer be skipped during test execution.

Changes

Performance Test Waiver Removal

Layer / File(s)	Summary
Remove kimi-k25-thinking perf test waiver tests/integration/test_lists/waives.txt	Delete the waiver entry that skipped perf/test_perf_sanity.py::test_e2e for the aggr_upload kimi-k25-thinking configuration, enabling the test to run.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#14387: Removes a specific SKIP waiver entry from the same tests/integration/test_lists/waives.txt file.
• NVIDIA/TensorRT-LLM#14341: Modifies the same waivers file by removing a specific SKIP entry for an integration test scenario.

Suggested reviewers

• ZhanruiSunCh
• yuxianq

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and specifically describes the main change: removing a waive entry for a K2.5 thinking MTP3 perf sanity test, with proper ticket reference format [https://nvbugs/6185248][test].
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.
Description check	✅ Passed	The PR description clearly explains the issue, background, fix, and rationale for the separate unwaive PR with detailed timeline and verification steps.

_{✏️ 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 #49937 [ run ] triggered by Bot. Commit: 858a79f Link to invocation

[tensorrt-cicd]

PR_Github #49937 [ run ] completed with state SUCCESS. Commit: 858a79f
/LLM/main/L0_MergeRequest_PR pipeline #39509 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

[tianyuxbear]

/bot run --extra-stage "GB200-4_GPUs-PyTorch-PerfSanity-Post-Merge-1" --disable-fail-fast

[tensorrt-cicd]

PR_Github #50036 [ run ] triggered by Bot. Commit: 858a79f Link to invocation

[tensorrt-cicd]

PR_Github #50036 [ run ] completed with state SUCCESS. Commit: 858a79f
/LLM/main/L0_MergeRequest_PR pipeline #39599 completed with status: 'SUCCESS'

CI Report

Link to invocation