[https://nvbugs/6535779][fix] Remove stale Qwen3.5 and DeepSeekV32 waivers

[VALLIS-NERIA]

What

• Remove the three stale Qwen3.5-397B-A17B waivers associated with NVBug 6535779.
• Remove the stale B300 DeepSeekV32 chunked-prefill latency waiver associated with NVBug 6483369, moved from [https://nvbugs/6483369][fix] Keep the engineer-confirmed one-line stale-waiver removal, and add a… #17438.
• Restore CI coverage for all four cases.
• Keep [https://nvbugs/6483369][fix] Keep the engineer-confirmed one-line stale-waiver removal, and add a… #17438's MoE unit-test changes out of scope.

Why

NVBugs 6535779 and 6483369 are fixed, so retaining these entries hides valid integration coverage. The NVBug 6483369 waiver removal is moved from #17438; its MoE unit-test changes are intentionally discarded. This fresh change supersedes #17611 because that branch now conflicts with the current main waiver list.

Validation

• Commit-time pre-commit suite for the moved waiver deletion: all hooks passed, including duplicate-waiver and 2,073-entry AST validation.
• Exact B300 DeepSeekV32 case from [https://nvbugs/6483369][fix] Keep the engineer-confirmed one-line stale-waiver removal, and add a… #17438: MMLU 87.99, GSM8K 96.25; 1 passed.
• python3 -m pre_commit run --files tests/integration/test_lists/waives.txt
• Full B200 x4 E2E:
  • TestQwen3_5_397B_A17B::test_nvfp4_mtp3_gdn_replay_tep4
  • TP4/EP4, MTP3, cached GDN replay, V2 cache manager, full GSM8K (1,319 samples)
  • accuracy: 95.53 average (91.797 threshold)
  • 1 passed in 809.08 s; executor evaluation time 74.219 s
  • logs confirmed MambaHybridCacheManagerV2, KVCacheV2Scheduler, cached replay, and replay state updates enabled

Replay manager performance sanity check

Qwen3.5-35B-A3B-FP8 on one B200, MTP3 cached replay, fixed 128-request synthetic workload (ISL 2048 / OSL 512, concurrency 64), two reversed-order A/B pairs:

• V1 mean: 11.345 req/s, 5,808.7 output tok/s
• V2 mean: 12.697 req/s, 6,501.1 output tok/s
• V2: +11.9% throughput, -10.7% total latency, -8.7% average request latency
• Both pairwise comparisons favored V2 (+10.5% and +13.4%)

This is a directional executor-level comparison; GPU clocks were not locked, and the second pair used an otherwise idle GPU on a shared host.

Dev Engineer Review

• Removed four stale waiver entries from tests/integration/test_lists/waives.txt.
• Removed three Qwen3.5-397B-A17B NVFP4 waivers for fixed NVBug 6535779.
• Removed one B300 DeepSeekV32 chunked-prefill latency waiver for fixed NVBug 6483369.
• Restored CI coverage for all four cases.
• Added no new waivers.
• Made no code, API, or performance changes.
• Preserved unrelated MoE unit-test changes outside this PR scope.
• CI failed and requires follow-up before merge.

QA Engineer Review

• Modified only tests/integration/test_lists/waives.txt.
• Modified no test-db/ or qa/ files.
• Removed three Qwen3.5-397B-A17B waiver entries and one B300 DeepSeekV32 waiver entry.
• Added no test-list entries.
• CBTS coverage data is unavailable.
• Verdict: needs follow-up.

[VALLIS-NERIA]

/bot run --extra-stage "DGX_B200-4_GPUs-PyTorch-Post-Merge-1, DGX_B200-4_GPUs-PyTorch-Post-Merge-2, DGX_B200-4_GPUs-PyTorch-Post-Merge-3, DGX_B200-4_GPUs-PyTorch-Post-Merge-4"

[tensorrt-cicd]

PR_Github #66118 [ run ] triggered by Bot. Commit: d3358f0 Link to invocation

[tensorrt-cicd]

PR_Github #66118 [ run ] completed with state FAILURE. Commit: d3358f0
/LLM/main/L0_MergeRequest_PR pipeline #53793 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

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 7e26695d-fb70-488d-90ae-565dfc15d461

📥 Commits

Reviewing files that changed from the base of the PR and between 5972811 and 3784d5a.

📒 Files selected for processing (1)

• tests/integration/test_lists/waives.txt

🚧 Files skipped from review as they are similar to previous changes (1)

• tests/integration/test_lists/waives.txt

---

Walkthrough

The integration test waive list removes three Qwen3.5 397B NVFP4 skip entries and one B300 DeepSeekV32 chunked-prefill skip entry.

Changes

Integration test waive-list cleanup

Layer / File(s)	Summary
Remove obsolete waive entries tests/integration/test_lists/waives.txt	The list removes three Qwen3.5 397B entries and one B300 DeepSeekV32 multi-GPU chunked-prefill latency entry.

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

Merge Risk: ⚪ Minimal · up to 3784d

This change restores coverage by removing four stale test waivers, and no actionable merge-blocking risk remains after normal checks and review.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title uses the required NVBugs URL and [fix] format and accurately identifies removal of stale Qwen3.5 and DeepSeekV32 waivers.
Description check	✅ Passed	The description clearly covers the issue, solution, scope, and validation with relevant test results.
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.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[VALLIS-NERIA]

/bot run

[tensorrt-cicd]

PR_Github #66465 [ run ] triggered by Bot. Commit: 3784d5a Link to invocation

[tensorrt-cicd]

PR_Github #66465 [ run ] completed with state FAILURE. Commit: 3784d5a
/LLM/main/L0_MergeRequest_PR pipeline #54107 completed with status: 'UNSTABLE'

CI Report

⚠️ Multi-GPU Label Required:
Multi-GPU tests require the ci: full pre-merge approved label on this PR. Ask a member of NVIDIA/trt-llm-ci-approvers to add the label, then re-trigger CI with the same bot command (no rebase needed).

⚠️ 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

Link to invocation