[https://nvbugs/6707518][fix] Fix Kimi K3 spec dec test

[mikeiovine]

Description

Fix the harness. Logprobs with spec dec has never been supported, but it previously used to warn instead of crashing.

#18401 broke this test by introducing the hard error. Fix it by just disabling logprobs when spec dec is on. The behavior of the test is unchanged.

Test Coverage

Existing tests.

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.

Dev Engineer Review

• The harness disables logprob requests during speculative decoding.
• Generation and JSON output use the adjusted logprob setting.
• Empty logprob lists are omitted from JSON output.
• Documentation explains the one-sided parity check.
• The change prevents the hard error caused by unsupported speculative-decoding logprobs.
• No public API or configuration changes were found.
• The scope is limited to the Kimi K3 test harness and its speculative-decoding test.

QA Engineer Review

• Modified test: test_kimi_k3_specdec documentation.
• No test functions were added or removed.
• Existing tests cover speculative decoding and provide regression coverage for the logprob limitation.
• No test-list files were modified.
• Verdict: sufficient.

[mikeiovine]

/bot run

[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: 5fb9b3ec-d02f-46c1-a96c-a745d0621b53

📥 Commits

Reviewing files that changed from the base of the PR and between f7b7f22 and e5b281c.

📒 Files selected for processing (2)

• tests/integration/defs/kimi_k3_sa_harness.py
• tests/integration/defs/test_kimi_k3_specdec.py

🚧 Files skipped from review as they are similar to previous changes (2)

• tests/integration/defs/test_kimi_k3_specdec.py
• tests/integration/defs/kimi_k3_sa_harness.py

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

---

Walkthrough

The Kimi K3 speculative-decoding harness disables logprob collection for spec runs, reports one-sided parity, omits empty logprob fields, and documents aggregate divergence handling.

Changes

Speculative-decoding logprob handling

Layer / File(s)	Summary
Speculative logprob request and serialization flow tests/integration/defs/kimi_k3_sa_harness.py	The harness disables logprob collection for speculative runs, uses the adjusted flag for generation and output dumping, and omits empty logprob lists.
Speculative parity test contract tests/integration/defs/test_kimi_k3_specdec.py	The test documentation describes baseline-distribution checks, warnings for isolated non-ties, and failure for aggregate non-tie dominance.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: ⚪ Minimal · up to e5b28

This updates the speculative-decoding test harness to avoid unsupported logprob requests; no current merge-blocking risk is identified.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 2 functions across 2 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title follows the required ticket and type format and clearly identifies the Kimi K3 speculative decoding test fix.
Description check	✅ Passed	The description explains the issue, identifies the cause, states the solution, and lists existing test coverage. The checklist is complete and marked as reviewed.
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.

• Fix all pre-merge checks with AI

✨ 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 #71333 [ run ] triggered by Bot. Commit: 6aa64bb Link to invocation

[sylvesterkaczmarek]

Changing the dump condition from comp.logprobs is not None to truthiness hides an unexpected empty logprob result. On a baseline run where logprobs were requested, comp.logprobs=[] should remain visible (or fail) rather than becoming indistinguishable from 'logprobs were not requested', otherwise the parity fixture can silently lose coverage.

[tensorrt-cicd]

PR_Github #71333 [ run ] completed with state SUCCESS. Commit: 6aa64bb
/LLM/main/L0_MergeRequest_PR pipeline #58456 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]

Note

GitHub couldn't provide a complete incremental comparison for this pull request, so CodeRabbit is performing a full review instead. This review may take a little longer.

[mikeiovine]

/bot run

[tensorrt-cicd]

PR_Github #71557 [ run ] triggered by Bot. Commit: e5b281c Link to invocation

[mikeiovine]

Changing the dump condition from comp.logprobs is not None to truthiness hides an unexpected empty logprob result. On a baseline run where logprobs were requested, comp.logprobs=[] should remain visible (or fail) rather than becoming indistinguishable from 'logprobs were not requested', otherwise the parity fixture can silently lose coverage.

I don't think that's right, baseline runs that request logprobs but return nothing should still fail after this change

[tensorrt-cicd]

PR_Github #71557 [ run ] completed with state SUCCESS. Commit: e5b281c
/LLM/main/L0_MergeRequest_PR pipeline #58645 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

[mikeiovine]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #71597 [ run ] triggered by Bot. Commit: e5b281c Link to invocation

[tensorrt-cicd]

PR_Github #71597 [ run ] completed with state SUCCESS. Commit: e5b281c
/LLM/main/L0_MergeRequest_PR pipeline #58684 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

[mikeiovine]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #71625 [ run ] triggered by Bot. Commit: e5b281c Link to invocation

[tensorrt-cicd]

PR_Github #71625 [ run ] completed with state SUCCESS. Commit: e5b281c
/LLM/main/L0_MergeRequest_PR pipeline #58712 completed with status: 'SUCCESS'

CI Report

Link to invocation