[https://nvbugs/6428087][fix] Extend the PP send payload with the flag (backward-compatible 2/3-tuple…

[trtllm-agent]

Summary

• Root cause: PP send/recv drops SampleStateTorch.use_host_stop_criteria, so non-last PP ranks default to False and call process_draft_tokens with an empty finish_reasons list from the last rank's host fast path.
• Fix: Extend the PP send payload with the flag (backward-compatible 2/3-tuple handling) so all PP ranks pick the same update branch; remove the waiver.
• Automated fix generated by repair-bot

Test plan

• Verify fix on the same GPU type as the original failure
• Check for regressions in related tests

Links

• Bug: https://nvbugs/6428087

Summary by CodeRabbit

• Bug Fixes

  • Improved consistency of generation stopping behavior across pipeline stages, reducing cases where host-based stop criteria could be lost during distributed execution.

• Tests

  • Updated integration test waivers to reflect a revised supported test configuration.

[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: f8c8b5cb-ffe8-48d2-b4b2-2f97b8b26e5e

📥 Commits

Reviewing files that changed from the base of the PR and between 5ce293f and c9b9bea.

📒 Files selected for processing (2)

• tensorrt_llm/_torch/pyexecutor/py_executor.py
• tests/integration/test_lists/waives.txt

💤 Files with no reviewable changes (1)

• tests/integration/test_lists/waives.txt

---

📝 Walkthrough

Walkthrough

The PR extends the PP (pipeline parallelism) ring broadcast for SAMPLE_STATE to propagate a use_host_stop_criteria flag alongside existing host and result-diff data across non-last PP ranks, and updates a test waiver entry for a DeepSeekV3Lite accuracy test with different parameter settings.

Changes

Sample State Broadcast Fix

Layer / File(s)	Summary
Propagate use_host_stop_criteria in PP ring broadcast tensorrt_llm/_torch/pyexecutor/py_executor.py	Send path now includes use_host_stop_criteria (via getattr) in the SAMPLE_STATE payload; receive path unpacks it and assigns it to sample_state.use_host_stop_criteria when present.
Test waiver update tests/integration/test_lists/waives.txt	Replaced the waived parameter combination for TestDeepSeekV3Lite::test_nvfp4_4gpus with moe_backend=CUTEDSL (tp2pp2), changing fp8kv, attention_dp, cuda_graph, overlap_scheduler, and torch_compile settings.

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

Possibly related PRs

• NVIDIA/TensorRT-LLM#16103: Both PRs modify tests/integration/test_lists/waives.txt to adjust DeepSeekV3Lite test waiver configurations.
• NVIDIA/TensorRT-LLM#16127: Directly related to the SAMPLE_STATE use_host_stop_criteria propagation fix, addressing the same regression this PR resolves.

Suggested reviewers: shuyixiong, mingyangHao, brb-nv

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title is specific and matches the main bug fix, including the NVBugs reference and fix type.
Description check	✅ Passed	It covers the summary, test plan, and bug link, but it omits the PR checklist section from the template.
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.}