[NVBUG-6448152][test] TEST ONLY native midpoint async-consensus discriminator

[chienchunhung]

TEST ONLY — DO NOT REVIEW OR MERGE.

This draft is the sole C++ source discriminator for the separate later-main throughput regression. It does not change the production proposal in the official C++ async-consensus review.

Factor under test

• Native source parent: 3ab71753
• Treatment: the official async-consensus factor applied byte-for-byte
• Stable patch ID: 549e0085a06212cde0c47ab47072ad6acdcd1f58
• Current CI image, exact workload YAML, selector, harness, and timeouts are unchanged
• No diagnostic instrumentation or additional product changes

This checkpoint is the audited compatible midpoint between the first current-runtime-compatible boundary and the measured-slow native checkpoint.

Required evidence

Run only:

GB300-12_GPUs-3_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE2-GPU8-Post-Merge-1

Accept only an exact-head run with the exact selector, coordinator active on all four CTX PP ranks, every request successful, clean shutdown, and a valid official output-token metric. Any timeout or failed request censors throughput.

Interpretation

• At least 1402 output tok/s: regression is later than this checkpoint.
• About 800 output tok/s: regression is already present at or before this checkpoint.
• Build/import failure: compatibility-only and not performance evidence.

Close this diagnostic immediately after terminal evidence; preserve the branch for reproducibility.

[chienchunhung]

/bot run --disable-fail-fast --stage-list "GB300-12_GPUs-3_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE2-GPU8-Post-Merge-1"

[tensorrt-cicd]

PR_Github #61314 [ run ] triggered by Bot. Commit: 42cf8ff Link to invocation

[tensorrt-cicd]

PR_Github #61314 [ run ] completed with state FAILURE. Commit: 42cf8ff
/LLM/main/L0_MergeRequest_PR pipeline #49545 (Partly Tested) 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