[NVBUG-6448152][test] TEST ONLY; DO NOT REVIEW: current-runtime source midpoint

[chienchunhung]

TEST ONLY; DO NOT REVIEW

This draft is a bounded midpoint-checkpoint discriminator for the remaining
current-tree throughput regression. It is not a proposed product change.

What this isolates

• Product checkpoint: legacy C++ module removal checkpoint.
• Treatment: the byte-identical asynchronous-consensus factor from the
  official C++ asynchronous-consensus change.
• Runtime: unchanged current image build 202607151440-16194 for x86 and SBSA.
• Workload: unchanged exact GB300 PP4 perf-sanity YAML and selector.
• CI mergeability: current main is recorded only as a no-tree parent; the factor
  tree is unchanged.

The checkpoint is exactly halfway through the 92-commit source interval that
already uses this current runtime image: 46 commits precede it and 46 follow it.
The current image tags, both exact workload YAML blobs, and every preimage used
by the asynchronous-consensus factor match the current official base. There is
no compatibility patch, timeout change, workload change, or semantic adaptation.

Interpretation

• At least 1402 output tokens/s establishes a fast endpoint under the current
  runtime and localizes the regression to the 46 later source commits.
• Approximately 800 output tokens/s shows that the slowdown was already
  present at this checkpoint and moves the next discriminator into the earlier
  half.
• Any partial recovery will be reported relative to the current 794.61
  output tokens/s result and the historical 1548.84 output tokens/s result.
• A build or infrastructure failure is inconclusive and will not be interpreted
  as performance evidence.

Verification

• Exact factor diff SHA-256:
  25e394aac54636e88433755595813bd89704cfa2861bd9167303a7beb9694867.
• The factor commit and no-tree merge commit have the same tree:
  33cc3f639ff6b4f1415e919c4c3fd8d2b9a2c718.
• git diff --check, clang-format, CMake format, codespell, DCO, and the
  applicable pre-commit checks passed.
• Targeted run pending for
  GB300-12_GPUs-3_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE2-GPU8-Post-Merge-1.

[chienchunhung]

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

[tensorrt-cicd]

PR_Github #61099 [ run ] triggered by Bot. Commit: dec24cf Link to invocation

[tensorrt-cicd]

PR_Github #61099 [ run ] completed with state FAILURE. Commit: dec24cf
/LLM/main/L0_MergeRequest_PR pipeline #49353 (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

[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 #61103 [ run ] triggered by Bot. Commit: dec24cf Link to invocation

[tensorrt-cicd]

PR_Github #61103 [ run ] completed with state SUCCESS. Commit: dec24cf
/LLM/main/L0_MergeRequest_PR pipeline #49357 (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