perf(agentx): Add B300 vLLM MTP TP4 data points

[ivanium]

Summary / 摘要

• Append GPU-resident TP4 concurrency 1, 2, 4, and 6 points to dsv4-fp4-b300-vllm-agentic-mtp.

• Add one perf-changelog.yaml entry at the physical end of the file with append-only: true.

• 为 dsv4-fp4-b300-vllm-agentic-mtp 追加 GPU-resident TP4 并发 1、2、4 和 6 测试点。

• 仅在 perf-changelog.yaml 文件末尾追加一条记录，并设置 append-only: true。

Validation / 验证

• Official current-main-to-head validate_perf_changelog.py: passed / 官方 current-main-to-head validate_perf_changelog.py：通过
• Exact-key generation: 11 base points → 15 head points; four additions and zero removals / 精确配置生成：11 个基础点 → 15 个分支点；新增 4 个，删除 0 个
• YAML parsing, Bash syntax, and git diff --check: passed / YAML 解析、Bash 语法和 git diff --check：通过
• Focused changelog and gating tests: 118 passed / 变更日志与门禁专项测试：118 项通过

[github-actions]

Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs

---

感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档

[claude]

Claude finished @ivanium's task in 1m 11s —— View job

---

PR Review

• Gather context and read the PR diff
• Verify the new image is publicly accessible on Docker Hub
• Validate perf-changelog entry placement and format

LGTM - no blocking issues found

• The only functional change is the image tag bump for dsv4-fp4-b300-vllm-agentic-mtp (426e59f → d62ad46) in configs/nvidia-master.yaml; no recipe, model, or matrix values were touched.
• vllm/vllm-openai:nightly-dev-x86_64-cu13.0.1-d62ad46 is a valid public Docker Hub reference. (Note: outbound network access wasn't available in this review environment, so I could not independently confirm the manifest — relying on the author's stated manifest check.)
• The perf-changelog.yaml entry is correctly appended to the end of the file, references the matching config key, and links this PR (perf(agentx): Add B300 vLLM MTP TP4 data points #2647).

[claude]

LGTM — straightforward image tag bump with matching changelog entry.

What was reviewed: confirmed the dsv4-fp4-b300-vllm-agentic-mtp image tag update in configs/nvidia-master.yaml (426e59f → d62ad46) is the only recipe field touched, and that perf-changelog.yaml has a matching new entry with correct config-key and PR link.

Extended reasoning...

Overview

This PR bumps the Docker image tag for the dsv4-fp4-b300-vllm-agentic-mtp AgentX recipe in configs/nvidia-master.yaml (from nightly-dev-x86_64-cu13.0.1-426e59f to ...-d62ad46) and adds a corresponding entry to perf-changelog.yaml. No other fields (tp/ep, model, runner, concurrency sweep) were changed.

Security risks

None. This only changes a pinned third-party vLLM nightly image tag used for internal benchmark runs; no code execution paths, auth, or credentials are touched.

Level of scrutiny

Low. This is a mechanical, single-field config change following the repository's established pattern for image bumps (same pattern as recent PRs #2621, #2619, #2571). The PR description confirms the manifest exists and YAML/bash validation passed.

Other factors

No prior reviews or outstanding comments on this PR besides the standard automated recipe-reminder bot comment. The change is small, self-contained, and low blast-radius (affects only this one benchmark recipe's image pin).

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32133529869
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32133529869

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32134936706
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32134936706

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32193512662
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32193512662

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32233742015
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32233742015

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32243066761
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32243066761

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32243066761
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32243066761

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32315429801
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32315429801

[xinli-sw]

/reuse-sweep-run

[cquil11]

/stage-results 32315429801

[github-actions]

@cquil11 staged run 32315429801: https://inferencemax-app-git-staging-semianalysisai.vercel.app/inference?i_dates=2026-08-19~r32315429801

This run remains available across future /stage-results requests. Staging the same run ID again updates its staged data. Staging workflow