[Klaud Cold] Revert "Add Kimi K2.6 NVFP4 B300 EAGLE3 AgentX benchmark" (#2158) / 回滚 Kimi K2.6 NVFP4 B300 EAGLE3 AgentX 基准测试（#2158）

[functionstackx]

Reverts #2158 (commit ececf32).

What this removes

• kimik2.5-fp4-b300-vllm-agentic-mtp config from configs/nvidia-master.yaml
• benchmarks/single_node/agentic/kimik2.5_fp4_b300_mtp.sh
• Kimi-K2.6-NVFP4 from the STAGED_MODELS allow-list in runners/launch_b300-nv.sh (no other config uses that model)
• The Add Kimi K2.6 NVFP4 B300 EAGLE3 AgentX benchmark / 新增 Kimi K2.6 NVFP4 B300 EAGLE3 AgentX 基准测试 #2158 entry in perf-changelog.yaml
• The exclude_none=True serialization change in utils/process_changelog.py

Conflict resolution

perf-changelog.yaml conflicted because #2222 (ingest recovery for #2158) appended a duplicate entry after it. The #2222 entry is kept — it belongs to a different commit that is not being reverted here.

Verification

• utils/test_process_changelog.py: 12 passed locally against the reverted code.
• No remaining references to kimik2.5-fp4-b300-vllm-agentic-mtp or Kimi-K2.6-NVFP4 outside historical changelog entries.

Expected CI behavior

The changelog validator treats perf-changelog.yaml as append-only, so a removal-only diff fails its check by design. This matches revert precedent #2164, which also removed the reverted entry; its push-to-main Run Sweep failed benignly with "No additions found in the changelog file". Merging via admin accordingly.

中文：回滚 #2158（Kimi K2.6 NVFP4 B300 EAGLE3 AgentX 基准测试）。保留 #2222 的数据摄取恢复条目；由于变更日志为只追加格式，删除条目会导致校验检查失败，与 #2164 回滚先例一致，故采用管理员合并。

🤖 Generated with Claude Code

[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

[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

[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

[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]

LGTM — clean, mechanical revert that matches the stated scope.

• Verified configs/nvidia-master.yaml, the benchmark script, and the runners/launch_b300-nv.sh allow-list entry are removed together with no other config referencing Kimi-K2.6-NVFP4.
• Confirmed the surviving kimik2.5-fp4-b300-vllm-agentic-mtp reference in perf-changelog.yaml belongs to PR #2222 (kept intentionally, per the conflict-resolution note) and is not a stray leftover.
• The process_changelog.py revert (dropping exclude_none=True) is a straight one-line rollback matching the reverted commit.

Extended reasoning...

Overview

Straight revert of PR #2158 (Kimi K2.6 NVFP4 B300 EAGLE3 AgentX benchmark), removing the master-config entry, the benchmark shell script, the STAGED_MODELS allow-list entry, the changelog entry, and the exclude_none=True serialization tweak in utils/process_changelog.py. No new logic is introduced.

Security risks

None. This is benchmark-configuration and CI-tooling code with no auth, crypto, or user-facing surface.

Level of scrutiny

Low. Reverts are inherently low-risk when they cleanly undo a prior commit, and this diff is a subset/mirror of the original PR's changes. I independently confirmed no dangling references to the removed config/model remain outside the deliberately-kept PR #2222 changelog entry.

Other factors

The PR body clearly explains the expected benign changelog-validator failure (append-only check) with precedent (#2164), and notes local test verification (test_process_changelog.py, 12 passed). No outstanding reviewer comments to address.