[on june 22, 2026, current gptoss submiters were given notice that it will be deprepcated on july 6th, and were given an chance for comment period and everyone reacted notice] Deprecate gpt-oss-120b benchmark configs / 弃用 gpt-oss-120b 基准测试配置

[functionstackx]

on june 22, 2026, current gptoss submiters were given notice that it will be deprepcated on july 6th, and were given an chance for comment period and everyone reacted with :ack: and ✅ to the notice

Summary

• Deprecates gpt-oss-120b following the MiniMax M2.5/M2.7 pattern from Deprecate MiniMax M2.5 / MiniMax M2.7 model architecture benchmark configs #1874:
  • All 10 config entries (gptoss-fp4-{b200,h100,h200}-{vllm,trt}, gptoss-fp4-gb200-dynamo-trt, gptoss-fp4-{mi300x,mi325x,mi355x}-vllm, gptoss-fp4-mi355x-atom) removed from the active master configs and archived verbatim (with comments) under configs/deprecated/nvidia-gptoss-master.yaml / configs/deprecated/amd-gptoss-master.yaml, so sweep generation no longer selects them.
  • All 10 scripts moved into deprecated/ subdirectories (9 × single_node/fixed_seq_len/, 1 × multi_node/).
• One improvement over Deprecate MiniMax M2.5 / MiniMax M2.7 model architecture benchmark configs #1874: the moved scripts' benchmark_lib.sh source paths are corrected for the extra directory level — the same git mv staleness that broke the DSV4 MTP scripts in fix: move dsv4-fp4-mi355x MTP scripts into fixed_seq_len/ #1860 (fixed in [Klaud Cold] [AMD] DeepSeek-V4 FP4 MI355X vLLM MTP: bump image to latest nightly / DeepSeek-V4 FP4 MI355X vLLM MTP：升级镜像至最新 nightly #1981/[Klaud Cold] Fix benchmark_lib source paths broken by #1860 script move / 修复 #1860 移动脚本导致的 benchmark_lib 引用路径错误 #2053). Note the M2.5 deprecated scripts still carry stale paths (dormant, archive-only).
• No perf-changelog.yaml entry, matching Deprecate MiniMax M2.5 / MiniMax M2.7 model architecture benchmark configs #1874 (removals don't trigger sweeps).

⚠️ Heads-up: open PR #2051 ("gpt-oss-fp4-mi355x W4A8 MoE optimizations") modifies gptoss-fp4-mi355x-vllm, which this PR removes — merging this deprecation supersedes/conflicts with #2051. Decide which one proceeds.

中文说明

• 按 Deprecate MiniMax M2.5 / MiniMax M2.7 model architecture benchmark configs #1874 的 MiniMax M2.5/M2.7 模式弃用 gpt-oss-120b：10 个配置条目从活动 master 配置移除并原样归档至 configs/deprecated/（含注释），扫描生成不再选中它们；全部 10 个脚本移入 deprecated/ 子目录。
• 相比 Deprecate MiniMax M2.5 / MiniMax M2.7 model architecture benchmark configs #1874 的改进：已修正移动脚本中 benchmark_lib.sh 的相对路径（fix: move dsv4-fp4-mi355x MTP scripts into fixed_seq_len/ #1860 同类问题；M2.5 的归档脚本仍带旧路径，仅归档不运行，属休眠问题）。
• 与 Deprecate MiniMax M2.5 / MiniMax M2.7 model architecture benchmark configs #1874 一致，不添加 perf-changelog.yaml 条目（移除操作不触发基准测试）。
• ⚠️ 注意：开放中的 PR [Klaud Cold] [AMD] gpt-oss-fp4-mi355x (vllm): W4A8 moe optimizations and vllm image bump / gpt-oss-fp4-mi355x（vLLM）：W4A8 MoE 优化与 vLLM 镜像升级 #2051（gpt-oss W4A8 MoE 优化）修改了本 PR 移除的 gptoss-fp4-mi355x-vllm 条目，两者冲突 - 需决定保留哪个。

🤖 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

[claude]

LGTM — mechanical deprecation following the #1874 pattern (config entries archived to configs/deprecated/, scripts moved with correctly-adjusted benchmark_lib.sh paths).

Extended reasoning...

Overview

Deprecates 10 gpt-oss-120b benchmark recipes: 10 config entries removed from active configs/amd-master.yaml and configs/nvidia-master.yaml and archived verbatim under new configs/deprecated/{amd,nvidia}-gptoss-master.yaml files, and the 10 corresponding shell scripts git mv-ed into deprecated/ subdirectories (9 single_node/fixed_seq_len + 1 multi_node). Also strips some blank-line whitespace between neighboring YAML entries.

Security risks

None. This is a config/benchmark-recipe deprecation touching only sweep definitions and their launch scripts. No auth, crypto, permissions, or user-facing surfaces are involved.

Level of scrutiny

Low. The changes are purely mechanical: identical YAML blocks moved between files, and shell scripts moved into a subdirectory with ../../benchmark_lib.sh → ../../../benchmark_lib.sh adjustments to compensate for the extra directory level. Sweep generation only reads configs/{amd,nvidia}-master.yaml (verified via .claude/mcp/mcp_utils.py), so archived entries under configs/deprecated/ won't be selected. This follows the established pattern from #1874 (MiniMax M2.5/M2.7 deprecation) with a small improvement — the relative source paths in the archived scripts are actually correct here, whereas M2.5's archived scripts still carry stale paths (dormant, since they're archive-only).

Other factors

No bugs were flagged by the bug hunting system. No perf-changelog.yaml entry is needed for a deprecation (matches #1874 precedent — removals don't trigger sweeps). The PR description explicitly flags a coordination concern with open PR #2051 (which modifies gptoss-fp4-mi355x-vllm, an entry this PR removes) — that's a merge-ordering call a human can make when picking which to merge first, not a code-correctness issue with this PR. Some historical launcher-dispatch code in runners/launch_*.sh still references gptoss as a supported MODEL_PREFIX, but since nothing in the active configs sets that prefix anymore, the branch is dormant and harmless.