Bump vLLM version for DSV4 B200 disagg

[RohitNagraj]

Bumps the DeepSeek-V4-Pro B200 disaggregated low-latency vLLM recipe (disagg-b200-low-latency.yaml) from vllm/vllm-openai:v0.20.1 to v0.23.0.

Recreated from #1898 on a branch in this repository. Supersedes and closes #1898.

---

Note

Medium Risk
Config-only, but a multi-minor vLLM image bump on a complex disagg B200 benchmark can change runtime behavior and compatibility with the pinned Dynamo wheel until the job is revalidated.

Overview
Updates the DeepSeek-V4-Pro B200 disaggregated low-latency Slurm recipe (disagg-b200-low-latency.yaml) to run on vLLM 0.23.0 instead of 0.20.x.

model.container and identity.container.image now use vllm/vllm-openai:v0.23.0, and identity.frameworks.vllm is set to 0.23.0 so launch metadata matches the container. Dynamo wheel and all vLLM/serving flags are unchanged.

^{Reviewed by Cursor Bugbot for commit bb6fef2. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thanks for the contribution! For vLLM & SGLang, please ensure that your recipes is similar to the official vLLM recipes and/or the SGLang cookbook

• https://docs.sglang.io/cookbook/intro
• https://recipes.vllm.ai/

If it is not, please create a PR first before we can merge your single node PR into the master branch. Let's ensure that the documentation is first class such that the entire ML community can benefit from your hard work! Thank you

• https://github.com/vllm-project/recipes
• https://github.com/sgl-project/sglang/tree/main/docs_new

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. If re-running failed jobs is attempted, PR authors are responsible for ensuring it passes. See GitHub's docs on re-running failed jobs: https://docs.github.com/en/actions/how-tos/manage-workflow-runs/re-run-workflows-and-jobs#re-running-failed-jobs-in-a-workflow

As a rule of thumb, generally, PR authors should request a review & get a PR approval from the respective companies' CODEOWNERS before requesting a review from core maintainers.

If additional help is needed, PR authors can reach out to core maintainers over Slack.

---

感谢你的贡献！对于 vLLM 与 SGLang，请确保你的 recipe 与官方 vLLM recipes 和/或 SGLang cookbook 保持一致

• https://docs.sglang.io/cookbook/intro
• https://recipes.vllm.ai/

如果不一致，请先创建一个 PR，之后我们才能将你的单节点 PR 合并到 master 分支。让我们确保文档保持一流水准，使整个 ML 社区都能从你的辛勤工作中受益！谢谢

• https://github.com/vllm-project/recipes
• https://github.com/sgl-project/sglang/tree/main/docs_new

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。如果选择重新运行失败的任务，PR 作者有责任确保其最终通过。参见 GitHub 关于重新运行失败任务的文档：https://docs.github.com/en/actions/how-tos/manage-workflow-runs/re-run-workflows-and-jobs#re-running-failed-jobs-in-a-workflow

一般而言，PR 作者应先向相应公司的 CODEOWNERS 请求审阅并获得 PR 批准，然后再请求核心维护者审阅。

如需更多帮助，PR 作者可通过 Slack 联系核心维护者。

[github-actions]

Thanks for the contribution! For vLLM & SGLang, please ensure that your recipes is similar to the official vLLM recipes and/or the SGLang cookbook

• https://docs.sglang.io/cookbook/intro
• https://recipes.vllm.ai/

If it is not, please create a PR first before we can merge your single node PR into the master branch. Let's ensure that the documentation is first class such that the entire ML community can benefit from your hard work! Thank you

• https://github.com/vllm-project/recipes
• https://github.com/sgl-project/sglang/tree/main/docs_new

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. If re-running failed jobs is attempted, PR authors are responsible for ensuring it passes. See GitHub's docs on re-running failed jobs: https://docs.github.com/en/actions/how-tos/manage-workflow-runs/re-run-workflows-and-jobs#re-running-failed-jobs-in-a-workflow

As a rule of thumb, generally, PR authors should request a review & get a PR approval from the respective companies' CODEOWNERS before requesting a review from core maintainers.

If additional help is needed, PR authors can reach out to core maintainers over Slack.

---

感谢你的贡献！对于 vLLM 与 SGLang，请确保你的 recipe 与官方 vLLM recipes 和/或 SGLang cookbook 保持一致

• https://docs.sglang.io/cookbook/intro
• https://recipes.vllm.ai/

如果不一致，请先创建一个 PR，之后我们才能将你的单节点 PR 合并到 master 分支。让我们确保文档保持一流水准，使整个 ML 社区都能从你的辛勤工作中受益！谢谢

• https://github.com/vllm-project/recipes
• https://github.com/sgl-project/sglang/tree/main/docs_new

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。如果选择重新运行失败的任务，PR 作者有责任确保其最终通过。参见 GitHub 关于重新运行失败任务的文档：https://docs.github.com/en/actions/how-tos/manage-workflow-runs/re-run-workflows-and-jobs#re-running-failed-jobs-in-a-workflow

一般而言，PR 作者应先向相应公司的 CODEOWNERS 请求审阅并获得 PR 批准，然后再请求核心维护者审阅。

如需更多帮助，PR 作者可通过 Slack 联系核心维护者。

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=28078051095
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=28078051095

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=28078051095
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=28078051095

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=28078051095
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=28078051095

[RohitNagraj]

/reuse-sweep-run

[functionstackx]

hi @RohitNagraj can u ask an NVIDIA CODEOWNER like @Ankur-singh for an PR review https://github.com/SemiAnalysisAI/InferenceX/blob/main/docs/PR_REVIEW_CHECKLIST.md

[Ankur-singh]

As a PR reviewer and CODEOWNER, I have reviewed this and have:

• Verified that the general code quality meets the InferenceX standard and does not make the code quality any worse.
• Verified that this PR has passed PR validation.
• Verified that this PR passes evals.
• If an company claims that they support vLLM/SGLang as first class LLM inference engines on their hardware, I have have verified that the respective vLLM/SGLang submission has been made before additional frameworks (TRT-LLM, ATOM, etc.). The only exceptions are for new hardware, such as MI455X UALoE72, Vera Rubin NVL72, Rubin NVL8, etc., and for new model architectures where there is an actual reason why vLLM/SGLang does not fundamentally support them yet.
• Verified that the single-node recipes are similar to the official vLLM recipes and/or theSGLang cookbook:
  • If they are not, I have verified that a PR has been opened in vLLM recipe repo or SGLang repo and linked it below in the additional detail section:
• If any of the above criteria cannot reasonably be satisfied, I have provided additional reasoning below.

Additional detail section:

This is a dis-agg config

Signed: ankur-singh

[Klaud-Cold]

@Ankur-singh BLOCKED — Check 3 fails: no recipe link in the sign-off's "Additional detail section" (only "This is a dis-agg config"). A link to a vLLM-recipes/SGLang-cookbook PR or published page is required by the merge standard; please add one (or the upstream commit this is mirrored from) and re-sign.

• Check 0 — PASS: @Ankur-singh is a direct CODEOWNER of .github/configs/nvidia-master.yaml; the benchmarks/** and perf-changelog.yaml changes fall under catch-all * @InferenceX/core and are covered.
• Check 1 — PASS: pinned head 319f7e8 (current PR tip) has green, executed multi-node 8k1k / (3 configs) and multi-node eval / (3 configs). run 28078051095. (single-node */ / eval / are skipped as expected for a multi-node disagg config.)
• Check 2 — PASS: gsm8k em_strict 0.972 / 0.951 / 0.954 across the three configs; eval run image vllm/vllm-openai:v0.23.0 matches this PR's config.
• Check 3 — FAIL: link absent. Server-arg comparison not performed (no recipe to compare against); the recipe yaml only references an internal NVIDIA/srt-slurm branch, not a public recipe. Note: this PR also lowers max-num-batched-tokens 32768->16384 and gpu-memory-utilization 0.95->0.9 on the megamoe recipes (OOM mitigation, per perf-changelog) beyond the image bump.

[cquil11]

LGTM

[Ankur-singh]

/reuse-sweep-run

[Ankur-singh]

/reuse-sweep-run