dsr1-fp8-b300-sglang-mtp: bump SGLang image to v0.5.15.post1-cu130

[hshrivastava-droid]

No description provided.

[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 — straightforward SGLang image version bump with matching changelog entry.

What was reviewed: the dsr1-fp8-b300-sglang-mtp image tag change (v0.5.12-cu130 → v0.5.15.post1-cu130) and its perf-changelog entry — description and pr-link are consistent and correctly reference this PR (#2313).

Extended reasoning...

Overview

This PR is a single-recipe SGLang container image bump for dsr1-fp8-b300-sglang-mtp, from lmsysorg/sglang:v0.5.12-cu130 to lmsysorg/sglang:v0.5.15.post1-cu130, plus the corresponding perf-changelog.yaml entry documenting the change and linking to the upstream SGLang release notes. A stale comment referencing an older v0.5.10.post1-cu130 bump note was also removed, which is appropriate cleanup since it no longer matches the current image tag.

Security risks

None. This only changes a pinned third-party container image tag and adds a changelog record — no code execution paths, auth, or secrets are touched.

Level of scrutiny

Low. This matches the repo's well-established pattern for routine inference-engine image bumps (as also used by the nuke skill), and the change is entirely data (YAML config), not logic. The bug hunting system found no issues, and the one candidate flagged as 'ruled out' (template placeholder pr-link) does not actually apply here — the pr-link field is correctly populated with #2313.

Other factors

No CODEOWNER-sensitive or critical-path code is touched. The PR carries full-sweep-enabled, so the benchmark sweep will validate the new image tag empirically post-merge. This is a one-line functional change with a clean, consistent changelog entry — safe to approve without further human review.

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30029792076
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=30029792076

[Ankur-singh]

/reuse-sweep-run 30029792076

[Ankur-singh]

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

• Verified that as of the moment of typing this, this is the latest version of PR_REVIEW_CHECKLIST.md
• 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. Please link to GitHub Action workflow that shows this. — https://github.com/SemiAnalysisAI/InferenceX/actions/runs/30029792076
• Verified that this PR passes evals. Please link to GitHub Action workflow that shows this. — https://github.com/SemiAnalysisAI/InferenceX/actions/runs/30029792076
• Verified that speculative decoding PRs uses chat templates to align the AL distribution to real world
• For agentic workloads: verified that speculative-decoding configs (EAGLE / MTP / draft models) run with simulated synthetic acceptance, with the acceptance-length value taken from the committed golden AL curve in golden_al_distribution/ for that model, thinking mode, and draft length. A submission may choose any supported draft length, but it may not substitute a different acceptance target.
• Verified that the model architecture isn't changed with benchmark hacks like using --hf-overrides to skipping indexer for every x layers on models that don't natively support this. As a general rule, we won't accept optimizations that reduces the number of model architecture FLOPs. Anything that makes that same computation run faster is fair game; FLOPs at lower precisions is fine, given that the config passes private evals. As an general north star princple, we should only use optimizations which is used in production by customers that care about accuracy
• If an company claims that they support vLLM/SGLang as first class LLM inference engines on their hardware, I have verified that the respective vLLM submission made using upstream https://hub.docker.com/u/vllm docker repo, upstream SGLang https://hub.docker.com/u/lmsysorg docker repo. 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 as supported by vLLM/SGLang community maintainers
• If an company claims that they support vLLM/SGLang as first class upstream in-tree 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 every single-node vLLM/SGLang recipe in this PR is documented in the official vLLM recipes and/or the SGLang cookbook:
  • I linked the corresponding upstream PR in the vLLM recipe repo or SGLang repo and verified that it is MERGED before this InferenceX PR merges. An opened, draft, or closed-without-merge upstream PR does not satisfy this requirement. If the matching recipe was already published, I linked the published recipe/cookbook page in the additional detail section below.
• Verified that this PR does not patch the inference engine or serving stack — the pinned image must run as shipped. This covers .patch files / git apply / patch, inline patches embedded in benchmark scripts (e.g. a python3/sed heredoc that rewrites installed engine sources before serving), in-place edits of site-packages, monkey-patching, overwriting container files, and installing forked/rebuilt engine wheels on top of the pinned image. The only exception is a patch covered by a filled-out waiver at docs/waiver/<PR_NUMBER>.md — named after the PR that introduces the patch and filed in that same PR, stating what is patched, why the unmodified upstream image cannot run this benchmark, the upstream PR/issue link, and the removal plan — which I have linked 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:

• Scope: single-config image bump for dsr1-fp8-b300-sglang-mtp — SGLang container lmsysorg/sglang:v0.5.12-cu130 → lmsysorg/sglang:v0.5.15.post1-cu130 (upstream lmsysorg repo; release notes). Also removes a stale in-config comment about a prior image version. No benchmark-script changes.
• Validation & evals: Run Sweep https://github.com/SemiAnalysisAI/InferenceX/actions/runs/30029792076 (head fe0085c7) is green top-level with non-skipped single-node 8k1k benchmark jobs (c1–c512) and eval jobs (c128, c512) passing for dsr1 fp8 b300 sgl TP8 mtp on the new image; collect-evals/collect-results/compare-results all succeeded.
• Chat template (spec decode): dsr1_fp8_b300_mtp.sh runs EAGLE speculative decoding and passes --use-chat-template in run_benchmark_serving, aligning the AL distribution to real-world chat usage.
• Recipe: documented in the SGLang cookbook — https://cookbook.sglang.io/autoregressive/DeepSeek/DeepSeek-R1 . The B300 config reuses the existing DSR1 FP8 B200 SGLang MTP recipe as-is until B300-specific tuning is available (noted in the config/script comment).
• No engine patching: the script runs python3 -m sglang.launch_server with the pinned image as shipped — no patch files, sed/heredoc rewrites, site-packages edits, or forked wheels.
• Agentic item (unchecked): not applicable — this is a single-node fixed-seq-len (8k1k) benchmark, not an agentic workload; agentic/golden-AL jobs are correctly skipped in the sweep.

Signed: Ankur-singh

[Klaud-Cold]

✅✅✅ Verdict: PASS ✅✅✅

✅ Check 0 (CODEOWNER): PASS — Ankur-singh is a listed owner of configs/nvidia-master.yaml; perf-changelog.yaml falls to the catch-all, covered by any recognized CODEOWNER.
✅ Check 1 (sweep on in-PR commit): PASS — head fe0085c7 carries green, executed single-node 8k1k / (c1–c512) and eval / (c128, c512) check-runs from run 30029792076.
✅ Check 2 (evals pass): PASS — gsm8k em_strict 0.9538 (c128) / 0.9568 (c512), n_eff=1319, above the 0.91 dsr1 bar; run used lmsysorg/sglang:v0.5.15.post1-cu130, the exact image this PR pins.
✅ Check 3 (recipe): PASS — published SGLang cookbook DSR1 page linked; major args match (DeepSeek-R1-0528, TP8, fp8, kv-cache fp8_e4m3, EAGLE MTP, --fp8-gemm-backend flashinfer_trtllm per the B200 FP8 recipe this config documents reusing). This PR changes no server args — only the image tag, which is InferenceX-specific. Informational: --attention-backend trtllm_mla (Blackwell MLA default) and --enable-flashinfer-allreduce-fusion are perf toggles not in the cookbook's non-MTP B300 entry; the B200-recipe reuse is noted in the config comment pending B300-specific cookbook tuning.
✅ Check 4 (reuse command): PASS — /reuse-sweep-run 30029792076 posted by Ankur-singh (COLLABORATOR).
✅ Check 5 (latest template): PASS — all current-template items present and checked; the agentic item is unchecked with an N/A explanation in the additional detail section.
✅ Check 6 (upstream image / engine-first): PASS — lmsysorg/sglang:v0.5.15.post1-cu130 is the upstream lmsysorg repo on established B300; framework is SGLang, so engine-first ordering is moot.
✅ Check 7 (no architecture hacks): PASS — no --hf-overrides or FLOPs-reducing knobs in the diff or launch script.
✅ Check 8 (spec-decode chat template): PASS — dsr1_fp8_b300_mtp.sh benchmarks via run_benchmark_serving ... --use-chat-template.
✅ Check 9 (no engine patches): PASS — pinned image runs as shipped; only client-side pip install datasets pandas (harness deps).
➖ Check 10 (agentic golden AL): N/A — fixed-seq-len 8k1k config, no agentic spec-decode changes.