Repository navigation
[Klaud Cold] Update DSR1 FP8 B300 SGLang image to v0.5.21 / [Klaud Cold] 将 DSR1 FP8 B300 SGLang 镜像更新至 v0.5.21 - #3862
Klaud-Cold wants to merge 2 commits into
Conversation
Pin lmsysorg/sglang:v0.5.21-cu130 by digest and rename the removed cuda-graph-max-bs server argument to cuda-graph-max-bs-decode. 将 dsr1-fp8-b300-sglang 更新至 SGLang v0.5.21-cu130：按 digest 固定 lmsysorg/sglang:v0.5.21-cu130 镜像，并将已移除的 cuda-graph-max-bs 参数改名为 cuda-graph-max-bs-decode。 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Thanks for the contribution!
中文感谢你的贡献！
|
|
Initial attempt · Passed · Run 37895732376 / attempt 1 · 2026-10-09 07:21 UTC
Note: All rows: request errors unavailable.
Note: All rows: Δ N/A: no matched eval baseline. Next: Add the perf-changelog entry and run the final full sweep. 中文初次尝试 · 已通过 · Run 37895732376 / attempt 1 · 2026-10-09 07:21 UTC |
为 dsr1-fp8-b300-sglang 的 SGLang v0.5.21 更新添加 perf-changelog 条目。 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Klaud-Cold
added
the
full-sweep-fail-fast
|
Final full sweep · Passed · Run 37898647160 / attempt 1 · 2026-10-09 07:55 UTC
Note: All rows: request errors unavailable.
Note: All rows: Δ N/A: no matched eval baseline. Note: 1k/1k c1 TP8 EP1 375312, 1k/1k c2 TP8 EP1 cecc47, 1k/1k c4 TP8 EP1 e827e5, 1k/1k c8 TP8 EP1 6b9dae, 1k/1k c16 TP8 EP1 0d4049, 1k/1k c32 TP8 EP1 30a8bd, 1k/1k c64 TP8 EP1 fea825: Single-turn 1k1k retired 2026-07-17 in #2263 after this baseline; excluded from coverage. Next: Mark ready for maintainer review. 中文最终完整 sweep · 已通过 · Run 37898647160 / attempt 1 · 2026-10-09 07:55 UTC |
|
View unofficial run (performance): https://inferencex.semianalysis.com/inference?unofficialRun=37898647160 View unofficial run (accuracy): https://inferencex.semianalysis.com/evaluation?unofficialRun=37898647160 |
|
validated · Repairs: 0 · Runs: 37895732376, 37898616486, 37898647160 中文validated · 修复次数：0 · 运行：37895732376, 37898616486, 37898647160 |
|
/use 37898647160 |
|
@Klaud-Cold staged run 37898647160: https://inferencemax-app-git-staging-semianalysisai.vercel.app/inference?i_dates=2026-10-09~r37898647160 This run remains available across future |
|
Claude finished @Klaud-Cold's task in 12s —— View job LGTM - no blocking issues found |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
Looks good, straightforward recipe/image-bump change. Reviewed: the image bump is digest-pinned and kept identical between the recipe's model.container and master config's image for dsr1-fp8-b300-sglang; the cuda-graph-max-bs to cuda-graph-max-bs-decode rename is applied consistently at both occurrences (TP8 base and TP4 zip-override); and the perf-changelog entry is appended at the tail, English-only, and follows the existing format with no prior bytes modified.
Extended reasoning...
Three-file config change: a single-node SGLang recipe YAML, its matching master-config entry, and an append-only perf-changelog. No security-sensitive surface (auth, crypto, permissions) is touched. Verified via git diff that container/image stay in sync, the flag rename is applied at both sites, and the changelog addition is purely additive at the end of the file with one block for this PR.
This branch has not been deployed
1 participant
Goal: Update SGLang image from
lmsysorg/sglang:v0.5.12-cu130tolmsysorg/sglang:v0.5.21-cu130@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469.Baseline: 2026-05-22 ·
lmsysorg/sglang:v0.5.12-cu130Mean latency · Sources: API 1, API 2
Note: 1k/1k c1 TP8 EP1 375312, 1k/1k c2 TP8 EP1 cecc47, 1k/1k c4 TP8 EP1 e827e5, 1k/1k c8 TP8 EP1 6b9dae, 1k/1k c16 TP8 EP1 0d4049, 1k/1k c32 TP8 EP1 30a8bd, 1k/1k c64 TP8 EP1 fea825: Single-turn 1k1k retired 2026-07-17 in #2263 after this baseline; excluded from coverage.
Eval: N/A
中文
**目标：**将 SGLang 镜像从
lmsysorg/sglang:v0.5.12-cu130更新为lmsysorg/sglang:v0.5.21-cu130@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469。**基线：**2026-05-22 ·
lmsysorg/sglang:v0.5.12-cu130平均延迟 · 来源： API 1, API 2；数值及异常说明见上表。