Repository navigation
[AgentX] MiniMax-M3.1-Flash B300 SGLang: add TP2 shortest-prefill-first points / [AgentX] MiniMax-M3.1-Flash B300 SGLang:新增 shortest-prefill-first 的 TP2 点 - #4647
Conversation
|
Thanks for the contribution!
中文感谢你的贡献!
|
There was a problem hiding this comment.
I reviewed this PR and found no bugs. The new B300 recipe and the nvidia-master.yaml rows have code owners, so one of them should still take a look before merge.
What was reviewed:
- The zipped override lists line up: 5 entries each for max-running-requests, cuda-graph-max-bs-decode and CONC. Each nvidia-master row's index and conc-list match its CONC.
- The interval-32 override matches the zipped [2] point except for prefill-decode-interval. The existing agentic-tp2.yaml row is unchanged.
- The changelog block is appended at the tail. Earlier entries already use the
pull/XXXplaceholder andappend-only: true, and that fits a change that only adds new points to an existing curve.
Extended reasoning...
The PR adds one new single-node srt-slurm recipe for MiniMax-M3.1-Flash FP4 on B300 SGLang, six search-space rows under an existing nvidia-master key, and one perf-changelog block appended at the tail. None of this is security-sensitive: it is benchmark configuration with no embedded shell. The recipe does not set image, model, precision, sequence lengths or benchmark type, and it does not hard-code acceptance lengths. CODEOWNERS assigns the NVIDIA master config and the b300-* recipe paths to specific owners, and that alone ruled out an automated approval.
This review covers commit de75f0c, which is no longer the latest commit on this pull request; later commits are not covered by it.
|
View unofficial run (performance): https://inferencex.semianalysis.com/inference?unofficialRun=40895671658 View unofficial run (accuracy): https://inferencex.semianalysis.com/evaluation?unofficialRun=40895671658 |
|
/use 40895671658 |
|
@ispobock staged run 40895671658. This run remains available across future |
…sglang-tp2-spf # Conflicts: # inferencex-e2e/perf-changelog.yaml
Resolve perf-changelog.yaml by taking main's file and re-appending this PR's entry. 中文:合并 origin/main;perf-changelog.yaml 采用 main 版本,并在末尾重新追加本 PR 的条目。
|
Ingest report — ✅ Results ingested into the Ingest workflow · Merge Ingest run · Benchmark source run
benchmarkFiles 6, rawBenchmarkRows 6, coveredByPerJobFiles 0, purgedPoints 0, failedRunRows 0, runStatsNew 0, runStatsDuplicate 0, evalSamplesNew 0, gpuMetricSamplesNew 78796, powerPublicationPoints 6, changelogEntriesWritten 1 Counts cover only this run. Rows: benchmark_results, eval_results, agentic rows linked to a trace, gpu_metrics series. This comment updates if the ingest is rerun. 中文Ingest 报告 — ✅ 结果已写入 上表仅统计本次运行写入的数据,各行单位依次为 benchmark_results 行、eval_results 行、已关联 trace 的 agentic 行和 gpu_metrics 序列。重新运行 ingest 时本评论会原地更新。 |
Resolve the perf-changelog conflict with main's new entries (#4647, #4627, #4596, #4526, #4664) using prepare_perf_changelog_merge resolve-conflict: main's bytes are kept and this PR's entry is re-appended unchanged. The recipe, the master entry and the changelog entry are byte-identical to the swept commit dd7a1f1 (run 40782441830, pinned by /reuse-sweep-run). 用 prepare_perf_changelog_merge resolve-conflict 解决与 main 新条目 (#4647、#4627、#4596、#4526、#4664)的 perf-changelog 冲突:保留 main 的原始字节,本 PR 的条目原样追加在末尾。recipe、master 条目和 changelog 条目与 sweep 所用提交 dd7a1f1(run 40782441830,已由 /reuse-sweep-run 固定)逐字节一致。 [skip-sweep] Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- perf-changelog: resolve the conflict with main's new entries (#4524, #4599, #4647, #4627, #4596, #4526, #4664) using prepare_perf_changelog_merge resolve-conflict; main's bytes are kept and this PR's entry is re-appended unchanged. - amd-master.yaml: #4664 deleted dsv4-fp4-mi355x-sglang-agentic-mtp right after this PR's glm5.2-fp4-mi355x-atom-disagg-agentic-mtp block. Keep the GLM-5.2 block byte-for-byte and take main's deletion. - main's srt-slurm pin moves to v2.65.2 (#4601); 573 and 604-atom-tp-from-pp.patch still apply in order on 59920ea1. All six files this PR changes are byte-identical to the previous head 1a2707f. - perf-changelog:用 prepare_perf_changelog_merge resolve-conflict 解决与 main 新条目(#4524、#4599、#4647、#4627、#4596、#4526、#4664)的冲突; 保留 main 的原始字节,本 PR 的条目原样追加在末尾。 - amd-master.yaml:#4664 删除了紧跟在本 PR 的 glm5.2-fp4-mi355x-atom-disagg-agentic-mtp 之后的 dsv4-fp4-mi355x-sglang-agentic-mtp。GLM-5.2 条目逐字节保留,同时采纳 main 的删除。 - main 的 srt-slurm 固定版本升到 v2.65.2(#4601);573 和 604-atom-tp-from-pp.patch 仍能在 59920ea1 上按顺序打上。 本 PR 修改的六个文件与上一个 head 1a2707f 逐字节一致。 [skip-sweep] Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add TP2 points with DRAM KV offload (L2 HiCache) at conc 40 to 56 to the existing B300 SGLang AgentX curve, in a new
agentic-tp2-spf.yaml. They use--schedule-policy shortest-prefill-firstwith--prefill-decode-interval 16, 8192-token prefill chunks,--mem-fraction-static 0.92and--stream-interval 20; a second conc 48 point uses an interval of 32. Existing points are unchanged.The serving flags are in the SGLang cookbook: sgl-project/sglang#47303
中文
在现有 B300 SGLang AgentX 曲线上新增 conc 40 到 56 的 TP2 点(DRAM KV offload,L2 HiCache),配置放在新的
agentic-tp2-spf.yaml中。这些点使用--schedule-policy shortest-prefill-first和--prefill-decode-interval 16,prefill chunk 为 8192 token,--mem-fraction-static 0.92,--stream-interval 20;另有一个 conc 48 点的 interval 为 32。现有点不变。服务参数已同步到 SGLang cookbook:sgl-project/sglang#47303
AI model disclosure
Prepared with Claude Code using
claude-opus-5-5. Reviewed by the human author before submission.