Update DSR1 B200 FP4 dynamo-sglang MTP disagg sweep points#2113
Update DSR1 B200 FP4 dynamo-sglang MTP disagg sweep points#2113RohitNagraj wants to merge 8 commits into
Conversation
…sglang-mtp-sweep-update # Conflicts: # perf-changelog.yaml
|
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 For PR verification, add the 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 如需进行 PR 验证，请为此 PR 添加 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 |
2 similar comments
|
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 For PR verification, add the 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 如需进行 PR 验证，请为此 PR 添加 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 |
|
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 For PR verification, add the 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 如需进行 PR 验证，请为此 PR 添加 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[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=28885926270 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=28887126043 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=28891078749 |
The 1k1k points referenced the external srt-slurm recipe, which carries no dynamo pin and falls back to an incompatible dynamo release. Restore the in-repo 1k1k recipes (dynamo hash-pinned, use_chat_template for MTP) and repoint the config, matching how the 8k1k points already resolve.
…sglang-mtp-sweep-update # Conflicts: # perf-changelog.yaml
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=28916523710 |
1 participant
Updates the 8k/1k MTP disaggregated sweep for
dsr1-fp4-b200-dynamo-sglang-mtp.Low-latency points (pruned):
1p5d-tp8→ conc 4, 8, 16, 321p3d-tp8→ conc 32, 641p1d-tp8→ conc 32High-throughput points (added):
1p1d(DEP4 prefill / DEP8 decode) @ conc 5124p1d(DEP4 prefill / DEP8 decode) @ conc 5124p1d(DEP4 prefill / DEP8 decode) @ conc 2048 — dedicated tune (max-running-requests 2048, decode scheduler-recv-interval 1, prefill/decode stream-interval 100/34,SGLANG_HACK_SEQ_BOOTSTRAP_ROOM)Adds three new disagg recipe files. The 1k/1k scenario is unchanged.