perf-changelog: re-run DSR1 SGLang agg configs (B200/B300, FP8/FP4, no-MTP/MTP)#1502
Conversation
Re-runs DSR1 SGLang agg configs on B200/B300 (FP8/FP4, no-MTP/MTP) to pick up the tokenizer fix from #1381.
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=26049882389 |
- runners/launch_b300-nv.sh: remove --nodelist=b300-[001-006,008-012,017-020] from salloc so jobs can land on any healthy B300 node. - perf-changelog.yaml: restore ~18 entries that were unintentionally dropped during a prior rebase; net effect of this branch is now just the new DSR1 SGLang agg re-run entry.
… into rerun-dsr1-sglang-agg-b200-b300 # Conflicts: # perf-changelog.yaml
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=26129774152 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=26129885146 |
|
/reuse-sweep-run |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=26143743050 |
* dsr1: extend sglang agg sweeps to conc 1,2 (B200/B300, FP8/FP4, no-MTP/MTP) Lowers conc-start from 4 to 1 in all six DSR1 SGLang agg search-spaces re-run by #1502: dsr1-fp{4,8}-b{200,300}-sglang and dsr1-fp8-b{200,300}-sglang-mtp. With step-factor 2 this adds conc=1 and conc=2 data points; conc-end is unchanged. * perf-changelog: extend DSR1 SGLang agg sweeps to conc 1,2 * perf-changelog: set PR link to #1534 * dsr1-sglang: truncate sweep to conc=1 and conc=2 only Sets conc-start=1, conc-end=2 in every search-space across the six DSR1 SGLang agg configs (B200/B300, FP8/FP4, no-MTP/MTP). Also updates the perf-changelog entry to match.
2 participants
Summary
Re-runs DSR1 SGLang agg configs on B200/B300 (FP8/FP4, no-MTP/MTP) to pick up the tokenizer fix from #1381.