sglang/ GB200 perf

Regressions

Auto-flagged by rolling median absolute deviation (|z| > 2.5, 30-day window).

dsr1-fp4-1k1k-mid-curve·peak_output_tokens_per_s·conc 2,048

+14.1%·z=2.7·detected 7 hours ago

dsr1-fp4-1k1k-mid-curve·median_ttft_ms·conc 2,048

+11.0%·z=8.0·detected 7 hours ago

dsr1-fp4-1k1k-mid-curve·median_e2el_ms·conc 2,048

-14.9%·z=-3.6·detected 7 hours ago

dsr1-fp4-1k1k-mid-curve·mean_e2el_ms·conc 2,048

-13.5%·z=-2.9·detected 7 hours ago

dsr1-fp4-1k1k-mid-curve·duration·conc 2,048

-12.5%·z=-2.7·detected 7 hours ago

dsr1-fp4-1k1k-mid-curve·std_ttft_ms·conc 4,096

+13.9%·z=3.5·detected 7 hours ago

dsr1-fp4-1k1k-mid-curve·std_e2el_ms·conc 4,096

-10.0%·z=-3.2·detected 7 hours ago

dsr1-fp4-1k1k-mid-curve·median_ttft_ms·conc 4,096

+83.9%·z=4.9·detected 7 hours ago

dsr1-fp4-1k1k-mid-curve·mean_ttft_ms·conc 4,096

+73.5%·z=4.5·detected 7 hours ago

dsr1-fp8-1k1k-max-tpt·std_ttft_ms·conc 1,024

+15.7%·z=3.1·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·median_e2el_ms·conc 1,024

-5.3%·z=-12.1·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·mean_e2el_ms·conc 1,024

-4.4%·z=-10.4·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·duration·conc 1,024

-4.3%·z=-12.5·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·peak_output_tokens_per_s·conc 4,096

+7.3%·z=7.3·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·median_e2el_ms·conc 4,096

-5.5%·z=-48.9·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·mean_e2el_ms·conc 4,096

-5.1%·z=-23.1·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·duration·conc 4,096

-4.8%·z=-19.7·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·median_e2el_ms·conc 2,048

-5.8%·z=-61.7·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·mean_e2el_ms·conc 2,048

-5.3%·z=-44.9·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·duration·conc 2,048

-5.0%·z=-55.8·detected 8 hours ago

dsr1-fp8-1k1k-ultra-tpt·p99_e2el_ms·conc 4,096

-1.8%·z=-4.7·detected 8 hours ago

dsr1-fp8-1k1k-ultra-tpt·median_e2el_ms·conc 4,096

-1.5%·z=-5.4·detected 8 hours ago

dsr1-fp8-1k1k-ultra-tpt·mean_e2el_ms·conc 4,096

-1.5%·z=-6.5·detected 8 hours ago

dsr1-fp8-1k1k-ultra-tpt·duration·conc 4,096

-1.4%·z=-6.2·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·std_ttft_ms·conc 1,024

+11.6%·z=3.3·detected 1 day ago

dsr1-fp8-1k1k-max-tpt·duration·conc 2,048

+0.2%·z=7.5·detected 2 days ago

dsr1-fp8-1k1k-max-tpt·std_ttft_ms·conc 1,024

+19.9%·z=7.2·detected 2 days ago

dsr1-fp8-1k1k-max-tpt·duration·conc 2,048

+0.3%·z=12.4·detected 3 days ago

dsr1-fp8-1k1k-max-tpt·std_ttft_ms·conc 1,024

+10.4%·z=5.2·detected 3 days ago

dsr1-fp8-1k1k-max-tpt·std_ttft_ms·conc 1,024

+6.5%·z=4.7·detected 4 days ago

dsr1-fp8-1k1k-max-tpt·std_ttft_ms·conc 1,024

+3.8%·z=4.5·detected 5 days ago

dsr1-fp4-1k1k-mid-curve·median_e2el_ms·conc 4,096

+8.8%·z=3.0·detected 7 hours ago

dsr1-fp8-1k1k-max-tpt·p99_e2el_ms·conc 2,048

-1.3%·z=-3.3·detected 8 hours ago

dsr1-fp8-1k1k-max-tpt·duration·conc 2,048

+0.1%·z=3.5·detected 1 day ago