Matrix A - run 37718273354
       │                                      c1@16k                                      │                                     c16@16k                                      │                                     c64@16k                                      
engine │ prefill tok/s decode/stream  decode total          TTFT          TPOT agg@900 tok/s │ prefill tok/s decode/stream  decode total          TTFT          TPOT agg@900 tok/s │ prefill tok/s decode/stream  decode total          TTFT          TPOT agg@900 tok/s 
─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
arbi   │   69562/67216       392/373     3135/2992       225/236       2.5/2.7       355/339 │   17557/16338      114/92.4     5910/4761       360/355      8.7/10.8     1723/1407 │     1911/1704     62.1/41.4     4040/2756     8520/9183     16.1/24.1     2395/1830 
vllm   │   66518/60963       369/368     2924/2903       244/253       2.7/2.7       332/328 │   17356/16826      120/91.6     6040/4717       444/387      8.3/10.9     1760/1411 │     2111/1809     54.9/39.2     4000/2710     7486/8560     18.2/25.5     2449/1832 
sglang │   59147/56743       415/398     3323/3186       266/283       2.4/2.5       368/353 │   19878/18612      133/97.8     6704/4944       398/306      7.5/10.2     1969/1483 │     2329/2116     60.6/38.3     4086/2564     6708/7372     16.5/26.1     2619/1831 
cells: tkv/bf16 · TTFT,TPOT=ms median · prefill=prompt tok / TTFT · decode/stream=1000/TPOT · decode total=all streams over their decode spans · agg=all streams @ 900 tok, prefill INCLUDED · best / worst>20% / worst-close / plain mid