# Wed Oct 28 03:03:39 PM +03 2026 mode=single server=127.0.0.1:18020
ROW cohort C1 real prompts T=default | e2e=225.03 tok/s | decode(C/meanTPOT)=228.8 | tok/step=3.34 | meanTTFT=83.71 ms
ROW cohort C2 real prompts T=default | e2e=347.20 tok/s | decode(C/meanTPOT)=404.9 | tok/step=3.08 | meanTTFT=118.61 ms
ROW cohort C4 real prompts T=default | e2e=588.57 tok/s | decode(C/meanTPOT)=723.3 | tok/step=3.23 | meanTTFT=137.09 ms
ROW cohort C8 real prompts T=default | e2e=538.21 tok/s | decode(C/meanTPOT)=901.9 | tok/step=3.32 | meanTTFT=1457.91 ms
ROW cohort C1 real prompts T=0 | e2e=224.09 tok/s | decode(C/meanTPOT)=231.5 | tok/step=3.28 | meanTTFT=81.45 ms
ROW cohort C2 real prompts T=0 | e2e=384.69 tok/s | decode(C/meanTPOT)=433.8 | tok/step=3.25 | meanTTFT=119.97 ms
ROW cohort C4 real prompts T=0 | e2e=607.66 tok/s | decode(C/meanTPOT)=735.3 | tok/step=3.24 | meanTTFT=135.86 ms
ROW cohort C8 real prompts T=0 | e2e=612.97 tok/s | decode(C/meanTPOT)=1000.0 | tok/step=3.24 | meanTTFT=814.05 ms
# raw logs in /home/zeka/ai/benchmarks/hyperqwen-5090-2026-10-28/results/single-dflash2/run2
