framework,version,device,op_name,kernel_source,batch_size,isl,num_heads,num_key_value_heads,head_dim,beam_width,attn_dtype,kv_cache_dtype,step,latency
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,96,1,128,1,float16,float16,0,120.8553466796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,96,2,128,1,float16,float16,0,121.32850341796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,96,4,128,1,float16,float16,0,125.4003173828125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,96,96,128,1,float16,float16,0,132.56558837890626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,96,96,128,1,float16,float16,0,66.06190795898438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,96,1,128,1,float16,float16,0,61.54518432617188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,96,2,128,1,float16,float16,0,60.88299560546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,96,2,128,1,float16,float16,0,243.1684326171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,96,1,128,1,float16,float16,0,244.7326904296875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,96,8,128,1,float16,float16,0,124.719384765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,96,96,128,1,float16,float16,0,33.229391479492186
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,96,4,128,1,float16,float16,0,61.98819580078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,96,4,128,1,float16,float16,0,250.304345703125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,96,8,128,1,float16,float16,0,63.41259765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,96,8,128,1,float16,float16,0,253.7759765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,96,1,128,1,float16,float16,0,30.07520751953125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,96,2,128,1,float16,float16,0,30.563156127929688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,96,4,128,1,float16,float16,0,30.901953125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,96,8,128,1,float16,float16,0,31.02864990234375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,96,1,128,1,float16,float16,0,69.85819091796876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,96,2,128,1,float16,float16,0,69.70242919921876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,96,4,128,1,float16,float16,0,70.82183837890625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,96,96,128,1,float16,float16,0,78.16509399414062
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,96,96,128,1,float16,float16,0,39.207400512695315
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,96,1,128,1,float16,float16,0,34.6943603515625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,96,2,128,1,float16,float16,0,34.888214111328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,96,1,128,1,float16,float16,0,139.621630859375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,96,2,128,1,float16,float16,0,142.153369140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,96,4,128,1,float16,float16,0,144.683251953125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,96,96,128,1,float16,float16,0,19.246611022949217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,96,8,128,1,float16,float16,0,146.34327392578126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,96,1,128,1,float16,float16,0,17.537326049804687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,96,8,128,1,float16,float16,0,72.12713623046875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,96,2,128,1,float16,float16,0,17.726931762695312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,96,4,128,1,float16,float16,0,17.970872497558595
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,96,4,128,1,float16,float16,0,35.342462158203126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,96,8,128,1,float16,float16,0,35.60911865234375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,96,8,128,1,float16,float16,0,17.577046203613282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,96,1,128,1,float16,float16,0,48.97829895019531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,96,2,128,1,float16,float16,0,49.209588623046876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,96,4,128,1,float16,float16,0,50.07047119140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,96,96,128,1,float16,float16,0,57.11148681640625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,96,96,128,1,float16,float16,0,27.530538940429686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,96,1,128,1,float16,float16,0,24.321963500976562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,96,2,128,1,float16,float16,0,24.926010131835938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,96,2,128,1,float16,float16,0,98.81826782226562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,96,1,128,1,float16,float16,0,100.402392578125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,96,4,128,1,float16,float16,0,101.22146606445312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,96,96,128,1,float16,float16,0,14.12151641845703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,96,8,128,1,float16,float16,0,102.647412109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,96,8,128,1,float16,float16,0,50.45530700683594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,96,1,128,1,float16,float16,0,12.144519805908203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,96,2,128,1,float16,float16,0,12.455429077148438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,96,4,128,1,float16,float16,0,24.957537841796874
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,96,8,128,1,float16,float16,0,25.27613220214844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,96,4,128,1,float16,float16,0,12.574759674072265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,96,8,128,1,float16,float16,0,12.525459289550781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,96,1,128,1,float16,float16,0,63.73524169921875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,96,2,128,1,float16,float16,0,64.7322509765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,96,4,128,1,float16,float16,0,66.57736206054688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,96,96,128,1,float16,float16,0,73.85716552734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,96,1,128,1,float16,float16,0,32.526486206054685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,96,2,128,1,float16,float16,0,32.02421264648437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,96,96,128,1,float16,float16,0,37.963531494140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,96,2,128,1,float16,float16,0,128.1260986328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,96,1,128,1,float16,float16,0,129.99947509765624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,96,96,128,1,float16,float16,0,18.750700378417967
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,96,4,128,1,float16,float16,0,131.06192626953126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,96,8,128,1,float16,float16,0,65.84114379882813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,96,8,128,1,float16,float16,0,135.0112060546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,96,8,128,1,float16,float16,0,32.688409423828126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,96,4,128,1,float16,float16,0,33.142236328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,96,96,128,1,float16,float16,0,8.930281829833984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,96,1,128,1,float16,float16,0,15.827809143066407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,96,2,128,1,float16,float16,0,16.324905395507812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,96,1,128,1,float16,float16,0,8.160662078857422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,96,4,128,1,float16,float16,0,16.534552001953124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,96,8,128,1,float16,float16,0,16.526687622070312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,96,2,128,1,float16,float16,0,7.9701484680175785
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,96,4,128,1,float16,float16,0,8.297859191894531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,96,8,128,1,float16,float16,0,8.072283172607422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,96,1,128,1,float16,float16,0,36.60723876953125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,96,2,128,1,float16,float16,0,37.801596069335936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,96,4,128,1,float16,float16,0,37.7826904296875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,96,96,128,1,float16,float16,0,43.698263549804686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,96,1,128,1,float16,float16,0,18.162681579589844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,96,96,128,1,float16,float16,0,19.903482055664064
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,96,2,128,1,float16,float16,0,18.046420288085937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,96,1,128,1,float16,float16,0,73.77433471679687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,96,2,128,1,float16,float16,0,75.50941162109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,96,96,128,1,float16,float16,0,10.248678588867188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,96,4,128,1,float16,float16,0,77.593408203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,96,8,128,1,float16,float16,0,77.2220703125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,96,8,128,1,float16,float16,0,37.917117309570315
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,96,4,128,1,float16,float16,0,18.938270568847656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,96,1,128,1,float16,float16,0,8.880294036865234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,96,2,128,1,float16,float16,0,9.239046478271485
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,96,8,128,1,float16,float16,0,19.77912292480469
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,96,1,128,1,float16,float16,0,4.52379035949707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,96,96,128,1,float16,float16,0,5.359190368652344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,96,2,128,1,float16,float16,0,4.3374481201171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,96,4,128,1,float16,float16,0,4.443454360961914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,96,4,128,1,float16,float16,0,9.266342163085938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,96,8,128,1,float16,float16,0,9.286577606201172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,96,8,128,1,float16,float16,0,4.651391983032227
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,96,1,128,1,float16,float16,0,35.002926635742185
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,96,2,128,1,float16,float16,0,34.790887451171876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,96,96,128,1,float16,float16,0,42.24431457519531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,96,4,128,1,float16,float16,0,34.00852661132812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,96,96,128,1,float16,float16,0,19.820480346679688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,96,1,128,1,float16,float16,0,17.055441284179686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,96,2,128,1,float16,float16,0,17.73763427734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,96,2,128,1,float16,float16,0,68.31886596679688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,96,1,128,1,float16,float16,0,69.08373413085937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,96,4,128,1,float16,float16,0,69.91966552734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,96,8,128,1,float16,float16,0,71.01826171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,96,96,128,1,float16,float16,0,10.098411560058594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,96,8,128,1,float16,float16,0,36.51026611328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,96,1,128,1,float16,float16,0,8.473462677001953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,96,2,128,1,float16,float16,0,8.618643188476563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,96,1,128,1,float16,float16,0,4.1841888427734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,96,4,128,1,float16,float16,0,18.078182983398438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,96,96,128,1,float16,float16,0,5.063814544677735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,96,4,128,1,float16,float16,0,8.45050277709961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,96,8,128,1,float16,float16,0,17.647189331054687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,96,1,128,1,float16,float16,0,1.8405567169189454
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,96,2,128,1,float16,float16,0,3.956337738037109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,96,96,128,1,float16,float16,0,2.460887908935547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,96,8,128,1,float16,float16,0,8.63968505859375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,96,4,128,1,float16,float16,0,4.0630958557128904
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,96,2,128,1,float16,float16,0,1.9134000778198241
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,96,4,128,1,float16,float16,0,2.0538656234741213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,96,8,128,1,float16,float16,0,4.155297470092774
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,96,8,128,1,float16,float16,0,1.9503040313720703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,96,1,128,1,float16,float16,0,19.850804138183594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,96,2,128,1,float16,float16,0,20.501057434082032
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,96,4,128,1,float16,float16,0,20.59107208251953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,96,96,128,1,float16,float16,0,25.330482482910156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,96,1,128,1,float16,float16,0,9.663581085205077
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,96,96,128,1,float16,float16,0,13.035874938964843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,96,2,128,1,float16,float16,0,10.024364471435547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,96,1,128,1,float16,float16,0,39.73389587402344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,96,2,128,1,float16,float16,0,40.73548583984375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,96,4,128,1,float16,float16,0,40.67567443847656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,96,8,128,1,float16,float16,0,20.869972229003906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,96,4,128,1,float16,float16,0,10.017334747314454
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,96,8,128,1,float16,float16,0,42.61497192382812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,96,96,128,1,float16,float16,0,6.069254302978516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,96,1,128,1,float16,float16,0,4.687017440795898
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,96,2,128,1,float16,float16,0,4.770753479003906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,96,1,128,1,float16,float16,0,2.1675392150878907
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,96,4,128,1,float16,float16,0,4.72740478515625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,96,96,128,1,float16,float16,0,3.049752044677734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,96,8,128,1,float16,float16,0,10.296720123291015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,96,2,128,1,float16,float16,0,2.114718437194824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,96,8,128,1,float16,float16,0,4.824524688720703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,96,2,128,1,float16,float16,0,1.0205023765563965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,96,4,128,1,float16,float16,0,2.2292207717895507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,96,1,128,1,float16,float16,0,1.123579216003418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,96,4,128,1,float16,float16,0,1.1783344268798828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,96,96,128,1,float16,float16,0,1.5521519660949707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,96,8,128,1,float16,float16,0,1.1886992454528809
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,96,8,128,1,float16,float16,0,2.326510429382324
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,96,1,128,1,float16,float16,0,18.59947967529297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,96,2,128,1,float16,float16,0,18.57896728515625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,96,4,128,1,float16,float16,0,19.396949768066406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,96,96,128,1,float16,float16,0,26.476028442382812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,96,1,128,1,float16,float16,0,9.641529846191407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,96,96,128,1,float16,float16,0,13.005728149414063
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,96,1,128,1,float16,float16,0,37.45081787109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,96,2,128,1,float16,float16,0,9.265337371826172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,96,2,128,1,float16,float16,0,37.590902709960936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,96,4,128,1,float16,float16,0,38.80931091308594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,96,8,128,1,float16,float16,0,39.56572265625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,96,8,128,1,float16,float16,0,20.53339080810547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,96,4,128,1,float16,float16,0,9.832563018798828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,96,1,128,1,float16,float16,0,4.3043567657470705
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,96,2,128,1,float16,float16,0,4.397179031372071
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,96,1,128,1,float16,float16,0,2.0408416748046876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,96,8,128,1,float16,float16,0,9.610384368896485
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,96,96,128,1,float16,float16,0,3.2160289764404295
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,96,4,128,1,float16,float16,0,4.437611389160156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,96,96,128,1,float16,float16,0,6.337388610839843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,96,2,128,1,float16,float16,0,2.134806442260742
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,96,4,128,1,float16,float16,0,2.0518720626831053
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,96,8,128,1,float16,float16,0,4.671284866333008
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,96,1,128,1,float16,float16,0,0.9685664176940918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,96,96,128,1,float16,float16,0,1.6589040756225586
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,96,8,128,1,float16,float16,0,2.159449577331543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,96,4,128,1,float16,float16,0,0.9740287780761718
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,96,2,128,1,float16,float16,0,1.1002016067504883
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,96,4,128,1,float16,float16,0,0.5200975894927978
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,96,8,128,1,float16,float16,0,1.1210432052612305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,96,1,128,1,float16,float16,0,0.5596704006195068
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,96,96,128,1,float16,float16,0,0.8697135925292969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,96,2,128,1,float16,float16,0,0.5187488079071045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,96,8,128,1,float16,float16,0,0.5975008010864258
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,96,1,128,1,float16,float16,0,10.182746887207031
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,96,2,128,1,float16,float16,0,10.814415740966798
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,96,4,128,1,float16,float16,0,11.189593505859374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,96,96,128,1,float16,float16,0,16.446070861816406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,96,2,128,1,float16,float16,0,21.763998413085936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,96,1,128,1,float16,float16,0,22.238442993164064
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,96,8,128,1,float16,float16,0,11.485414123535156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,96,4,128,1,float16,float16,0,23.286579895019532
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,96,1,128,1,float16,float16,0,4.995624160766601
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,96,96,128,1,float16,float16,0,8.245038604736328
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,96,8,128,1,float16,float16,0,24.171566772460938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,96,1,128,1,float16,float16,0,2.4650384902954103
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,96,2,128,1,float16,float16,0,2.3208992004394533
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,96,2,128,1,float16,float16,0,5.221660614013672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,96,4,128,1,float16,float16,0,2.8145168304443358
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,96,96,128,1,float16,float16,0,4.14958381652832
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,96,4,128,1,float16,float16,0,5.663111877441406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,96,1,128,1,float16,float16,0,1.23614559173584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,96,8,128,1,float16,float16,0,5.8309070587158205
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,96,8,128,1,float16,float16,0,2.5129648208618165
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,96,96,128,1,float16,float16,0,2.3002256393432616
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,96,2,128,1,float16,float16,0,1.2175647735595703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,96,1,128,1,float16,float16,0,0.634819221496582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,96,4,128,1,float16,float16,0,1.2334223747253419
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,96,8,128,1,float16,float16,0,1.1327327728271483
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,96,96,128,1,float16,float16,0,1.150857639312744
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,96,1,128,1,float16,float16,0,0.316593599319458
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,96,96,128,1,float16,float16,0,0.5535151958465576
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,96,8,128,1,float16,float16,0,0.6014224052429199
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,96,4,128,1,float16,float16,0,0.714031982421875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,96,2,128,1,float16,float16,0,0.6983535766601563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,96,8,128,1,float16,float16,0,0.33071041107177734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,96,2,128,1,float16,float16,0,0.3598543882369995
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,96,4,128,1,float16,float16,0,0.35635039806365965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,96,1,128,1,float16,float16,0,10.517393493652344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,96,2,128,1,float16,float16,0,10.414952087402344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,96,4,128,1,float16,float16,0,10.921198272705078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,96,1,128,1,float16,float16,0,5.041035079956055
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,96,96,128,1,float16,float16,0,19.806956481933593
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,96,1,128,1,float16,float16,0,21.6152587890625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,96,2,128,1,float16,float16,0,21.801275634765624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,96,96,128,1,float16,float16,0,10.055387115478515
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,96,2,128,1,float16,float16,0,5.01592788696289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,96,8,128,1,float16,float16,0,11.135517120361328
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,96,4,128,1,float16,float16,0,22.66401062011719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,96,8,128,1,float16,float16,0,22.949208068847657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,96,1,128,1,float16,float16,0,2.1327184677124023
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,96,2,128,1,float16,float16,0,2.334529685974121
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,96,4,128,1,float16,float16,0,2.3901567459106445
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,96,8,128,1,float16,float16,0,2.7536144256591797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,96,96,128,1,float16,float16,0,2.688385581970215
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,96,1,128,1,float16,float16,0,1.1529024124145508
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,96,4,128,1,float16,float16,0,4.843360137939453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,96,2,128,1,float16,float16,0,1.1813039779663086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,96,4,128,1,float16,float16,0,1.276473617553711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,96,8,128,1,float16,float16,0,5.640998458862304
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,96,96,128,1,float16,float16,0,5.234723281860352
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,96,1,128,1,float16,float16,0,0.5780416011810303
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,96,4,128,1,float16,float16,0,0.6025599956512451
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,96,8,128,1,float16,float16,0,1.280628776550293
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,96,2,128,1,float16,float16,0,0.6538271903991699
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,96,8,128,1,float16,float16,0,0.6593247890472412
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,96,96,128,1,float16,float16,0,1.378659152984619
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,96,2,128,1,float16,float16,0,0.3067008018493652
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,96,4,128,1,float16,float16,0,0.31227359771728513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,96,1,128,1,float16,float16,0,0.29787840843200686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,96,1,128,1,float16,float16,0,0.17574720382690429
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,96,8,128,1,float16,float16,0,0.36555359363555906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,96,4,128,1,float16,float16,0,0.1667215943336487
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,96,8,128,1,float16,float16,0,0.16285279989242554
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,96,96,128,1,float16,float16,0,0.6897264003753663
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,96,2,128,1,float16,float16,0,0.15682560205459595
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,96,96,128,1,float16,float16,0,0.32073919773101806
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,96,1,128,1,float16,float16,0,6.410304260253906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,96,2,128,1,float16,float16,0,6.496669006347656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,96,4,128,1,float16,float16,0,6.0543663024902346
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,96,1,128,1,float16,float16,0,13.551724243164063
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,96,2,128,1,float16,float16,0,13.861805725097657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,96,4,128,1,float16,float16,0,13.731126403808593
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,96,1,128,1,float16,float16,0,3.256243133544922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,96,8,128,1,float16,float16,0,14.658709716796874
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,96,8,128,1,float16,float16,0,7.307777404785156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,96,1,128,1,float16,float16,0,1.4124159812927246
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,96,2,128,1,float16,float16,0,1.5338255882263183
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,96,2,128,1,float16,float16,0,3.037552070617676
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,96,8,128,1,float16,float16,0,3.2205646514892576
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,96,96,128,1,float16,float16,0,9.908280181884766
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,96,4,128,1,float16,float16,0,3.3402801513671876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,96,4,128,1,float16,float16,0,1.4719552040100097
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,96,8,128,1,float16,float16,0,1.5449695587158203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,96,1,128,1,float16,float16,0,0.7855775833129883
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,96,4,128,1,float16,float16,0,0.9036383628845215
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,96,2,128,1,float16,float16,0,0.7076191902160645
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,96,96,128,1,float16,float16,0,4.95807991027832
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,96,1,128,1,float16,float16,0,0.37712318897247316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,96,2,128,1,float16,float16,0,0.39378719329833983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,96,8,128,1,float16,float16,0,0.7930191993713379
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,96,4,128,1,float16,float16,0,0.3792959928512573
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,96,96,128,1,float16,float16,0,19.556733703613283
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,96,8,128,1,float16,float16,0,0.40181918144226075
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,96,1,128,1,float16,float16,0,0.19317760467529296
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,96,96,128,1,float16,float16,0,2.656270408630371
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,96,96,128,1,float16,float16,0,1.2829855918884276
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,96,96,128,1,float16,float16,0,0.6595263957977295
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,96,2,128,1,float16,float16,0,0.21283841133117676
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,96,1,128,1,float16,float16,0,0.0947376012802124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,96,96,128,1,float16,float16,0,0.24993600845336914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,96,2,128,1,float16,float16,0,0.09423519968986512
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,96,96,128,1,float16,float16,0,0.06757919788360596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,96,2,128,1,float16,float16,0,0.052407997846603396
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,96,8,128,1,float16,float16,0,0.21392478942871093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,96,4,128,1,float16,float16,0,0.0545199990272522
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,96,8,128,1,float16,float16,0,0.10271040201187134
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,96,1,128,1,float16,float16,0,0.05124160051345825
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,96,8,128,1,float16,float16,0,0.05591840147972107
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,96,4,128,1,float16,float16,0,0.20335841178894043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,96,4,128,1,float16,float16,0,0.0940671980381012
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,96,2,128,1,float16,float16,0,5.531777572631836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,96,4,128,1,float16,float16,0,5.724193572998047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,96,1,128,1,float16,float16,0,5.692480087280273
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,96,2,128,1,float16,float16,0,2.7577680587768554
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,96,8,128,1,float16,float16,0,6.369867324829102
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,96,1,128,1,float16,float16,0,1.282420825958252
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,96,2,128,1,float16,float16,0,1.4862159729003905
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,96,4,128,1,float16,float16,0,3.0465423583984377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,96,1,128,1,float16,float16,0,2.590059280395508
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,96,8,128,1,float16,float16,0,3.2374862670898437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,96,4,128,1,float16,float16,0,1.4266240119934082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,96,96,128,1,float16,float16,0,9.841022491455078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,96,1,128,1,float16,float16,0,0.6776080131530762
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,96,2,128,1,float16,float16,0,0.7310688018798828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,96,1,128,1,float16,float16,0,0.3092223882675171
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,96,4,128,1,float16,float16,0,0.7670639991760254
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,96,8,128,1,float16,float16,0,1.6594768524169923
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,96,8,128,1,float16,float16,0,0.8002304077148438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,96,2,128,1,float16,float16,0,0.33285601139068605
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,96,1,128,1,float16,float16,0,0.1230512022972107
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,96,4,128,1,float16,float16,0,0.33511838912963865
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,96,96,128,1,float16,float16,0,5.024803161621094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,96,8,128,1,float16,float16,0,0.41346240043640137
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,96,4,128,1,float16,float16,0,0.141102397441864
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,96,2,128,1,float16,float16,0,0.13441920280456543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,96,96,128,1,float16,float16,0,0.6429840087890625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,96,8,128,1,float16,float16,0,0.16556479930877685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,96,96,128,1,float16,float16,0,2.505999946594238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,96,2,128,1,float16,float16,0,0.06501280069351197
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,96,4,128,1,float16,float16,0,0.06246560215950012
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,96,96,128,1,float16,float16,0,0.046367999911308286
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,96,96,128,1,float16,float16,0,1.3528559684753418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,96,96,128,1,float16,float16,0,0.2347248077392578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,96,2,128,1,float16,float16,0,0.03848640024662018
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,96,8,128,1,float16,float16,0,0.06382399797439575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,96,4,128,1,float16,float16,0,0.037571200728416444
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,96,1,128,1,float16,float16,0,0.0616703987121582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,96,8,128,1,float16,float16,0,0.03692319989204407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,96,1,128,1,float16,float16,0,0.03551360070705414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,96,4,128,1,float16,float16,0,0.024391999840736388
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,96,96,128,1,float16,float16,0,0.029416000843048094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,96,1,128,1,float16,float16,0,2.6347455978393555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,96,2,128,1,float16,float16,0,2.712486457824707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,96,4,128,1,float16,float16,0,2.948252868652344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,96,8,128,1,float16,float16,0,3.2300865173339846
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,96,8,128,1,float16,float16,0,0.02481440007686615
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,96,2,128,1,float16,float16,0,0.02468159943819046
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,96,1,128,1,float16,float16,0,0.024358400702476503
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,96,1,128,1,float16,float16,0,0.651038408279419
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,96,2,128,1,float16,float16,0,0.6491775989532471
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,96,2,128,1,float16,float16,0,1.378321647644043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,96,4,128,1,float16,float16,0,1.4908432006835937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,96,4,128,1,float16,float16,0,0.6862912178039551
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,96,8,128,1,float16,float16,0,0.7647776126861572
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,96,96,128,1,float16,float16,0,5.002660751342773
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,96,8,128,1,float16,float16,0,1.65579833984375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,96,1,128,1,float16,float16,0,0.3486032009124756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,96,2,128,1,float16,float16,0,0.32125120162963866
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,96,4,128,1,float16,float16,0,0.33455839157104494
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,96,96,128,1,float16,float16,0,1.2769856452941895
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,96,2,128,1,float16,float16,0,0.10057280063629151
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,96,1,128,1,float16,float16,0,0.0996016025543213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,96,96,128,1,float16,float16,0,2.5145919799804686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,96,4,128,1,float16,float16,0,0.11758719682693482
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,96,96,128,1,float16,float16,0,0.6377280235290528
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,96,1,128,1,float16,float16,0,0.04886560142040253
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,96,8,128,1,float16,float16,0,0.16286720037460328
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,96,96,128,1,float16,float16,0,0.21419520378112794
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,96,4,128,1,float16,float16,0,0.050108802318572995
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,96,96,128,1,float16,float16,0,0.039124798774719236
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,96,1,128,1,float16,float16,0,0.029108801484107973
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,96,2,128,1,float16,float16,0,0.029161599278450013
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,96,4,128,1,float16,float16,0,0.02948319911956787
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,96,2,128,1,float16,float16,0,0.049558401107788086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,96,8,128,1,float16,float16,0,0.029569599032402038
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,96,96,128,1,float16,float16,0,0.02574400007724762
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,96,8,128,1,float16,float16,0,0.050988799333572386
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,96,1,128,1,float16,float16,0,0.020363199710845947
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,96,2,128,1,float16,float16,0,0.020865599811077117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,96,8,128,1,float16,float16,0,0.020895999670028687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,96,96,128,1,float16,float16,0,0.015582400560379028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,96,4,128,1,float16,float16,0,0.02072319984436035
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,96,2,128,1,float16,float16,0,0.013199999928474426
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,96,4,128,1,float16,float16,0,0.013152000308036805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,96,1,128,1,float16,float16,0,0.013460800051689148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,96,8,128,1,float16,float16,0,0.013158400356769562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,96,1,128,1,float16,float16,0,1.3190976142883302
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,96,1,128,1,float16,float16,0,1.2792911529541016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,96,4,128,1,float16,float16,0,1.3841024398803712
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,96,2,128,1,float16,float16,0,1.3078944206237793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,96,1,128,1,float16,float16,0,0.653769588470459
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,96,2,128,1,float16,float16,0,0.6593728065490723
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,96,4,128,1,float16,float16,0,0.7187392234802246
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,96,8,128,1,float16,float16,0,0.7687424182891845
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,96,1,128,1,float16,float16,0,0.3169680118560791
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,96,96,128,1,float16,float16,0,2.495310401916504
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,96,8,128,1,float16,float16,0,0.4144911766052246
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,96,2,128,1,float16,float16,0,0.32168800830841066
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,96,96,128,1,float16,float16,0,1.26528959274292
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,96,1,128,1,float16,float16,0,0.10198080539703369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,96,2,128,1,float16,float16,0,0.10897279977798462
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,96,4,128,1,float16,float16,0,0.3486720085144043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,96,4,128,1,float16,float16,0,0.13532320261001587
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,96,8,128,1,float16,float16,0,0.1630735993385315
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,96,1,128,1,float16,float16,0,0.046891200542449954
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,96,96,128,1,float16,float16,0,0.22009119987487794
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,96,2,128,1,float16,float16,0,0.04764159917831421
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,96,4,128,1,float16,float16,0,0.04786880016326904
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,96,96,128,1,float16,float16,0,0.6443903923034668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,96,96,128,1,float16,float16,0,0.053020799160003663
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,96,1,128,1,float16,float16,0,0.027721598744392395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,96,2,128,1,float16,float16,0,0.027902400493621825
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,96,8,128,1,float16,float16,0,0.028379198908805848
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,96,96,128,1,float16,float16,0,0.029716798663139345
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,96,4,128,1,float16,float16,0,0.02774080038070679
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,96,2,128,1,float16,float16,0,0.020022399723529816
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,96,4,128,1,float16,float16,0,0.019998399913311003
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,96,8,128,1,float16,float16,0,0.020136000216007234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,96,96,128,1,float16,float16,0,0.02070239931344986
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,96,1,128,1,float16,float16,0,0.012851199507713318
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,96,2,128,1,float16,float16,0,0.012731200456619263
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,96,4,128,1,float16,float16,0,0.012779200077056884
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,96,8,128,1,float16,float16,0,0.012894399464130402
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,96,96,128,1,float16,float16,0,0.013687999546527862
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,96,1,128,1,float16,float16,0,0.011897599697113037
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,96,2,128,1,float16,float16,0,0.012006399780511856
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,96,4,128,1,float16,float16,0,0.0119439996778965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,96,8,128,1,float16,float16,0,0.012031999975442886
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,96,8,128,1,float16,float16,0,1.5877887725830078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,96,1,128,1,float16,float16,0,0.637340784072876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,96,4,128,1,float16,float16,0,0.6811423778533936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,96,2,128,1,float16,float16,0,0.6500319957733154
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,96,1,128,1,float16,float16,0,0.31744320392608644
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,96,8,128,1,float16,float16,0,0.7536416053771973
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,96,2,128,1,float16,float16,0,0.3209439992904663
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,96,4,128,1,float16,float16,0,0.3374671936035156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,96,96,128,1,float16,float16,0,1.1934752464294434
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,96,8,128,1,float16,float16,0,0.3703984022140503
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,96,1,128,1,float16,float16,0,0.09380000233650207
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,96,8,128,1,float16,float16,0,0.3655519962310791
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,96,96,128,1,float16,float16,0,0.5748735904693604
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,96,2,128,1,float16,float16,0,0.11399999856948853
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,96,4,128,1,float16,float16,0,0.15259360074996947
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,96,2,128,1,float16,float16,0,0.04726400077342987
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,96,96,128,1,float16,float16,0,0.2668400049209595
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,96,8,128,1,float16,float16,0,0.048563200235366824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,96,4,128,1,float16,float16,0,0.04796960055828094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,96,1,128,1,float16,float16,0,0.047126400470733645
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,96,8,128,1,float16,float16,0,0.04829759895801544
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,96,96,128,1,float16,float16,0,0.0804319977760315
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,96,2,128,1,float16,float16,0,0.02744640111923218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,96,4,128,1,float16,float16,0,0.027425599098205567
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,96,1,128,1,float16,float16,0,0.019782400131225585
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,96,96,128,1,float16,float16,0,0.045948800444602964
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,96,1,128,1,float16,float16,0,0.019606399536132812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,96,2,128,1,float16,float16,0,0.0196943998336792
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,96,8,128,1,float16,float16,0,0.0198512002825737
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,96,4,128,1,float16,float16,0,0.020095999538898467
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,96,1,128,1,float16,float16,0,0.012692800164222718
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,96,2,128,1,float16,float16,0,0.012681600451469422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,96,8,128,1,float16,float16,0,0.012751999497413635
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,96,4,128,1,float16,float16,0,0.013174399733543396
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,96,96,128,1,float16,float16,0,0.01906079947948456
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,96,1,128,1,float16,float16,0,0.011713600158691407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,96,4,128,1,float16,float16,0,0.01183520033955574
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,96,8,128,1,float16,float16,0,0.011856000125408172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,96,2,128,1,float16,float16,0,0.012140800058841706
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,96,96,128,1,float16,float16,0,0.013172799348831176
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,96,1,128,1,float16,float16,0,0.01162559986114502
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,96,8,128,1,float16,float16,0,0.011720000207424164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,96,4,128,1,float16,float16,0,0.011716800183057785
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,96,2,128,1,float16,float16,0,0.012025599926710128
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,96,1,128,1,float16,float16,0,0.3176527976989746
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,96,2,128,1,float16,float16,0,0.3242176055908203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,96,8,128,1,float16,float16,0,0.36442079544067385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,96,1,128,1,float16,float16,0,0.09928640127182006
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,96,4,128,1,float16,float16,0,0.11839519739151001
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,96,2,128,1,float16,float16,0,0.1199504017829895
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,96,96,128,1,float16,float16,0,0.825318431854248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,96,8,128,1,float16,float16,0,0.16075680255889893
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,96,2,128,1,float16,float16,0,0.04863199889659882
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,96,1,128,1,float16,float16,0,0.04789760112762451
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,96,4,128,1,float16,float16,0,0.04898079931735992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,96,96,128,1,float16,float16,0,0.3976128101348877
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,96,8,128,1,float16,float16,0,0.15886240005493163
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,96,8,128,1,float16,float16,0,0.059192001819610596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,96,96,128,1,float16,float16,0,0.143230402469635
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,96,2,128,1,float16,float16,0,0.027801600098609925
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,96,1,128,1,float16,float16,0,0.028036800026893616
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,96,4,128,1,float16,float16,0,0.02791680097579956
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,96,1,128,1,float16,float16,0,0.027225598692893982
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,96,8,128,1,float16,float16,0,0.03330079913139343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,96,96,128,1,float16,float16,0,0.07455199956893921
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,96,8,128,1,float16,float16,0,0.027801600098609925
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,96,1,128,1,float16,float16,0,0.019571200013160706
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,96,4,128,1,float16,float16,0,0.019704000651836397
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,96,2,128,1,float16,float16,0,0.01993599981069565
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,96,1,128,1,float16,float16,0,0.012624000012874604
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,96,2,128,1,float16,float16,0,0.012583999335765839
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,96,8,128,1,float16,float16,0,0.01977439969778061
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,96,96,128,1,float16,float16,0,0.02580159902572632
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,96,4,128,1,float16,float16,0,0.012916800379753113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,96,8,128,1,float16,float16,0,0.012732799351215362
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,96,96,128,1,float16,float16,0,0.024799999594688416
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,96,1,128,1,float16,float16,0,0.01186399981379509
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,96,2,128,1,float16,float16,0,0.012430399656295776
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,96,8,128,1,float16,float16,0,0.0118928000330925
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,96,96,128,1,float16,float16,0,0.01876640021800995
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,96,4,128,1,float16,float16,0,0.012017600238323212
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,96,1,128,1,float16,float16,0,0.011776000261306763
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,96,4,128,1,float16,float16,0,0.011644800007343293
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,96,96,128,1,float16,float16,0,0.013409599661827087
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,96,8,128,1,float16,float16,0,0.011649599671363831
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,96,2,128,1,float16,float16,0,0.011580800265073776
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,96,1,128,1,float16,float16,0,0.021678400039672852
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,96,4,128,1,float16,float16,0,0.011868800222873687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,96,2,128,1,float16,float16,0,0.0326335996389389
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,96,4,128,1,float16,float16,0,0.05561760067939758
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,96,4,128,1,float16,float16,0,0.36290559768676756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,96,1,128,1,float16,float16,0,0.013870400190353394
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,96,96,128,1,float16,float16,0,0.5075727939605713
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,96,2,128,1,float16,float16,0,0.0192208006978035
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,96,8,128,1,float16,float16,0,0.0540880024433136
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,96,96,128,1,float16,float16,0,0.2608896017074585
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,96,1,128,1,float16,float16,0,0.012380799651145935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,96,2,128,1,float16,float16,0,0.012289600074291229
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,96,4,128,1,float16,float16,0,0.017956799268722533
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,96,8,128,1,float16,float16,0,0.029756799340248108
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,96,96,128,1,float16,float16,0,0.13427040576934815
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,96,96,128,1,float16,float16,0,0.04286719858646393
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,96,1,128,1,float16,float16,0,0.01170400008559227
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,96,4,128,1,float16,float16,0,0.011603199690580369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,96,8,128,1,float16,float16,0,0.017496000230312347
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,96,2,128,1,float16,float16,0,0.011684799939393998
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,96,96,128,1,float16,float16,0,0.07069600224494935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,96,2,128,1,float16,float16,0,0.01104160025715828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,96,1,128,1,float16,float16,0,0.011475200206041336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,96,4,128,1,float16,float16,0,0.01106560006737709
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,96,2,128,1,float16,float16,0,0.011955200135707856
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,96,8,128,1,float16,float16,0,0.01154239997267723
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,96,96,128,1,float16,float16,0,0.041833600401878356
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,96,8,128,1,float16,float16,0,0.011590400338172912
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,96,1,128,1,float16,float16,0,0.011641599982976914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,96,2,128,1,float16,float16,0,0.011188799887895584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,96,8,128,1,float16,float16,0,0.01122720018029213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,96,1,128,1,float16,float16,0,0.011345600336790084
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,96,2,128,1,float16,float16,0,0.011217600107192994
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,96,8,128,1,float16,float16,0,0.09682400226593017
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,96,8,128,1,float16,float16,0,0.011190400272607804
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,96,96,128,1,float16,float16,0,0.01865759938955307
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,96,1,128,1,float16,float16,0,0.011403200030326844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,96,2,128,1,float16,float16,0,0.010878399759531022
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,96,8,128,1,float16,float16,0,0.01090560033917427
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,96,96,128,1,float16,float16,0,0.012928000092506409
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,96,1,128,1,float16,float16,0,0.011115200072526931
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,96,2,128,1,float16,float16,0,0.010820800065994262
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,96,4,128,1,float16,float16,0,0.010864000022411346
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,96,4,128,1,float16,float16,0,0.03078719973564148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,96,8,128,1,float16,float16,0,0.010830400139093399
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,96,1,128,1,float16,float16,0,0.011548800021409988
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,96,4,128,1,float16,float16,0,0.010953599959611893
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,96,96,128,1,float16,float16,0,0.024779200553894043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,96,4,128,1,float16,float16,0,0.011145599931478501
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,96,4,128,1,float16,float16,0,0.011052799969911575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,64,1,128,1,float16,float16,0,82.68960571289062
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,64,2,128,1,float16,float16,0,84.16561889648438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,64,4,128,1,float16,float16,0,84.53244018554688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,64,1,128,1,float16,float16,0,41.665505981445314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,64,2,128,1,float16,float16,0,170.7490966796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,64,1,128,1,float16,float16,0,171.037451171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,64,8,128,1,float16,float16,0,85.03199462890625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,64,4,128,1,float16,float16,0,170.06202392578126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,64,2,128,1,float16,float16,0,42.042807006835936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,64,8,128,1,float16,float16,0,173.17821044921874
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,64,2,128,1,float16,float16,0,20.374647521972655
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,64,1,128,1,float16,float16,0,20.535072326660156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,64,4,128,1,float16,float16,0,20.954771423339842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,64,4,128,1,float16,float16,0,43.524356079101565
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,64,8,128,1,float16,float16,0,21.716290283203126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,64,8,128,1,float16,float16,0,43.892764282226565
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,64,1,128,1,float16,float16,0,95.43135986328124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,64,64,128,1,float16,float16,0,21.658737182617188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,64,2,128,1,float16,float16,0,94.71848754882812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,64,64,128,1,float16,float16,0,45.58158569335937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,64,8,128,1,float16,float16,0,98.15897216796876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,64,4,128,1,float16,float16,0,99.64034423828124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,64,64,128,1,float16,float16,0,53.79727783203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,64,1,128,1,float16,float16,0,47.87965698242188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,64,2,128,1,float16,float16,0,47.9826904296875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,64,4,128,1,float16,float16,0,48.258905029296876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,64,1,128,1,float16,float16,0,24.014320373535156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,64,64,128,1,float16,float16,0,26.286392211914062
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,64,2,128,1,float16,float16,0,23.60426025390625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,64,64,128,1,float16,float16,0,89.79639282226563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,64,64,128,1,float16,float16,0,12.958224487304687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,64,2,128,1,float16,float16,0,11.396056365966796
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,64,1,128,1,float16,float16,0,12.204478454589843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,64,4,128,1,float16,float16,0,12.407292938232422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,64,8,128,1,float16,float16,0,49.648843383789064
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,64,4,128,1,float16,float16,0,24.601690673828124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,64,8,128,1,float16,float16,0,24.942184448242188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,64,8,128,1,float16,float16,0,11.760969543457032
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,64,1,128,1,float16,float16,0,33.3799560546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,64,2,128,1,float16,float16,0,33.868283081054685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,64,64,128,1,float16,float16,0,38.847119140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,64,4,128,1,float16,float16,0,34.04977722167969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,64,64,128,1,float16,float16,0,18.629548645019533
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,64,2,128,1,float16,float16,0,16.43382110595703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,64,1,128,1,float16,float16,0,16.413003540039064
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,64,1,128,1,float16,float16,0,66.2700927734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,64,2,128,1,float16,float16,0,66.53966064453125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,64,64,128,1,float16,float16,0,9.645104217529298
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,64,4,128,1,float16,float16,0,70.28902587890624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,64,1,128,1,float16,float16,0,8.073435211181641
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,64,8,128,1,float16,float16,0,34.35621948242188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,64,8,128,1,float16,float16,0,69.90821533203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,64,2,128,1,float16,float16,0,8.388410949707032
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,64,4,128,1,float16,float16,0,17.415443420410156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,64,8,128,1,float16,float16,0,17.251641845703126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,64,4,128,1,float16,float16,0,8.586408233642578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,64,8,128,1,float16,float16,0,8.656198120117187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,64,1,128,1,float16,float16,0,43.14847106933594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,64,2,128,1,float16,float16,0,44.61678466796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,64,4,128,1,float16,float16,0,45.79450378417969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,64,64,128,1,float16,float16,0,50.06562805175781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,64,1,128,1,float16,float16,0,21.97950439453125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,64,2,128,1,float16,float16,0,21.185032653808594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,64,64,128,1,float16,float16,0,25.328764343261717
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,64,1,128,1,float16,float16,0,86.26820678710938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,64,2,128,1,float16,float16,0,87.62166137695313
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,64,64,128,1,float16,float16,0,12.884959411621093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,64,8,128,1,float16,float16,0,89.14686279296875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,64,8,128,1,float16,float16,0,44.638632202148436
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,64,4,128,1,float16,float16,0,90.9729248046875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,64,1,128,1,float16,float16,0,10.685460662841797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,64,4,128,1,float16,float16,0,22.445806884765624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,64,8,128,1,float16,float16,0,22.453659057617188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,64,64,128,1,float16,float16,0,5.8635822296142575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,64,1,128,1,float16,float16,0,5.2620094299316404
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,64,2,128,1,float16,float16,0,10.784486389160156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,64,4,128,1,float16,float16,0,10.889358520507812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,64,8,128,1,float16,float16,0,10.814636993408204
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,64,2,128,1,float16,float16,0,5.1806079864501955
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,64,4,128,1,float16,float16,0,5.424844741821289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,64,8,128,1,float16,float16,0,5.167582321166992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,64,1,128,1,float16,float16,0,24.717388916015626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,64,2,128,1,float16,float16,0,25.535238647460936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,64,4,128,1,float16,float16,0,25.663699340820312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,64,64,128,1,float16,float16,0,29.281338500976563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,64,1,128,1,float16,float16,0,12.235649871826173
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,64,2,128,1,float16,float16,0,11.594158172607422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,64,64,128,1,float16,float16,0,14.899496459960938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,64,1,128,1,float16,float16,0,49.606076049804685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,64,2,128,1,float16,float16,0,51.587493896484375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,64,4,128,1,float16,float16,0,51.490545654296874
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,64,64,128,1,float16,float16,0,7.333718109130859
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,64,8,128,1,float16,float16,0,25.558865356445313
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,64,4,128,1,float16,float16,0,12.523390197753907
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,64,1,128,1,float16,float16,0,6.062731170654297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,64,8,128,1,float16,float16,0,53.0125732421875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,64,2,128,1,float16,float16,0,6.119361495971679
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,64,8,128,1,float16,float16,0,12.565151977539063
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,64,64,128,1,float16,float16,0,3.6394561767578124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,64,1,128,1,float16,float16,0,2.8114479064941404
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,64,2,128,1,float16,float16,0,2.86038875579834
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,64,4,128,1,float16,float16,0,6.262361526489258
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,64,4,128,1,float16,float16,0,2.8591808319091796
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,64,8,128,1,float16,float16,0,6.318719863891602
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,64,8,128,1,float16,float16,0,3.0087247848510743
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,64,1,128,1,float16,float16,0,23.428952026367188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,64,2,128,1,float16,float16,0,23.62382354736328
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,64,4,128,1,float16,float16,0,24.31134033203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,64,64,128,1,float16,float16,0,28.349508666992186
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,64,1,128,1,float16,float16,0,11.712529754638672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,64,64,128,1,float16,float16,0,14.40709991455078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,64,2,128,1,float16,float16,0,12.034019470214844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,64,1,128,1,float16,float16,0,45.10052490234375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,64,2,128,1,float16,float16,0,46.18824768066406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,64,4,128,1,float16,float16,0,47.2343994140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,64,8,128,1,float16,float16,0,48.299618530273435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,64,8,128,1,float16,float16,0,24.927928161621093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,64,64,128,1,float16,float16,0,6.8500419616699215
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,64,1,128,1,float16,float16,0,5.05230712890625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,64,4,128,1,float16,float16,0,12.208710479736329
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,64,2,128,1,float16,float16,0,5.546702575683594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,64,8,128,1,float16,float16,0,11.965265655517578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,64,1,128,1,float16,float16,0,2.5462671279907227
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,64,2,128,1,float16,float16,0,2.614784049987793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,64,64,128,1,float16,float16,0,3.325116729736328
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,64,4,128,1,float16,float16,0,5.615196609497071
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,64,4,128,1,float16,float16,0,2.762603187561035
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,64,64,128,1,float16,float16,0,1.6999296188354491
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,64,1,128,1,float16,float16,0,1.329916763305664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,64,8,128,1,float16,float16,0,5.665100860595703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,64,8,128,1,float16,float16,0,2.6688032150268555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,64,2,128,1,float16,float16,0,1.226198387145996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,64,4,128,1,float16,float16,0,1.4026448249816894
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,64,8,128,1,float16,float16,0,1.4418496131896972
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,64,1,128,1,float16,float16,0,12.99793701171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,64,2,128,1,float16,float16,0,13.652568054199218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,64,64,128,1,float16,float16,0,17.146241760253908
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,64,4,128,1,float16,float16,0,13.553764343261719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,64,64,128,1,float16,float16,0,8.410321807861328
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,64,1,128,1,float16,float16,0,6.485697937011719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,64,2,128,1,float16,float16,0,26.19317626953125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,64,1,128,1,float16,float16,0,26.788873291015626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,64,4,128,1,float16,float16,0,27.589840698242188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,64,8,128,1,float16,float16,0,14.10535125732422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,64,2,128,1,float16,float16,0,6.3869377136230465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,64,8,128,1,float16,float16,0,28.383096313476564
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,64,1,128,1,float16,float16,0,3.3268096923828123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,64,64,128,1,float16,float16,0,4.2925537109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,64,64,128,1,float16,float16,0,2.0603471755981446
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,64,2,128,1,float16,float16,0,3.124662399291992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,64,4,128,1,float16,float16,0,6.726464080810547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,64,8,128,1,float16,float16,0,3.1780303955078124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,64,4,128,1,float16,float16,0,3.0935247421264647
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,64,8,128,1,float16,float16,0,6.8418830871582035
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,64,1,128,1,float16,float16,0,1.523958396911621
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,64,2,128,1,float16,float16,0,1.390505599975586
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,64,2,128,1,float16,float16,0,0.822214412689209
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,64,64,128,1,float16,float16,0,1.1123663902282714
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,64,4,128,1,float16,float16,0,1.365556812286377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,64,8,128,1,float16,float16,0,0.8129391670227051
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,64,8,128,1,float16,float16,0,1.5350192070007325
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,64,4,128,1,float16,float16,0,0.9343615531921386
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,64,1,128,1,float16,float16,0,0.7868752002716064
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,64,1,128,1,float16,float16,0,12.284142303466798
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,64,2,128,1,float16,float16,0,12.925920104980468
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,64,4,128,1,float16,float16,0,12.530919647216797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,64,64,128,1,float16,float16,0,17.074122619628906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,64,1,128,1,float16,float16,0,5.931081771850586
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,64,64,128,1,float16,float16,0,8.599479675292969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,64,1,128,1,float16,float16,0,24.37323913574219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,64,2,128,1,float16,float16,0,5.964286422729492
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,64,2,128,1,float16,float16,0,25.085600280761717
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,64,8,128,1,float16,float16,0,13.312939453125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,64,4,128,1,float16,float16,0,27.314544677734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,64,8,128,1,float16,float16,0,26.723760986328124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,64,1,128,1,float16,float16,0,2.777950477600098
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,64,2,128,1,float16,float16,0,2.823054313659668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,64,64,128,1,float16,float16,0,4.213032150268555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,64,4,128,1,float16,float16,0,6.22070541381836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,64,1,128,1,float16,float16,0,1.4064895629882812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,64,4,128,1,float16,float16,0,2.9270448684692383
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,64,8,128,1,float16,float16,0,6.522294616699218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,64,64,128,1,float16,float16,0,2.177569580078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,64,2,128,1,float16,float16,0,1.5112496376037599
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,64,8,128,1,float16,float16,0,3.024211120605469
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,64,4,128,1,float16,float16,0,1.2668031692504882
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,64,1,128,1,float16,float16,0,0.7364463806152344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,64,4,128,1,float16,float16,0,0.7074687957763672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,64,2,128,1,float16,float16,0,0.7904096126556397
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,64,8,128,1,float16,float16,0,1.5584464073181152
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,64,8,128,1,float16,float16,0,0.7718175888061524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,64,2,128,1,float16,float16,0,0.3792448043823242
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,64,1,128,1,float16,float16,0,0.40050082206726073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,64,8,128,1,float16,float16,0,0.37364959716796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,64,4,128,1,float16,float16,0,0.3547647953033447
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,64,64,128,1,float16,float16,0,0.6580175876617431
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,64,64,128,1,float16,float16,0,1.0956048011779784
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,64,1,128,1,float16,float16,0,6.814701080322266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,64,2,128,1,float16,float16,0,7.482276916503906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,64,4,128,1,float16,float16,0,6.840366363525391
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,64,64,128,1,float16,float16,0,10.94863510131836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,64,1,128,1,float16,float16,0,3.5325183868408203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,64,64,128,1,float16,float16,0,5.5308574676513675
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,64,2,128,1,float16,float16,0,15.140345764160156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,64,1,128,1,float16,float16,0,15.604176330566407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,64,4,128,1,float16,float16,0,15.268125915527344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,64,8,128,1,float16,float16,0,7.672672271728516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,64,2,128,1,float16,float16,0,2.891969680786133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,64,8,128,1,float16,float16,0,15.964923095703124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,64,1,128,1,float16,float16,0,1.627222442626953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,64,2,128,1,float16,float16,0,1.5532336235046387
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,64,4,128,1,float16,float16,0,1.655931282043457
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,64,8,128,1,float16,float16,0,1.8706592559814452
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,64,64,128,1,float16,float16,0,1.4348272323608398
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,64,4,128,1,float16,float16,0,3.587246322631836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,64,8,128,1,float16,float16,0,3.5125568389892576
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,64,64,128,1,float16,float16,0,2.753414344787598
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,64,1,128,1,float16,float16,0,0.8511856079101563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,64,1,128,1,float16,float16,0,0.4609936237335205
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,64,4,128,1,float16,float16,0,0.8074943542480468
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,64,64,128,1,float16,float16,0,0.7086160182952881
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,64,2,128,1,float16,float16,0,0.8930912017822266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,64,4,128,1,float16,float16,0,0.4257391929626465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,64,8,128,1,float16,float16,0,0.9024191856384277
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,64,2,128,1,float16,float16,0,0.5169760227203369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,64,64,128,1,float16,float16,0,0.3783616065979004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,64,8,128,1,float16,float16,0,0.47499680519104004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,64,4,128,1,float16,float16,0,0.22614879608154298
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,64,2,128,1,float16,float16,0,0.214684796333313
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,64,8,128,1,float16,float16,0,0.245849609375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,64,1,128,1,float16,float16,0,0.21731359958648683
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,64,1,128,1,float16,float16,0,7.0809165954589846
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,64,2,128,1,float16,float16,0,6.890662384033203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,64,64,128,1,float16,float16,0,13.1066162109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,64,2,128,1,float16,float16,0,14.67293701171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,64,1,128,1,float16,float16,0,14.75347900390625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,64,4,128,1,float16,float16,0,15.146266174316406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,64,4,128,1,float16,float16,0,7.523808288574219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,64,8,128,1,float16,float16,0,15.623139953613281
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,64,8,128,1,float16,float16,0,7.025392150878906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,64,1,128,1,float16,float16,0,3.1546335220336914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,64,1,128,1,float16,float16,0,1.9064992904663085
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,64,64,128,1,float16,float16,0,6.591909027099609
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,64,4,128,1,float16,float16,0,3.4708736419677733
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,64,2,128,1,float16,float16,0,3.645979309082031
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,64,64,128,1,float16,float16,0,3.3696128845214846
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,64,2,128,1,float16,float16,0,1.5629440307617188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,64,8,128,1,float16,float16,0,3.6866832733154298
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,64,1,128,1,float16,float16,0,0.8568991661071778
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,64,2,128,1,float16,float16,0,0.929257583618164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,64,8,128,1,float16,float16,0,1.6095632553100585
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,64,4,128,1,float16,float16,0,0.7495632171630859
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,64,8,128,1,float16,float16,0,0.9064656257629394
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,64,1,128,1,float16,float16,0,0.4106287956237793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,64,64,128,1,float16,float16,0,0.888424015045166
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,64,2,128,1,float16,float16,0,0.4465792179107666
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,64,64,128,1,float16,float16,0,1.8159584045410155
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,64,4,128,1,float16,float16,0,0.5038559913635254
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,64,4,128,1,float16,float16,0,0.23094398975372316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,64,2,128,1,float16,float16,0,0.23967199325561522
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,64,8,128,1,float16,float16,0,0.4230224132537842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,64,64,128,1,float16,float16,0,0.41469922065734866
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,64,64,128,1,float16,float16,0,0.16531840562820435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,64,1,128,1,float16,float16,0,0.12232799530029297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,64,2,128,1,float16,float16,0,0.13635679483413696
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,64,8,128,1,float16,float16,0,0.19761120080947875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,64,4,128,1,float16,float16,0,0.12439999580383301
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,64,8,128,1,float16,float16,0,0.12360479831695556
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,64,4,128,1,float16,float16,0,1.6832319259643556
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,64,1,128,1,float16,float16,0,0.20312800407409667
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,64,1,128,1,float16,float16,0,4.086305618286133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,64,2,128,1,float16,float16,0,3.6977134704589845
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,64,4,128,1,float16,float16,0,4.373665618896484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,64,1,128,1,float16,float16,0,8.738091278076173
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,64,2,128,1,float16,float16,0,8.986402893066407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,64,4,128,1,float16,float16,0,9.548492431640625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,64,8,128,1,float16,float16,0,9.56559066772461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,64,8,128,1,float16,float16,0,4.671988677978516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,64,1,128,1,float16,float16,0,2.135684776306152
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,64,2,128,1,float16,float16,0,2.037188720703125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,64,1,128,1,float16,float16,0,0.9826800346374511
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,64,4,128,1,float16,float16,0,2.0931215286254883
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,64,2,128,1,float16,float16,0,1.0267168045043946
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,64,8,128,1,float16,float16,0,2.36572151184082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,64,1,128,1,float16,float16,0,0.5366943836212158
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,64,8,128,1,float16,float16,0,1.0970239639282227
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,64,4,128,1,float16,float16,0,1.0517647743225098
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,64,64,128,1,float16,float16,0,12.969834899902343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,64,2,128,1,float16,float16,0,0.5140384197235107
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,64,64,128,1,float16,float16,0,3.435348892211914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,64,64,128,1,float16,float16,0,6.5515602111816404
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,64,8,128,1,float16,float16,0,0.5665599822998046
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,64,64,128,1,float16,float16,0,1.7718751907348633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,64,64,128,1,float16,float16,0,0.8591360092163086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,64,2,128,1,float16,float16,0,0.267305588722229
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,64,1,128,1,float16,float16,0,0.11519520282745362
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,64,2,128,1,float16,float16,0,0.12117600440979004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,64,64,128,1,float16,float16,0,0.378385591506958
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,64,4,128,1,float16,float16,0,0.1309167981147766
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,64,1,128,1,float16,float16,0,0.2482991933822632
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,64,8,128,1,float16,float16,0,0.11810879707336426
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,64,8,128,1,float16,float16,0,0.29306559562683104
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,64,2,128,1,float16,float16,0,0.06593279838562012
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,64,8,128,1,float16,float16,0,0.0711072027683258
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,64,1,128,1,float16,float16,0,0.03850559890270233
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,64,64,128,1,float16,float16,0,0.04683519899845123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,64,4,128,1,float16,float16,0,0.07466239929199218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,64,2,128,1,float16,float16,0,0.04067519903182983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,64,4,128,1,float16,float16,0,0.04506399929523468
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,64,8,128,1,float16,float16,0,0.040571200847625735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,64,4,128,1,float16,float16,0,0.5367119789123536
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,64,4,128,1,float16,float16,0,0.26956000328063967
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,64,64,128,1,float16,float16,0,0.09670559763908386
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,64,1,128,1,float16,float16,0,0.0661728024482727
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,64,2,128,1,float16,float16,0,3.6785743713378904
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,64,1,128,1,float16,float16,0,3.8499664306640624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,64,4,128,1,float16,float16,0,4.004779052734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,64,2,128,1,float16,float16,0,1.838150405883789
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,64,1,128,1,float16,float16,0,1.8075599670410156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,64,8,128,1,float16,float16,0,4.6752174377441404
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,64,1,128,1,float16,float16,0,0.9639856338500976
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,64,4,128,1,float16,float16,0,2.024140739440918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,64,2,128,1,float16,float16,0,0.9235615730285645
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,64,4,128,1,float16,float16,0,1.0183631896972656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,64,8,128,1,float16,float16,0,2.44549446105957
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,64,1,128,1,float16,float16,0,0.4382175922393799
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,64,8,128,1,float16,float16,0,1.2652544021606444
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,64,2,128,1,float16,float16,0,0.47054400444030764
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,64,4,128,1,float16,float16,0,0.5091119766235351
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,64,8,128,1,float16,float16,0,0.5560512065887451
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,64,64,128,1,float16,float16,0,1.6677919387817384
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,64,1,128,1,float16,float16,0,0.2105952024459839
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,64,64,128,1,float16,float16,0,3.348231887817383
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,64,2,128,1,float16,float16,0,0.20815999507904054
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,64,4,128,1,float16,float16,0,0.22214560508728026
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,64,2,128,1,float16,float16,0,0.07962719798088073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,64,8,128,1,float16,float16,0,0.2523519992828369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,64,64,128,1,float16,float16,0,0.3671056032180786
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,64,4,128,1,float16,float16,0,0.08094239830970765
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,64,8,128,1,float16,float16,0,0.08235200047492981
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,64,1,128,1,float16,float16,0,0.08081120252609253
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,64,1,128,1,float16,float16,0,0.04580320119857788
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,64,2,128,1,float16,float16,0,0.04577760100364685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,64,4,128,1,float16,float16,0,0.046312001347541806
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,64,64,128,1,float16,float16,0,0.03483999967575073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,64,1,128,1,float16,float16,0,0.028486400842666626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,64,2,128,1,float16,float16,0,0.028439998626708984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,64,8,128,1,float16,float16,0,0.04725280106067657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,64,4,128,1,float16,float16,0,0.02847839891910553
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,64,64,128,1,float16,float16,0,6.534078216552734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,64,8,128,1,float16,float16,0,0.028932800889015196
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,64,64,128,1,float16,float16,0,0.020526400208473204
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,64,1,128,1,float16,float16,0,0.01720159947872162
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,64,2,128,1,float16,float16,0,0.017339199781417847
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,64,8,128,1,float16,float16,0,0.017552000284194947
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,64,4,128,1,float16,float16,0,0.017268800735473634
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,64,1,128,1,float16,float16,0,1.765652847290039
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,64,2,128,1,float16,float16,0,1.8747392654418946
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,64,4,128,1,float16,float16,0,2.039695930480957
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,64,8,128,1,float16,float16,0,2.363259124755859
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,64,64,128,1,float16,float16,0,0.8516976356506347
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,64,64,128,1,float16,float16,0,0.0678704023361206
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,64,1,128,1,float16,float16,0,0.8990351676940918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,64,2,128,1,float16,float16,0,0.9397120475769043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,64,1,128,1,float16,float16,0,0.4239039897918701
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,64,4,128,1,float16,float16,0,1.0507360458374024
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,64,2,128,1,float16,float16,0,0.44463839530944826
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,64,8,128,1,float16,float16,0,1.1813568115234374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,64,4,128,1,float16,float16,0,0.4790192127227783
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,64,8,128,1,float16,float16,0,0.5503344058990478
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,64,1,128,1,float16,float16,0,0.20091519355773926
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,64,64,128,1,float16,float16,0,3.3193504333496096
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,64,2,128,1,float16,float16,0,0.21397120952606202
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,64,64,128,1,float16,float16,0,1.6654991149902343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,64,4,128,1,float16,float16,0,0.22434399127960206
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,64,2,128,1,float16,float16,0,0.06266080141067505
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,64,8,128,1,float16,float16,0,0.2532383918762207
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,64,64,128,1,float16,float16,0,0.8504400253295898
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,64,64,128,1,float16,float16,0,0.3649168014526367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,64,1,128,1,float16,float16,0,0.03617919981479645
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,64,64,128,1,float16,float16,0,0.054180800914764404
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,64,1,128,1,float16,float16,0,0.06268960237503052
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,64,8,128,1,float16,float16,0,0.06476960182189942
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,64,4,128,1,float16,float16,0,0.03667519986629486
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,64,8,128,1,float16,float16,0,0.037136000394821164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,64,4,128,1,float16,float16,0,0.06395519971847534
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,64,64,128,1,float16,float16,0,0.0283488005399704
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,64,2,128,1,float16,float16,0,0.02184640020132065
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,64,2,128,1,float16,float16,0,0.03630400002002716
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,64,4,128,1,float16,float16,0,0.021966400742530822
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,64,8,128,1,float16,float16,0,0.022592000663280487
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,64,64,128,1,float16,float16,0,0.016920000314712524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,64,1,128,1,float16,float16,0,0.013990400731563568
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,64,1,128,1,float16,float16,0,0.021899199485778807
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,64,2,128,1,float16,float16,0,0.013574400544166565
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,64,8,128,1,float16,float16,0,0.013795199990272521
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,64,4,128,1,float16,float16,0,0.013692800700664521
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,64,1,128,1,float16,float16,0,0.012417600303888322
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,64,64,128,1,float16,float16,0,0.013966399431228637
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,64,2,128,1,float16,float16,0,0.012270399928092956
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,64,4,128,1,float16,float16,0,0.012372799962759019
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,64,8,128,1,float16,float16,0,0.012398400157690049
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,64,2,128,1,float16,float16,0,0.4431920051574707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,64,1,128,1,float16,float16,0,0.4290287971496582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,64,1,128,1,float16,float16,0,0.8666463851928711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,64,4,128,1,float16,float16,0,0.4776432037353516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,64,2,128,1,float16,float16,0,0.9148271560668946
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,64,1,128,1,float16,float16,0,0.2070688009262085
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,64,4,128,1,float16,float16,0,0.9762880325317382
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,64,8,128,1,float16,float16,0,1.140278434753418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,64,2,128,1,float16,float16,0,0.20692160129547119
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,64,8,128,1,float16,float16,0,0.5545599937438965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,64,8,128,1,float16,float16,0,0.2548288106918335
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,64,1,128,1,float16,float16,0,0.06148639917373657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,64,4,128,1,float16,float16,0,0.062009602785110474
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,64,4,128,1,float16,float16,0,0.22460319995880126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,64,64,128,1,float16,float16,0,1.6618719100952148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,64,64,128,1,float16,float16,0,0.8430080413818359
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,64,64,128,1,float16,float16,0,0.3752703905105591
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,64,8,128,1,float16,float16,0,0.06383519768714904
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,64,1,128,1,float16,float16,0,0.03458879888057709
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,64,64,128,1,float16,float16,0,0.07656800150871276
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,64,2,128,1,float16,float16,0,0.034587201476097104
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,64,1,128,1,float16,float16,0,0.020721599459648132
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,64,2,128,1,float16,float16,0,0.020883199572563172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,64,64,128,1,float16,float16,0,0.037520000338554384
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,64,4,128,1,float16,float16,0,0.02104319930076599
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,64,8,128,1,float16,float16,0,0.03543359935283661
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,64,64,128,1,float16,float16,0,0.02147040069103241
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,64,8,128,1,float16,float16,0,0.021216000616550445
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,64,1,128,1,float16,float16,0,0.013512000441551208
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,64,2,128,1,float16,float16,0,0.013390399515628815
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,64,8,128,1,float16,float16,0,0.013382400572299957
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,64,64,128,1,float16,float16,0,0.013763199746608733
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,64,1,128,1,float16,float16,0,0.012164799869060517
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,64,2,128,1,float16,float16,0,0.012086399644613267
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,64,4,128,1,float16,float16,0,0.012193600088357926
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,64,8,128,1,float16,float16,0,0.012227199971675873
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,64,64,128,1,float16,float16,0,0.012825599312782288
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,64,1,128,1,float16,float16,0,0.011828800290822982
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,64,8,128,1,float16,float16,0,0.011926399916410447
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,64,2,128,1,float16,float16,0,0.011851199716329575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,64,4,128,1,float16,float16,0,0.011846400052309036
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,64,1,128,1,float16,float16,0,0.4282815933227539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,64,2,128,1,float16,float16,0,0.44279842376708983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,64,4,128,1,float16,float16,0,0.4720320224761963
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,64,1,128,1,float16,float16,0,0.20787360668182372
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,64,8,128,1,float16,float16,0,0.5327648162841797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,64,2,128,1,float16,float16,0,0.2103343963623047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,64,8,128,1,float16,float16,0,0.25154240131378175
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,64,64,128,1,float16,float16,0,0.7800943851470947
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,64,1,128,1,float16,float16,0,0.06115520000457764
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,64,64,128,1,float16,float16,0,0.3740528106689453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,64,2,128,1,float16,float16,0,0.061312001943588254
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,64,4,128,1,float16,float16,0,0.06184160113334656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,64,2,128,1,float16,float16,0,0.06131680011749267
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,64,8,128,1,float16,float16,0,0.0635695993900299
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,64,1,128,1,float16,float16,0,0.034929600358009336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,64,64,128,1,float16,float16,0,0.12643680572509766
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,64,2,128,1,float16,float16,0,0.034841600060462954
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,64,4,128,1,float16,float16,0,0.03501279950141907
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,64,64,128,1,float16,float16,0,0.05897279977798462
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,64,1,128,1,float16,float16,0,0.020598399639129638
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,64,2,128,1,float16,float16,0,0.020662400126457214
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,64,4,128,1,float16,float16,0,0.020555199682712556
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,64,8,128,1,float16,float16,0,0.02114560008049011
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,64,1,128,1,float16,float16,0,0.013332800567150116
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,64,4,128,1,float16,float16,0,0.013332800567150116
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,64,2,128,1,float16,float16,0,0.013223999738693237
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,64,8,128,1,float16,float16,0,0.01343040019273758
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,64,4,128,1,float16,float16,0,0.013251200318336487
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,64,64,128,1,float16,float16,0,0.019011199474334717
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,64,1,128,1,float16,float16,0,0.011947199702262878
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,64,2,128,1,float16,float16,0,0.01202400028705597
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,64,8,128,1,float16,float16,0,0.012204799801111221
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,64,64,128,1,float16,float16,0,0.01276639997959137
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,64,2,128,1,float16,float16,0,0.011697600036859513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,64,1,128,1,float16,float16,0,0.01165279969573021
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,64,4,128,1,float16,float16,0,0.011771199852228164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,64,8,128,1,float16,float16,0,0.011774399876594543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,64,1,128,1,float16,float16,0,0.011622399836778641
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,64,64,128,1,float16,float16,0,0.012382400035858155
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,64,2,128,1,float16,float16,0,0.011664000153541566
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,64,4,128,1,float16,float16,0,0.01173280030488968
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,64,8,128,1,float16,float16,0,0.011681599915027619
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,64,1,128,1,float16,float16,0,0.2055408000946045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,64,2,128,1,float16,float16,0,0.21647679805755615
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,64,4,128,1,float16,float16,0,0.22711679935455323
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,64,4,128,1,float16,float16,0,0.22546560764312745
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,64,8,128,1,float16,float16,0,0.2557136058807373
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,64,1,128,1,float16,float16,0,0.061947202682495116
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,64,2,128,1,float16,float16,0,0.06183519959449768
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,64,4,128,1,float16,float16,0,0.06234560012817383
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,64,8,128,1,float16,float16,0,0.06393120288848878
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,64,64,128,1,float16,float16,0,0.5467855930328369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,64,1,128,1,float16,float16,0,0.034913599491119385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,64,64,128,1,float16,float16,0,0.21806719303131103
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,64,4,128,1,float16,float16,0,0.035209599137306216
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,64,8,128,1,float16,float16,0,0.036073601245880126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,64,2,128,1,float16,float16,0,0.0350271999835968
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,64,8,128,1,float16,float16,0,0.03590719997882843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,64,1,128,1,float16,float16,0,0.020916800200939178
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,64,2,128,1,float16,float16,0,0.020764799416065217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,64,64,128,1,float16,float16,0,0.09937919974327088
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,64,8,128,1,float16,float16,0,0.0213359996676445
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,64,64,128,1,float16,float16,0,0.05496159791946411
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,64,1,128,1,float16,float16,0,0.013153600692749023
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,64,64,128,1,float16,float16,0,0.032395198941230774
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,64,2,128,1,float16,float16,0,0.013212800025939941
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,64,4,128,1,float16,float16,0,0.013147200644016265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,64,8,128,1,float16,float16,0,0.013289600610733032
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,64,1,128,1,float16,float16,0,0.011985599994659424
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,64,2,128,1,float16,float16,0,0.011958400160074234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,64,4,128,1,float16,float16,0,0.011980800330638886
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,64,8,128,1,float16,float16,0,0.0121568001806736
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,64,64,128,1,float16,float16,0,0.01833920031785965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,64,4,128,1,float16,float16,0,0.012080000340938568
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,64,1,128,1,float16,float16,0,0.01159999966621399
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,64,4,128,1,float16,float16,0,0.011670400202274323
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,64,2,128,1,float16,float16,0,0.01167680025100708
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,64,64,128,1,float16,float16,0,0.01255040019750595
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,64,1,128,1,float16,float16,0,0.011585599929094314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,64,2,128,1,float16,float16,0,0.011667200177907944
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,64,8,128,1,float16,float16,0,0.01204800009727478
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,64,4,128,1,float16,float16,0,0.011617600172758102
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,64,8,128,1,float16,float16,0,0.011608000099658965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,64,1,128,1,float16,float16,0,0.011555200070142746
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,64,2,128,1,float16,float16,0,0.011598400026559829
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,64,4,128,1,float16,float16,0,0.01154559999704361
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,64,1,128,1,float16,float16,0,0.020601600408554077
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,64,4,128,1,float16,float16,0,0.05529919862747192
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,64,2,128,1,float16,float16,0,0.031732800602912906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,64,8,128,1,float16,float16,0,0.09700160026550293
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,64,1,128,1,float16,float16,0,0.013152000308036805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,64,2,128,1,float16,float16,0,0.018671999871730804
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,64,64,128,1,float16,float16,0,0.3428080081939697
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,64,4,128,1,float16,float16,0,0.030449599027633667
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,64,8,128,1,float16,float16,0,0.054199999570846556
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,64,64,128,1,float16,float16,0,0.17487200498580932
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,64,2,128,1,float16,float16,0,0.012030400335788727
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,64,4,128,1,float16,float16,0,0.017924800515174866
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,64,8,128,1,float16,float16,0,0.029791998863220214
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,64,64,128,1,float16,float16,0,0.09353280067443848
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,64,1,128,1,float16,float16,0,0.011481600254774094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,64,2,128,1,float16,float16,0,0.011427199840545655
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,64,4,128,1,float16,float16,0,0.02080959975719452
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,64,4,128,1,float16,float16,0,0.011608000099658965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,64,64,128,1,float16,float16,0,0.052753597497940063
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,64,8,128,1,float16,float16,0,0.01764000058174133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,64,1,128,1,float16,float16,0,0.011105599999427795
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,64,64,128,1,float16,float16,0,0.03039200007915497
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,64,2,128,1,float16,float16,0,0.011155200004577637
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,64,4,128,1,float16,float16,0,0.011060799658298492
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,64,1,128,1,float16,float16,0,0.011055999994277954
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,64,8,128,1,float16,float16,0,0.011684799939393998
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,64,64,128,1,float16,float16,0,0.029812800884246825
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,64,2,128,1,float16,float16,0,0.010891199856996537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,64,4,128,1,float16,float16,0,0.011240000277757645
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,64,64,128,1,float16,float16,0,0.018136000633239745
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,64,2,128,1,float16,float16,0,0.010886400192975997
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,64,1,128,1,float16,float16,0,0.01132960021495819
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,64,4,128,1,float16,float16,0,0.010894399881362916
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,64,64,128,1,float16,float16,0,0.012625600397586822
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,64,64,128,1,float16,float16,0,0.01233920007944107
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,64,8,128,1,float16,float16,0,0.011238399893045425
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,64,8,128,1,float16,float16,0,0.011577600240707397
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,64,2,128,1,float16,float16,0,0.01088000014424324
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,64,1,128,1,float16,float16,0,0.011255999654531479
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,64,8,128,1,float16,float16,0,0.010966400057077408
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,64,64,128,1,float16,float16,0,0.012070400267839431
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,64,2,128,1,float16,float16,0,0.010833600163459777
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,64,4,128,1,float16,float16,0,0.010739199817180634
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,64,1,128,1,float16,float16,0,0.011336000263690948
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,64,8,128,1,float16,float16,0,0.010849600285291671
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,64,1,128,1,float16,float16,0,0.011958400160074234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,64,4,128,1,float16,float16,0,0.03505760133266449
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,64,8,128,1,float16,float16,0,0.011155200004577637
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,64,4,128,1,float16,float16,0,0.011185599863529206
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,48,1,128,1,float16,float16,0,62.80623168945313
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,48,2,128,1,float16,float16,0,63.407080078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,48,4,128,1,float16,float16,0,63.5035400390625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,48,1,128,1,float16,float16,0,31.74486389160156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,48,2,128,1,float16,float16,0,126.5727294921875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,48,1,128,1,float16,float16,0,128.23038330078126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,48,4,128,1,float16,float16,0,128.3619384765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,48,8,128,1,float16,float16,0,65.81542358398437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,48,2,128,1,float16,float16,0,31.68184814453125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,48,8,128,1,float16,float16,0,133.2563232421875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,48,1,128,1,float16,float16,0,15.566316223144531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,48,2,128,1,float16,float16,0,15.555487060546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,48,4,128,1,float16,float16,0,16.021269226074217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,48,4,128,1,float16,float16,0,32.00316467285156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,48,8,128,1,float16,float16,0,32.97011108398438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,48,8,128,1,float16,float16,0,16.55518341064453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,48,1,128,1,float16,float16,0,71.62242431640625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,48,2,128,1,float16,float16,0,73.81624145507813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,48,4,128,1,float16,float16,0,73.45841064453126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,48,8,128,1,float16,float16,0,76.01373901367188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,48,1,128,1,float16,float16,0,35.86967468261719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,48,2,128,1,float16,float16,0,37.09742126464844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,48,4,128,1,float16,float16,0,36.70380859375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,48,8,128,1,float16,float16,0,37.821893310546876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,48,1,128,1,float16,float16,0,18.43203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,48,48,128,1,float16,float16,0,16.311311340332033
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,48,48,128,1,float16,float16,0,19.593116760253906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,48,2,128,1,float16,float16,0,18.04913330078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,48,48,128,1,float16,float16,0,9.528860473632813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,48,1,128,1,float16,float16,0,8.44168930053711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,48,4,128,1,float16,float16,0,18.520710754394532
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,48,2,128,1,float16,float16,0,8.787838745117188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,48,48,128,1,float16,float16,0,34.13717956542969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,48,8,128,1,float16,float16,0,19.13914794921875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,48,4,128,1,float16,float16,0,8.84622573852539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,48,8,128,1,float16,float16,0,9.169478607177734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,48,48,128,1,float16,float16,0,40.27308349609375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,48,1,128,1,float16,float16,0,24.98230743408203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,48,48,128,1,float16,float16,0,68.29867553710938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,48,48,128,1,float16,float16,0,27.886929321289063
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,48,2,128,1,float16,float16,0,24.89037170410156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,48,48,128,1,float16,float16,0,13.905552673339844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,48,1,128,1,float16,float16,0,51.34352416992188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,48,1,128,1,float16,float16,0,12.679289245605469
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,48,2,128,1,float16,float16,0,50.831243896484374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,48,4,128,1,float16,float16,0,51.9778076171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,48,8,128,1,float16,float16,0,54.058740234375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,48,4,128,1,float16,float16,0,26.205474853515625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,48,8,128,1,float16,float16,0,26.319625854492188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,48,2,128,1,float16,float16,0,12.273146820068359
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,48,48,128,1,float16,float16,0,6.763673400878906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,48,4,128,1,float16,float16,0,13.2945068359375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,48,1,128,1,float16,float16,0,6.090691375732422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,48,8,128,1,float16,float16,0,13.004179382324219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,48,2,128,1,float16,float16,0,5.993560028076172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,48,4,128,1,float16,float16,0,5.992244720458984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,48,8,128,1,float16,float16,0,6.146238327026367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,48,1,128,1,float16,float16,0,32.70552062988281
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,48,2,128,1,float16,float16,0,32.87559814453125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,48,4,128,1,float16,float16,0,33.07631225585938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,48,48,128,1,float16,float16,0,38.14784545898438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,48,1,128,1,float16,float16,0,15.832101440429687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,48,48,128,1,float16,float16,0,18.525614929199218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,48,2,128,1,float16,float16,0,16.063742065429686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,48,1,128,1,float16,float16,0,64.999755859375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,48,2,128,1,float16,float16,0,66.59374389648437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,48,48,128,1,float16,float16,0,9.141862487792968
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,48,4,128,1,float16,float16,0,68.837109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,48,8,128,1,float16,float16,0,34.24209899902344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,48,4,128,1,float16,float16,0,16.685879516601563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,48,8,128,1,float16,float16,0,70.68561401367188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,48,1,128,1,float16,float16,0,7.93109130859375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,48,8,128,1,float16,float16,0,16.998687744140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,48,2,128,1,float16,float16,0,8.069786834716798
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,48,48,128,1,float16,float16,0,4.618142318725586
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,48,1,128,1,float16,float16,0,3.9106304168701174
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,48,4,128,1,float16,float16,0,8.394931030273437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,48,4,128,1,float16,float16,0,4.003726577758789
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,48,2,128,1,float16,float16,0,3.953438568115234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,48,8,128,1,float16,float16,0,8.568465423583984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,48,8,128,1,float16,float16,0,3.960647964477539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,48,1,128,1,float16,float16,0,19.01688232421875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,48,2,128,1,float16,float16,0,19.231382751464842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,48,4,128,1,float16,float16,0,19.16313934326172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,48,48,128,1,float16,float16,0,22.114544677734376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,48,1,128,1,float16,float16,0,9.011119842529297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,48,2,128,1,float16,float16,0,9.27724609375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,48,48,128,1,float16,float16,0,11.225230407714843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,48,1,128,1,float16,float16,0,37.99629211425781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,48,2,128,1,float16,float16,0,37.983517456054685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,48,4,128,1,float16,float16,0,38.499365234375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,48,48,128,1,float16,float16,0,5.318632125854492
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,48,8,128,1,float16,float16,0,39.920693969726564
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,48,8,128,1,float16,float16,0,20.713018798828124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,48,4,128,1,float16,float16,0,9.602130889892578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,48,1,128,1,float16,float16,0,4.400521469116211
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,48,2,128,1,float16,float16,0,4.570094299316406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,48,8,128,1,float16,float16,0,10.069550323486329
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,48,4,128,1,float16,float16,0,4.464495849609375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,48,48,128,1,float16,float16,0,2.620716857910156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,48,2,128,1,float16,float16,0,2.2064544677734377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,48,1,128,1,float16,float16,0,2.1518016815185548
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,48,4,128,1,float16,float16,0,2.138505554199219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,48,8,128,1,float16,float16,0,4.718016052246094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,48,8,128,1,float16,float16,0,2.190398406982422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,48,1,128,1,float16,float16,0,17.172503662109374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,48,4,128,1,float16,float16,0,17.699491882324217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,48,2,128,1,float16,float16,0,18.04859619140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,48,48,128,1,float16,float16,0,20.355256652832033
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,48,1,128,1,float16,float16,0,8.748320007324219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,48,48,128,1,float16,float16,0,10.352974700927735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,48,2,128,1,float16,float16,0,8.898505401611327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,48,2,128,1,float16,float16,0,34.44404602050781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,48,1,128,1,float16,float16,0,34.56326599121094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,48,4,128,1,float16,float16,0,36.28424987792969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,48,48,128,1,float16,float16,0,5.428246307373047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,48,8,128,1,float16,float16,0,19.116064453125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,48,8,128,1,float16,float16,0,37.22308349609375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,48,4,128,1,float16,float16,0,9.03975830078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,48,1,128,1,float16,float16,0,4.195566558837891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,48,2,128,1,float16,float16,0,4.2372478485107425
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,48,8,128,1,float16,float16,0,9.036721801757812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,48,2,128,1,float16,float16,0,2.0146608352661133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,48,1,128,1,float16,float16,0,1.8727327346801759
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,48,48,128,1,float16,float16,0,2.4908287048339846
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,48,4,128,1,float16,float16,0,4.230204772949219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,48,1,128,1,float16,float16,0,1.0744815826416017
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,48,2,128,1,float16,float16,0,1.0423888206481933
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,48,48,128,1,float16,float16,0,1.3188672065734863
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,48,4,128,1,float16,float16,0,0.9491344451904297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,48,8,128,1,float16,float16,0,4.363283157348633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,48,8,128,1,float16,float16,0,2.165937614440918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,48,8,128,1,float16,float16,0,1.0546688079833983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,48,4,128,1,float16,float16,0,1.9508880615234374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,48,1,128,1,float16,float16,0,9.672566223144532
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,48,2,128,1,float16,float16,0,9.814107513427734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,48,4,128,1,float16,float16,0,9.612118530273438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,48,48,128,1,float16,float16,0,13.136265563964844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,48,48,128,1,float16,float16,0,6.110200119018555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,48,1,128,1,float16,float16,0,4.6102031707763675
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,48,2,128,1,float16,float16,0,4.847542572021484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,48,2,128,1,float16,float16,0,19.958404541015625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,48,1,128,1,float16,float16,0,20.41163330078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,48,4,128,1,float16,float16,0,21.157777404785158
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,48,8,128,1,float16,float16,0,21.394525146484376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,48,8,128,1,float16,float16,0,10.59200668334961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,48,1,128,1,float16,float16,0,2.276457595825195
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,48,2,128,1,float16,float16,0,2.286289596557617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,48,48,128,1,float16,float16,0,3.138191986083984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,48,4,128,1,float16,float16,0,4.954872131347656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,48,8,128,1,float16,float16,0,4.879803085327149
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,48,1,128,1,float16,float16,0,1.218684768676758
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,48,48,128,1,float16,float16,0,1.5415871620178223
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,48,4,128,1,float16,float16,0,2.366134452819824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,48,2,128,1,float16,float16,0,1.0776816368103028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,48,4,128,1,float16,float16,0,1.146667194366455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,48,8,128,1,float16,float16,0,1.189465618133545
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,48,48,128,1,float16,float16,0,0.8012607574462891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,48,1,128,1,float16,float16,0,0.56321120262146
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,48,2,128,1,float16,float16,0,0.6855167865753173
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,48,4,128,1,float16,float16,0,0.6340335845947266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,48,8,128,1,float16,float16,0,0.6018335819244385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,48,8,128,1,float16,float16,0,2.392975997924805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,48,1,128,1,float16,float16,0,9.539857482910156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,48,2,128,1,float16,float16,0,9.082561492919922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,48,48,128,1,float16,float16,0,13.240658569335938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,48,4,128,1,float16,float16,0,9.293128204345702
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,48,1,128,1,float16,float16,0,18.407255554199217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,48,48,128,1,float16,float16,0,6.364235305786133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,48,1,128,1,float16,float16,0,4.457633590698242
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,48,2,128,1,float16,float16,0,19.840988159179688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,48,4,128,1,float16,float16,0,20.140855407714845
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,48,8,128,1,float16,float16,0,10.215825653076172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,48,8,128,1,float16,float16,0,20.803782653808593
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,48,2,128,1,float16,float16,0,4.376811218261719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,48,1,128,1,float16,float16,0,2.067987251281738
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,48,2,128,1,float16,float16,0,2.0621055603027343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,48,48,128,1,float16,float16,0,3.1696319580078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,48,48,128,1,float16,float16,0,1.6341279983520507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,48,4,128,1,float16,float16,0,2.1167007446289063
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,48,4,128,1,float16,float16,0,4.669185638427734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,48,8,128,1,float16,float16,0,2.1560144424438477
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,48,1,128,1,float16,float16,0,1.2095328330993653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,48,2,128,1,float16,float16,0,1.0794063568115235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,48,48,128,1,float16,float16,0,0.854190444946289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,48,4,128,1,float16,float16,0,1.008148765563965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,48,8,128,1,float16,float16,0,1.030185604095459
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,48,1,128,1,float16,float16,0,0.6291903972625732
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,48,8,128,1,float16,float16,0,4.917136001586914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,48,2,128,1,float16,float16,0,0.5789343833923339
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,48,1,128,1,float16,float16,0,0.28387200832366943
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,48,2,128,1,float16,float16,0,0.3062623977661133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,48,48,128,1,float16,float16,0,0.4267695903778076
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,48,4,128,1,float16,float16,0,0.2970815896987915
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,48,8,128,1,float16,float16,0,0.3619632005691528
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,48,4,128,1,float16,float16,0,0.6028880119323731
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,48,8,128,1,float16,float16,0,0.5605584144592285
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,48,1,128,1,float16,float16,0,4.917964935302734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,48,2,128,1,float16,float16,0,5.115969467163086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,48,48,128,1,float16,float16,0,8.142140960693359
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,48,4,128,1,float16,float16,0,5.323748779296875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,48,1,128,1,float16,float16,0,10.904468536376953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,48,2,128,1,float16,float16,0,11.591854095458984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,48,4,128,1,float16,float16,0,11.474988555908203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,48,8,128,1,float16,float16,0,5.816612625122071
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,48,8,128,1,float16,float16,0,11.925023651123047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,48,1,128,1,float16,float16,0,2.291641616821289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,48,2,128,1,float16,float16,0,2.540727996826172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,48,1,128,1,float16,float16,0,1.2647151947021484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,48,48,128,1,float16,float16,0,4.121511840820313
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,48,4,128,1,float16,float16,0,1.2829919815063477
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,48,4,128,1,float16,float16,0,2.7747392654418945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,48,2,128,1,float16,float16,0,1.3870047569274901
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,48,8,128,1,float16,float16,0,1.3569055557250977
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,48,48,128,1,float16,float16,0,2.1437536239624024
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,48,2,128,1,float16,float16,0,0.6481328010559082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,48,8,128,1,float16,float16,0,2.9480255126953123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,48,1,128,1,float16,float16,0,0.3045727968215942
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,48,48,128,1,float16,float16,0,1.086961555480957
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,48,48,128,1,float16,float16,0,0.5443007946014404
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,48,1,128,1,float16,float16,0,0.5882031917572021
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,48,2,128,1,float16,float16,0,0.322323203086853
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,48,8,128,1,float16,float16,0,0.6325615882873535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,48,8,128,1,float16,float16,0,0.32763519287109377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,48,1,128,1,float16,float16,0,0.19260799884796143
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,48,48,128,1,float16,float16,0,0.27698559761047364
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,48,4,128,1,float16,float16,0,0.4110400199890137
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,48,4,128,1,float16,float16,0,0.1799999952316284
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,48,2,128,1,float16,float16,0,0.17352800369262694
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,48,8,128,1,float16,float16,0,0.18755040168762208
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,48,4,128,1,float16,float16,0,0.716483211517334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,48,1,128,1,float16,float16,0,5.089311981201172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,48,2,128,1,float16,float16,0,5.033924865722656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,48,4,128,1,float16,float16,0,5.432470321655273
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,48,1,128,1,float16,float16,0,10.808236694335937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,48,48,128,1,float16,float16,0,9.798044586181641
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,48,2,128,1,float16,float16,0,10.918206024169923
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,48,48,128,1,float16,float16,0,4.983295822143555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,48,4,128,1,float16,float16,0,11.420708465576173
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,48,8,128,1,float16,float16,0,5.826315307617188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,48,1,128,1,float16,float16,0,2.3319280624389647
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,48,8,128,1,float16,float16,0,12.607170867919923
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,48,1,128,1,float16,float16,0,1.1992303848266601
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,48,2,128,1,float16,float16,0,1.1942655563354492
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,48,2,128,1,float16,float16,0,2.4919904708862304
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,48,4,128,1,float16,float16,0,1.2811856269836426
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,48,4,128,1,float16,float16,0,2.3646175384521486
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,48,48,128,1,float16,float16,0,2.5475807189941406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,48,8,128,1,float16,float16,0,2.683345603942871
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,48,2,128,1,float16,float16,0,0.6437888145446777
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,48,1,128,1,float16,float16,0,0.6439919948577881
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,48,4,128,1,float16,float16,0,0.6886767864227294
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,48,8,128,1,float16,float16,0,1.3384063720703125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,48,48,128,1,float16,float16,0,1.2752623558044434
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,48,48,128,1,float16,float16,0,0.6687583923339844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,48,2,128,1,float16,float16,0,0.3342463970184326
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,48,8,128,1,float16,float16,0,0.651196813583374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,48,1,128,1,float16,float16,0,0.3262752056121826
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,48,2,128,1,float16,float16,0,0.17475199699401855
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,48,8,128,1,float16,float16,0,0.35227360725402834
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,48,4,128,1,float16,float16,0,0.15820800065994262
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,48,8,128,1,float16,float16,0,0.17134879827499389
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,48,1,128,1,float16,float16,0,0.17660000324249267
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,48,48,128,1,float16,float16,0,0.11240479946136475
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,48,2,128,1,float16,float16,0,0.09243680238723755
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,48,4,128,1,float16,float16,0,0.10321279764175414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,48,1,128,1,float16,float16,0,0.09899680018424988
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,48,8,128,1,float16,float16,0,0.1022752046585083
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,48,4,128,1,float16,float16,0,0.34358079433441163
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,48,48,128,1,float16,float16,0,0.30885279178619385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,48,1,128,1,float16,float16,0,2.977235221862793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,48,2,128,1,float16,float16,0,6.383975982666016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,48,1,128,1,float16,float16,0,6.552228546142578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,48,2,128,1,float16,float16,0,3.0007776260375976
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,48,4,128,1,float16,float16,0,7.028601837158203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,48,8,128,1,float16,float16,0,7.746094512939453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,48,1,128,1,float16,float16,0,1.656332778930664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,48,2,128,1,float16,float16,0,1.5109503746032715
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,48,4,128,1,float16,float16,0,1.4836015701293945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,48,4,128,1,float16,float16,0,3.4242321014404298
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,48,48,128,1,float16,float16,0,9.718023681640625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,48,1,128,1,float16,float16,0,0.7790751934051514
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,48,8,128,1,float16,float16,0,1.9591136932373048
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,48,8,128,1,float16,float16,0,3.8734527587890626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,48,2,128,1,float16,float16,0,0.7682928085327149
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,48,4,128,1,float16,float16,0,0.8461664199829102
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,48,48,128,1,float16,float16,0,2.492184066772461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,48,4,128,1,float16,float16,0,0.40839362144470215
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,48,2,128,1,float16,float16,0,0.4302063941955566
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,48,1,128,1,float16,float16,0,0.3980815887451172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,48,48,128,1,float16,float16,0,4.979164886474609
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,48,48,128,1,float16,float16,0,1.2686639785766602
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,48,1,128,1,float16,float16,0,0.19933439493179322
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,48,8,128,1,float16,float16,0,0.45064477920532225
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,48,8,128,1,float16,float16,0,1.0263855934143067
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,48,2,128,1,float16,float16,0,0.2142672061920166
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,48,4,128,1,float16,float16,0,0.2192431926727295
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,48,2,128,1,float16,float16,0,0.09703999757766724
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,48,48,128,1,float16,float16,0,0.6754720211029053
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,48,1,128,1,float16,float16,0,0.09755839705467224
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,48,48,128,1,float16,float16,0,0.24374721050262452
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,48,8,128,1,float16,float16,0,0.10344480276107788
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,48,1,128,1,float16,float16,0,0.05565919876098633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,48,4,128,1,float16,float16,0,0.09609760046005249
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,48,48,128,1,float16,float16,0,0.06690559983253479
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,48,4,128,1,float16,float16,0,0.05893599987030029
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,48,2,128,1,float16,float16,0,0.053288000822067264
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,48,1,128,1,float16,float16,0,0.03680320084095001
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,48,48,128,1,float16,float16,0,0.04328800141811371
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,48,8,128,1,float16,float16,0,0.038134399056434634
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,48,4,128,1,float16,float16,0,0.037355199456214905
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,48,2,128,1,float16,float16,0,0.035780799388885495
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,48,8,128,1,float16,float16,0,0.055308800935745236
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,48,1,128,1,float16,float16,0,1.3774543762207032
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,48,2,128,1,float16,float16,0,1.4227392196655273
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,48,1,128,1,float16,float16,0,2.731240081787109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,48,2,128,1,float16,float16,0,2.8443456649780274
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,48,8,128,1,float16,float16,0,0.23434081077575683
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,48,4,128,1,float16,float16,0,3.186840057373047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,48,4,128,1,float16,float16,0,1.597332763671875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,48,2,128,1,float16,float16,0,0.6810832023620605
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,48,1,128,1,float16,float16,0,0.7460400104522705
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,48,8,128,1,float16,float16,0,3.821780776977539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,48,8,128,1,float16,float16,0,1.9291040420532226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,48,4,128,1,float16,float16,0,0.8539024353027344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,48,1,128,1,float16,float16,0,0.3275264024734497
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,48,8,128,1,float16,float16,0,0.9585696220397949
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,48,48,128,1,float16,float16,0,4.90190544128418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,48,1,128,1,float16,float16,0,0.13191039562225343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,48,48,128,1,float16,float16,0,0.6339360237121582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,48,48,128,1,float16,float16,0,1.2625760078430175
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,48,4,128,1,float16,float16,0,0.16703039407730103
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,48,48,128,1,float16,float16,0,2.511027145385742
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,48,8,128,1,float16,float16,0,0.18963040113449098
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,48,2,128,1,float16,float16,0,0.16550400257110595
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,48,1,128,1,float16,float16,0,0.06456639766693115
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,48,4,128,1,float16,float16,0,0.06415039896965027
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,48,48,128,1,float16,float16,0,0.04626399874687195
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,48,48,128,1,float16,float16,0,0.2181999921798706
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,48,2,128,1,float16,float16,0,0.06331679821014405
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,48,8,128,1,float16,float16,0,0.0663968026638031
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,48,1,128,1,float16,float16,0,0.038863998651504514
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,48,2,128,1,float16,float16,0,0.037703999876976015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,48,4,128,1,float16,float16,0,0.03705599904060364
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,48,2,128,1,float16,float16,0,0.024382400512695312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,48,1,128,1,float16,float16,0,0.024700799584388734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,48,8,128,1,float16,float16,0,0.038108798861503604
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,48,4,128,1,float16,float16,0,0.024820800125598907
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,48,8,128,1,float16,float16,0,0.025879999995231627
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,48,48,128,1,float16,float16,0,0.01911199986934662
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,48,1,128,1,float16,float16,0,0.016622400283813475
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,48,2,128,1,float16,float16,0,0.016935999691486358
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,48,4,128,1,float16,float16,0,0.016758400201797485
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,48,8,128,1,float16,float16,0,0.017100800573825837
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,48,1,128,1,float16,float16,0,1.3176112174987793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,48,2,128,1,float16,float16,0,0.36763839721679686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,48,2,128,1,float16,float16,0,1.397753620147705
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,48,4,128,1,float16,float16,0,1.5652496337890625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,48,8,128,1,float16,float16,0,0.4469247817993164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,48,1,128,1,float16,float16,0,0.6808735847473144
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,48,2,128,1,float16,float16,0,0.6871808052062989
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,48,4,128,1,float16,float16,0,0.7639647960662842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,48,48,128,1,float16,float16,0,0.029334399104118346
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,48,1,128,1,float16,float16,0,0.32111520767211915
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,48,8,128,1,float16,float16,0,0.9248543739318847
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,48,8,128,1,float16,float16,0,1.9113935470581054
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,48,4,128,1,float16,float16,0,0.3723504066467285
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,48,48,128,1,float16,float16,0,1.2560799598693848
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,48,48,128,1,float16,float16,0,2.481550407409668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,48,1,128,1,float16,float16,0,0.10932960510253906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,48,8,128,1,float16,float16,0,0.4528351783752441
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,48,2,128,1,float16,float16,0,0.12867200374603271
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,48,48,128,1,float16,float16,0,0.6361567974090576
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,48,8,128,1,float16,float16,0,0.1864591956138611
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,48,1,128,1,float16,float16,0,0.049414399266242984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,48,48,128,1,float16,float16,0,0.20917279720306398
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,48,2,128,1,float16,float16,0,0.04986400008201599
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,48,4,128,1,float16,float16,0,0.05101439952850342
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,48,8,128,1,float16,float16,0,0.05235360264778137
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,48,48,128,1,float16,float16,0,0.039211198687553406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,48,4,128,1,float16,float16,0,0.17511839866638185
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,48,1,128,1,float16,float16,0,0.029139199852943422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,48,4,128,1,float16,float16,0,0.38441441059112547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,48,2,128,1,float16,float16,0,0.029148799180984498
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,48,4,128,1,float16,float16,0,0.029947200417518617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,48,8,128,1,float16,float16,0,0.03078719973564148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,48,1,128,1,float16,float16,0,0.020712000131607056
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,48,48,128,1,float16,float16,0,0.025854399800300597
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,48,4,128,1,float16,float16,0,0.021087999641895293
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,48,8,128,1,float16,float16,0,0.02144480049610138
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,48,1,128,1,float16,float16,0,0.013152000308036805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,48,2,128,1,float16,float16,0,0.013158400356769562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,48,2,128,1,float16,float16,0,0.021137599647045136
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,48,8,128,1,float16,float16,0,0.013305599987506866
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,48,48,128,1,float16,float16,0,0.01356160044670105
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,48,1,128,1,float16,float16,0,0.012193600088357926
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,48,4,128,1,float16,float16,0,0.012214399874210358
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,48,8,128,1,float16,float16,0,0.012347199767827988
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,48,2,128,1,float16,float16,0,0.012355200201272964
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,48,1,128,1,float16,float16,0,0.6641615867614746
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,48,2,128,1,float16,float16,0,0.6890160083770752
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,48,4,128,1,float16,float16,0,0.7673183917999268
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,48,8,128,1,float16,float16,0,0.9370767593383789
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,48,2,128,1,float16,float16,0,0.3369472026824951
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,48,1,128,1,float16,float16,0,0.3237344026565552
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,48,2,128,1,float16,float16,0,0.3381439924240112
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,48,1,128,1,float16,float16,0,0.11084320545196533
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,48,4,128,1,float16,float16,0,0.37310400009155276
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,48,8,128,1,float16,float16,0,0.4468128204345703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,48,48,128,1,float16,float16,0,0.6212287902832031
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,48,48,128,1,float16,float16,0,1.2484304428100585
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,48,4,128,1,float16,float16,0,0.15699039697647094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,48,8,128,1,float16,float16,0,0.18731520175933838
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,48,48,128,1,float16,float16,0,0.21615519523620605
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,48,2,128,1,float16,float16,0,0.04872480034828186
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,48,4,128,1,float16,float16,0,0.04988479912281037
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,48,48,128,1,float16,float16,0,0.05295040011405945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,48,8,128,1,float16,float16,0,0.051158398389816284
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,48,1,128,1,float16,float16,0,0.027822399139404298
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,48,2,128,1,float16,float16,0,0.02804799973964691
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,48,48,128,1,float16,float16,0,0.02961440086364746
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,48,48,128,1,float16,float16,0,0.015432000160217285
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,48,2,128,1,float16,float16,0,0.02006720006465912
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,48,4,128,1,float16,float16,0,0.020351999998092653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,48,4,128,1,float16,float16,0,0.013246400654315949
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,48,8,128,1,float16,float16,0,0.020499199628829956
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,48,48,128,1,float16,float16,0,0.02059040069580078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,48,1,128,1,float16,float16,0,0.0127920001745224
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,48,2,128,1,float16,float16,0,0.012702399492263794
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,48,4,128,1,float16,float16,0,0.012929600477218629
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,48,8,128,1,float16,float16,0,0.013279999792575835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,48,48,128,1,float16,float16,0,0.013104000687599182
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,48,1,128,1,float16,float16,0,0.011931200325489045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,48,2,128,1,float16,float16,0,0.011987199634313583
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,48,4,128,1,float16,float16,0,0.012044800072908401
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,48,8,128,1,float16,float16,0,0.01218079999089241
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,48,48,128,1,float16,float16,0,0.012470400333404541
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,48,1,128,1,float16,float16,0,0.01173280030488968
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,48,2,128,1,float16,float16,0,0.011851199716329575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,48,4,128,1,float16,float16,0,0.011767999827861786
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,48,1,128,1,float16,float16,0,0.3224287986755371
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,48,2,128,1,float16,float16,0,0.33726561069488525
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,48,4,128,1,float16,float16,0,0.3666320085525513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,48,8,128,1,float16,float16,0,0.4274288177490234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,48,2,128,1,float16,float16,0,0.13270080089569092
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,48,1,128,1,float16,float16,0,0.11119519472122193
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,48,48,128,1,float16,float16,0,0.5734384059906006
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,48,2,128,1,float16,float16,0,0.13347840309143066
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,48,1,128,1,float16,float16,0,0.048316800594329835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,48,8,128,1,float16,float16,0,0.19117759466171264
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,48,4,128,1,float16,float16,0,0.16580159664154054
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,48,4,128,1,float16,float16,0,0.028651198744773863
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,48,8,128,1,float16,float16,0,0.02969920039176941
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,48,4,128,1,float16,float16,0,0.049527999758720395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,48,2,128,1,float16,float16,0,0.048879998922348025
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,48,1,128,1,float16,float16,0,0.04867520034313202
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,48,48,128,1,float16,float16,0,0.2719248056411743
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,48,2,128,1,float16,float16,0,0.028265601396560668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,48,48,128,1,float16,float16,0,0.08070240020751954
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,48,8,128,1,float16,float16,0,0.03443520069122315
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,48,48,128,1,float16,float16,0,0.045771199464797976
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,48,4,128,1,float16,float16,0,0.028516799211502075
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,48,1,128,1,float16,float16,0,0.01967200040817261
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,48,2,128,1,float16,float16,0,0.019888000190258028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,48,4,128,1,float16,float16,0,0.019897599518299103
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,48,8,128,1,float16,float16,0,0.02024800032377243
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,48,48,128,1,float16,float16,0,0.02568640112876892
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,48,2,128,1,float16,float16,0,0.012729600071907043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,48,8,128,1,float16,float16,0,0.011856000125408172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,48,8,128,1,float16,float16,0,0.012894399464130402
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,48,48,128,1,float16,float16,0,0.018348799645900728
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,48,1,128,1,float16,float16,0,0.011787199974060058
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,48,2,128,1,float16,float16,0,0.011875200271606445
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,48,4,128,1,float16,float16,0,0.011966399848461151
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,48,8,128,1,float16,float16,0,0.012043199688196182
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,48,48,128,1,float16,float16,0,0.012195199728012085
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,48,1,128,1,float16,float16,0,0.011660800129175187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,48,4,128,1,float16,float16,0,0.011718399822711945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,48,2,128,1,float16,float16,0,0.011699199676513672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,48,48,128,1,float16,float16,0,0.011980800330638886
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,48,1,128,1,float16,float16,0,0.011627200245857238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,48,2,128,1,float16,float16,0,0.011707200109958649
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,48,4,128,1,float16,float16,0,0.011687999963760376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,48,8,128,1,float16,float16,0,0.01168000027537346
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,48,1,128,1,float16,float16,0,0.11283040046691895
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,48,2,128,1,float16,float16,0,0.14226880073547363
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,48,4,128,1,float16,float16,0,0.1685472011566162
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,48,1,128,1,float16,float16,0,0.020033599436283113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,48,8,128,1,float16,float16,0,0.19441599845886232
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,48,8,128,1,float16,float16,0,0.06163039803504944
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,48,1,128,1,float16,float16,0,0.028043198585510253
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,48,1,128,1,float16,float16,0,0.04908480048179627
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,48,4,128,1,float16,float16,0,0.06005439758300781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,48,48,128,1,float16,float16,0,0.39091200828552247
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,48,2,128,1,float16,float16,0,0.049692800641059874
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,48,8,128,1,float16,float16,0,0.06186400055885315
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,48,1,128,1,float16,float16,0,0.012647999823093415
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,48,2,128,1,float16,float16,0,0.028721600770950317
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,48,48,128,1,float16,float16,0,0.14387520551681518
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,48,1,128,1,float16,float16,0,0.028358399868011475
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,48,4,128,1,float16,float16,0,0.03386560082435608
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,48,4,128,1,float16,float16,0,0.012851199507713318
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,48,4,128,1,float16,float16,0,0.020094400644302367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,48,1,128,1,float16,float16,0,0.0199535995721817
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,48,8,128,1,float16,float16,0,0.034836798906326294
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,48,2,128,1,float16,float16,0,0.020267200469970704
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,48,8,128,1,float16,float16,0,0.020635199546813966
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,48,48,128,1,float16,float16,0,0.04276799857616424
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,48,4,128,1,float16,float16,0,0.012894399464130402
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,48,2,128,1,float16,float16,0,0.012945599853992462
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,48,48,128,1,float16,float16,0,0.02425439953804016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,48,8,128,1,float16,float16,0,0.013128000497817992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,48,1,128,1,float16,float16,0,0.011860799789428712
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,48,4,128,1,float16,float16,0,0.01189119964838028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,48,8,128,1,float16,float16,0,0.011751999706029892
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,48,2,128,1,float16,float16,0,0.012055999785661697
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,48,8,128,1,float16,float16,0,0.012140800058841706
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,48,48,128,1,float16,float16,0,0.018012799322605133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,48,1,128,1,float16,float16,0,0.011574400216341018
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,48,2,128,1,float16,float16,0,0.011928000301122666
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,48,4,128,1,float16,float16,0,0.011903999745845795
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,48,1,128,1,float16,float16,0,0.011590400338172912
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,48,48,128,1,float16,float16,0,0.012230399996042252
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,48,2,128,1,float16,float16,0,0.011990399658679962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,48,8,128,1,float16,float16,0,0.011417599767446518
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,48,8,128,1,float16,float16,0,0.011764799803495407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,48,4,128,1,float16,float16,0,0.011929599940776825
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,48,48,128,1,float16,float16,0,0.012067200243473053
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,48,2,128,1,float16,float16,0,0.011587200313806533
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,48,1,128,1,float16,float16,0,0.01185920014977455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,48,1,128,1,float16,float16,0,0.02028000056743622
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,48,2,128,1,float16,float16,0,0.031632000207901
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,48,8,128,1,float16,float16,0,0.011675199866294861
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,48,4,128,1,float16,float16,0,0.05537599921226501
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,48,1,128,1,float16,float16,0,0.013167999684810638
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,48,8,128,1,float16,float16,0,0.09717440009117126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,48,2,128,1,float16,float16,0,0.018838399648666383
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,48,48,128,1,float16,float16,0,0.26209440231323244
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,48,4,128,1,float16,float16,0,0.030695998668670656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,48,8,128,1,float16,float16,0,0.05448799729347229
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,48,48,128,1,float16,float16,0,0.134441602230072
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,48,4,128,1,float16,float16,0,0.01788319945335388
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,48,2,128,1,float16,float16,0,0.012329600006341934
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,48,48,128,1,float16,float16,0,0.07045599818229675
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,48,1,128,1,float16,float16,0,0.011657600104808808
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,48,2,128,1,float16,float16,0,0.01170239970088005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,48,4,128,1,float16,float16,0,0.011798399686813354
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,48,8,128,1,float16,float16,0,0.01767839938402176
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,48,48,128,1,float16,float16,0,0.04116640090942383
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,48,1,128,1,float16,float16,0,0.011121600121259689
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,48,2,128,1,float16,float16,0,0.011147200316190719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,48,4,128,1,float16,float16,0,0.011209599673748016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,48,48,128,1,float16,float16,0,0.07511519789695739
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,48,1,128,1,float16,float16,0,0.012579199671745301
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,48,8,128,1,float16,float16,0,0.01143840029835701
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,48,1,128,1,float16,float16,0,0.011027199774980545
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,48,48,128,1,float16,float16,0,0.023844799399375914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,48,2,128,1,float16,float16,0,0.011219199746847153
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,48,8,128,1,float16,float16,0,0.011052799969911575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,48,48,128,1,float16,float16,0,0.017927999794483184
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,48,1,128,1,float16,float16,0,0.011748799681663513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,48,8,128,1,float16,float16,0,0.011820799857378005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,48,4,128,1,float16,float16,0,0.01117599979043007
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,48,48,128,1,float16,float16,0,0.012187200039625168
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,48,1,128,1,float16,float16,0,0.011156799644231797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,48,2,128,1,float16,float16,0,0.01111840009689331
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,48,4,128,1,float16,float16,0,0.011126399785280228
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,48,4,128,1,float16,float16,0,0.011902400106191636
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,48,8,128,1,float16,float16,0,0.011134400218725204
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,48,48,128,1,float16,float16,0,0.012064000219106674
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,48,1,128,1,float16,float16,0,0.011204800009727478
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,48,2,128,1,float16,float16,0,0.011095999926328658
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,48,4,128,1,float16,float16,0,0.01104160025715828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,48,8,128,1,float16,float16,0,0.011228799819946289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,48,1,128,1,float16,float16,0,0.011868800222873687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,48,8,128,1,float16,float16,0,0.030083200335502623
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,48,4,128,1,float16,float16,0,0.011073599755764007
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,48,2,128,1,float16,float16,0,0.010924799740314484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,40,2,128,1,float16,float16,0,52.01539306640625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,40,1,128,1,float16,float16,0,53.350830078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,40,4,128,1,float16,float16,0,54.264154052734376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,40,1,128,1,float16,float16,0,26.407736206054686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,40,1,128,1,float16,float16,0,106.07672119140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,40,4,128,1,float16,float16,0,107.33232421875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,40,8,128,1,float16,float16,0,108.0630859375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,40,2,128,1,float16,float16,0,108.96256103515626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,40,8,128,1,float16,float16,0,53.88859252929687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,40,2,128,1,float16,float16,0,26.514401245117188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,40,2,128,1,float16,float16,0,13.083531188964844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,40,1,128,1,float16,float16,0,13.657382202148437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,40,4,128,1,float16,float16,0,13.903919982910157
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,40,4,128,1,float16,float16,0,27.691561889648437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,40,8,128,1,float16,float16,0,27.512753295898438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,40,8,128,1,float16,float16,0,13.612059020996094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,40,1,128,1,float16,float16,0,62.09541625976563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,40,2,128,1,float16,float16,0,61.14515380859375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,40,4,128,1,float16,float16,0,62.42366943359375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,40,8,128,1,float16,float16,0,62.44590454101562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,40,1,128,1,float16,float16,0,31.06212463378906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,40,2,128,1,float16,float16,0,30.777984619140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,40,4,128,1,float16,float16,0,31.261233520507812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,40,8,128,1,float16,float16,0,31.1206298828125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,40,1,128,1,float16,float16,0,15.025032043457031
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,40,2,128,1,float16,float16,0,15.332568359375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,40,4,128,1,float16,float16,0,15.4849609375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,40,1,128,1,float16,float16,0,7.23028793334961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,40,8,128,1,float16,float16,0,15.7716552734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,40,2,128,1,float16,float16,0,7.319857788085938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,40,4,128,1,float16,float16,0,7.526161956787109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,40,8,128,1,float16,float16,0,7.7554771423339846
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,40,40,128,1,float16,float16,0,13.656150817871094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,40,40,128,1,float16,float16,0,7.983493041992188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,40,40,128,1,float16,float16,0,16.4415283203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,40,40,128,1,float16,float16,0,27.983038330078124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,40,40,128,1,float16,float16,0,32.78877258300781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,40,1,128,1,float16,float16,0,42.49706115722656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,40,40,128,1,float16,float16,0,24.253938293457033
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,40,2,128,1,float16,float16,0,42.469174194335935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,40,1,128,1,float16,float16,0,21.763047790527345
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,40,40,128,1,float16,float16,0,57.1584716796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,40,2,128,1,float16,float16,0,21.66106719970703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,40,4,128,1,float16,float16,0,42.84505004882813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,40,40,128,1,float16,float16,0,11.73602752685547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,40,8,128,1,float16,float16,0,44.18030700683594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,40,1,128,1,float16,float16,0,10.136443328857421
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,40,4,128,1,float16,float16,0,21.922010803222655
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,40,2,128,1,float16,float16,0,10.232259368896484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,40,40,128,1,float16,float16,0,5.718172836303711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,40,1,128,1,float16,float16,0,5.010009765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,40,4,128,1,float16,float16,0,10.465102386474609
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,40,8,128,1,float16,float16,0,10.450920104980469
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,40,2,128,1,float16,float16,0,5.114767837524414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,40,8,128,1,float16,float16,0,22.44103088378906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,40,4,128,1,float16,float16,0,5.2063438415527346
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,40,8,128,1,float16,float16,0,5.2086433410644535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,40,2,128,1,float16,float16,0,27.668563842773438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,40,1,128,1,float16,float16,0,28.580307006835938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,40,4,128,1,float16,float16,0,28.214358520507812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,40,40,128,1,float16,float16,0,32.08157653808594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,40,40,128,1,float16,float16,0,15.894058227539062
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,40,1,128,1,float16,float16,0,13.361880493164062
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,40,2,128,1,float16,float16,0,13.744091796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,40,1,128,1,float16,float16,0,54.810150146484375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,40,2,128,1,float16,float16,0,55.18243408203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,40,40,128,1,float16,float16,0,7.930480194091797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,40,8,128,1,float16,float16,0,56.393505859375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,40,4,128,1,float16,float16,0,57.506951904296876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,40,8,128,1,float16,float16,0,28.771148681640625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,40,8,128,1,float16,float16,0,14.120831298828126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,40,4,128,1,float16,float16,0,14.79541015625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,40,1,128,1,float16,float16,0,6.653070068359375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,40,40,128,1,float16,float16,0,3.5211711883544923
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,40,2,128,1,float16,float16,0,6.696006774902344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,40,1,128,1,float16,float16,0,3.1538400650024414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,40,2,128,1,float16,float16,0,3.2242961883544923
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,40,4,128,1,float16,float16,0,6.640116882324219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,40,4,128,1,float16,float16,0,3.2673198699951174
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,40,8,128,1,float16,float16,0,7.097220611572266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,40,8,128,1,float16,float16,0,3.36024169921875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,40,1,128,1,float16,float16,0,15.022361755371094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,40,4,128,1,float16,float16,0,15.962144470214843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,40,2,128,1,float16,float16,0,15.73079833984375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,40,40,128,1,float16,float16,0,18.650816345214842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,40,1,128,1,float16,float16,0,7.58250732421875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,40,40,128,1,float16,float16,0,9.001630401611328
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,40,2,128,1,float16,float16,0,7.709521484375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,40,1,128,1,float16,float16,0,31.90185546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,40,2,128,1,float16,float16,0,31.4537353515625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,40,40,128,1,float16,float16,0,4.319073486328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,40,8,128,1,float16,float16,0,32.52450866699219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,40,8,128,1,float16,float16,0,16.567730712890626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,40,4,128,1,float16,float16,0,33.95695190429687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,40,4,128,1,float16,float16,0,8.309156799316407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,40,1,128,1,float16,float16,0,3.59107666015625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,40,2,128,1,float16,float16,0,3.6192577362060545
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,40,8,128,1,float16,float16,0,8.144355010986327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,40,40,128,1,float16,float16,0,2.4161136627197264
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,40,2,128,1,float16,float16,0,1.8609615325927735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,40,4,128,1,float16,float16,0,3.7058624267578124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,40,1,128,1,float16,float16,0,1.7848207473754882
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,40,4,128,1,float16,float16,0,1.9254720687866211
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,40,8,128,1,float16,float16,0,3.7963790893554688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,40,8,128,1,float16,float16,0,1.9163711547851563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,40,1,128,1,float16,float16,0,14.492170715332032
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,40,2,128,1,float16,float16,0,13.916883850097657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,40,4,128,1,float16,float16,0,14.992948913574219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,40,40,128,1,float16,float16,0,18.503317260742186
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,40,1,128,1,float16,float16,0,7.140254211425781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,40,40,128,1,float16,float16,0,8.421692657470704
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,40,2,128,1,float16,float16,0,7.372679901123047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,40,2,128,1,float16,float16,0,28.704776000976562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,40,1,128,1,float16,float16,0,29.233517456054688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,40,4,128,1,float16,float16,0,30.715365600585937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,40,8,128,1,float16,float16,0,15.938255310058594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,40,8,128,1,float16,float16,0,31.191561889648437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,40,40,128,1,float16,float16,0,4.405963134765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,40,4,128,1,float16,float16,0,7.465831756591797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,40,8,128,1,float16,float16,0,7.248442840576172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,40,1,128,1,float16,float16,0,3.3301456451416014
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,40,2,128,1,float16,float16,0,3.4057472229003904
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,40,2,128,1,float16,float16,0,1.709012794494629
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,40,1,128,1,float16,float16,0,1.6355968475341798
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,40,4,128,1,float16,float16,0,1.6708608627319337
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,40,8,128,1,float16,float16,0,1.801246452331543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,40,40,128,1,float16,float16,0,2.0699440002441407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,40,40,128,1,float16,float16,0,1.1204143524169923
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,40,4,128,1,float16,float16,0,3.716769790649414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,40,1,128,1,float16,float16,0,0.8439567565917969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,40,2,128,1,float16,float16,0,0.9025263786315918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,40,8,128,1,float16,float16,0,3.639766311645508
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,40,4,128,1,float16,float16,0,0.9068240165710449
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,40,8,128,1,float16,float16,0,0.948588752746582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,40,2,128,1,float16,float16,0,8.16031494140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,40,1,128,1,float16,float16,0,8.211166381835938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,40,4,128,1,float16,float16,0,8.522286224365235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,40,40,128,1,float16,float16,0,10.251326751708984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,40,1,128,1,float16,float16,0,3.773382568359375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,40,40,128,1,float16,float16,0,5.040671920776367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,40,2,128,1,float16,float16,0,3.958750534057617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,40,1,128,1,float16,float16,0,16.557102966308594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,40,2,128,1,float16,float16,0,17.40767822265625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,40,8,128,1,float16,float16,0,8.659751892089844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,40,4,128,1,float16,float16,0,17.595079040527345
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,40,4,128,1,float16,float16,0,3.9538528442382814
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,40,8,128,1,float16,float16,0,18.210362243652344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,40,1,128,1,float16,float16,0,1.8584255218505858
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,40,40,128,1,float16,float16,0,2.6067232131958007
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,40,8,128,1,float16,float16,0,4.248788833618164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,40,2,128,1,float16,float16,0,1.8579856872558593
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,40,40,128,1,float16,float16,0,1.343563175201416
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,40,2,128,1,float16,float16,0,1.0211647987365722
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,40,1,128,1,float16,float16,0,1.0178400039672852
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,40,4,128,1,float16,float16,0,0.9387999534606933
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,40,40,128,1,float16,float16,0,0.7387360095977783
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,40,1,128,1,float16,float16,0,0.5125199794769287
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,40,8,128,1,float16,float16,0,2.0923999786376952
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,40,4,128,1,float16,float16,0,0.5263376235961914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,40,8,128,1,float16,float16,0,1.0337807655334472
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,40,8,128,1,float16,float16,0,0.5424464225769043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,40,4,128,1,float16,float16,0,1.9916496276855469
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,40,2,128,1,float16,float16,0,0.4982367992401123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,40,1,128,1,float16,float16,0,7.8132575988769535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,40,2,128,1,float16,float16,0,7.3760223388671875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,40,4,128,1,float16,float16,0,7.651898956298828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,40,40,128,1,float16,float16,0,10.783956909179688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,40,2,128,1,float16,float16,0,15.929910278320312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,40,1,128,1,float16,float16,0,16.055955505371095
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,40,1,128,1,float16,float16,0,3.5920543670654297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,40,40,128,1,float16,float16,0,5.191471862792969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,40,4,128,1,float16,float16,0,17.457667541503906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,40,8,128,1,float16,float16,0,17.557240295410157
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,40,8,128,1,float16,float16,0,9.202098846435547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,40,1,128,1,float16,float16,0,1.782811164855957
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,40,2,128,1,float16,float16,0,3.724212646484375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,40,4,128,1,float16,float16,0,4.099124908447266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,40,40,128,1,float16,float16,0,2.66287841796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,40,4,128,1,float16,float16,0,1.825155258178711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,40,8,128,1,float16,float16,0,3.9720863342285155
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,40,2,128,1,float16,float16,0,2.1168752670288087
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,40,40,128,1,float16,float16,0,0.699289608001709
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,40,1,128,1,float16,float16,0,0.8746975898742676
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,40,4,128,1,float16,float16,0,0.9852543830871582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,40,2,128,1,float16,float16,0,0.8037023544311523
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,40,8,128,1,float16,float16,0,0.989094352722168
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,40,8,128,1,float16,float16,0,1.9229488372802734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,40,1,128,1,float16,float16,0,0.4608799934387207
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,40,2,128,1,float16,float16,0,0.44744319915771485
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,40,4,128,1,float16,float16,0,0.4761055946350098
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,40,8,128,1,float16,float16,0,0.4606959819793701
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,40,1,128,1,float16,float16,0,0.2750607967376709
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,40,2,128,1,float16,float16,0,0.2590751886367798
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,40,4,128,1,float16,float16,0,0.26155838966369627
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,40,8,128,1,float16,float16,0,0.26899359226226804
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,40,40,128,1,float16,float16,0,1.3745648384094238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,40,40,128,1,float16,float16,0,0.36275200843811034
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,40,1,128,1,float16,float16,0,3.9968143463134767
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,40,2,128,1,float16,float16,0,3.6005199432373045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,40,40,128,1,float16,float16,0,6.9121757507324215
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,40,4,128,1,float16,float16,0,4.528387069702148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,40,1,128,1,float16,float16,0,8.9126220703125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,40,2,128,1,float16,float16,0,9.347547149658203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,40,4,128,1,float16,float16,0,9.541153717041016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,40,1,128,1,float16,float16,0,1.78721923828125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,40,8,128,1,float16,float16,0,10.545893096923828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,40,2,128,1,float16,float16,0,2.2449520111083983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,40,40,128,1,float16,float16,0,3.4089729309082033
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,40,4,128,1,float16,float16,0,2.1458736419677735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,40,1,128,1,float16,float16,0,0.9910063743591309
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,40,40,128,1,float16,float16,0,1.7639152526855468
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,40,1,128,1,float16,float16,0,0.517310380935669
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,40,8,128,1,float16,float16,0,2.2824512481689454
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,40,4,128,1,float16,float16,0,1.151863956451416
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,40,2,128,1,float16,float16,0,1.1104080200195312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,40,8,128,1,float16,float16,0,1.071664047241211
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,40,2,128,1,float16,float16,0,0.5155632019042968
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,40,4,128,1,float16,float16,0,0.5393008232116699
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,40,40,128,1,float16,float16,0,0.46129441261291504
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,40,8,128,1,float16,float16,0,0.611356782913208
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,40,2,128,1,float16,float16,0,0.27624480724334716
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,40,8,128,1,float16,float16,0,0.2849855899810791
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,40,40,128,1,float16,float16,0,0.1913040041923523
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,40,1,128,1,float16,float16,0,0.1671072006225586
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,40,2,128,1,float16,float16,0,0.17129119634628295
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,40,4,128,1,float16,float16,0,0.16803200244903566
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,40,8,128,1,float16,float16,0,0.16860640048980713
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,40,8,128,1,float16,float16,0,4.770822525024414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,40,40,128,1,float16,float16,0,0.9865792274475098
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,40,1,128,1,float16,float16,0,0.285481595993042
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,40,1,128,1,float16,float16,0,3.7673519134521483
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,40,4,128,1,float16,float16,0,0.2738127946853638
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,40,2,128,1,float16,float16,0,4.119440078735352
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,40,1,128,1,float16,float16,0,8.798670196533203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,40,2,128,1,float16,float16,0,8.996208190917969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,40,40,128,1,float16,float16,0,8.170305633544922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,40,4,128,1,float16,float16,0,10.044403076171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,40,8,128,1,float16,float16,0,10.051964569091798
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,40,2,128,1,float16,float16,0,2.0763776779174803
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,40,1,128,1,float16,float16,0,2.1952287673950197
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,40,4,128,1,float16,float16,0,1.9879264831542969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,40,4,128,1,float16,float16,0,4.526776123046875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,40,8,128,1,float16,float16,0,4.766987228393555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,40,40,128,1,float16,float16,0,4.151222229003906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,40,8,128,1,float16,float16,0,2.3393856048583985
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,40,2,128,1,float16,float16,0,1.0852368354797364
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,40,1,128,1,float16,float16,0,0.9640576362609863
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,40,40,128,1,float16,float16,0,2.2044687271118164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,40,4,128,1,float16,float16,0,1.092516803741455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,40,4,128,1,float16,float16,0,0.5295392036437988
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,40,1,128,1,float16,float16,0,0.47508797645568845
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,40,8,128,1,float16,float16,0,1.1737248420715332
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,40,8,128,1,float16,float16,0,0.5923903942108154
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,40,2,128,1,float16,float16,0,0.5391407966613769
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,40,40,128,1,float16,float16,0,1.1345760345458984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,40,1,128,1,float16,float16,0,0.28396000862121584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,40,2,128,1,float16,float16,0,0.2565808057785034
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,40,40,128,1,float16,float16,0,0.5741936206817627
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,40,40,128,1,float16,float16,0,0.24248321056365968
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,40,8,128,1,float16,float16,0,0.30820479393005373
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,40,1,128,1,float16,float16,0,0.14364320039749146
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,40,4,128,1,float16,float16,0,0.2529632091522217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,40,8,128,1,float16,float16,0,0.14585920572280883
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,40,2,128,1,float16,float16,0,0.13983839750289917
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,40,2,128,1,float16,float16,0,0.09217280149459839
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,40,40,128,1,float16,float16,0,0.09862880110740661
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,40,1,128,1,float16,float16,0,0.08461599946022033
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,40,4,128,1,float16,float16,0,0.08999040126800537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,40,4,128,1,float16,float16,0,0.1669648051261902
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,40,8,128,1,float16,float16,0,0.08857439756393433
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,40,2,128,1,float16,float16,0,5.092063903808594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,40,1,128,1,float16,float16,0,5.474132919311524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,40,4,128,1,float16,float16,0,5.741910552978515
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,40,1,128,1,float16,float16,0,2.379376029968262
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,40,2,128,1,float16,float16,0,2.4740320205688477
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,40,8,128,1,float16,float16,0,6.522484588623047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,40,1,128,1,float16,float16,0,1.340129566192627
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,40,2,128,1,float16,float16,0,1.440401554107666
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,40,4,128,1,float16,float16,0,2.8474655151367188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,40,40,128,1,float16,float16,0,8.096087646484374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,40,8,128,1,float16,float16,0,3.2731502532958983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,40,1,128,1,float16,float16,0,0.6678703784942627
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,40,2,128,1,float16,float16,0,0.6085968017578125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,40,8,128,1,float16,float16,0,1.6558063507080079
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,40,4,128,1,float16,float16,0,0.7033023834228516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,40,1,128,1,float16,float16,0,0.3193520069122314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,40,2,128,1,float16,float16,0,0.3349503993988037
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,40,8,128,1,float16,float16,0,0.8227375984191895
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,40,40,128,1,float16,float16,0,2.0765647888183594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,40,40,128,1,float16,float16,0,4.160244750976562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,40,4,128,1,float16,float16,0,0.35405919551849363
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,40,40,128,1,float16,float16,0,1.0567616462707519
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,40,1,128,1,float16,float16,0,0.16019999980926514
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,40,2,128,1,float16,float16,0,0.14928480386734008
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,40,40,128,1,float16,float16,0,0.5283440113067627
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,40,4,128,1,float16,float16,0,0.17147840261459352
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,40,8,128,1,float16,float16,0,0.422708797454834
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,40,40,128,1,float16,float16,0,0.18286240100860596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,40,4,128,1,float16,float16,0,0.08851839900016785
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,40,1,128,1,float16,float16,0,0.08351839780807495
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,40,40,128,1,float16,float16,0,0.056828802824020384
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,40,2,128,1,float16,float16,0,0.08072800040245057
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,40,1,128,1,float16,float16,0,0.044147199392318724
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,40,8,128,1,float16,float16,0,0.08815199732780457
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,40,2,128,1,float16,float16,0,0.049527999758720395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,40,4,128,1,float16,float16,0,0.0503216028213501
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,40,40,128,1,float16,float16,0,0.03670400083065033
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,40,2,128,1,float16,float16,0,0.03192479908466339
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,40,4,128,1,float16,float16,0,0.03380639851093292
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,40,8,128,1,float16,float16,0,0.03460640013217926
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,40,1,128,1,float16,float16,0,0.031534400582313535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,40,4,128,1,float16,float16,0,1.4436688423156738
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,40,1,128,1,float16,float16,0,1.129032039642334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,40,1,128,1,float16,float16,0,2.281158447265625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,40,2,128,1,float16,float16,0,2.4235408782958983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,40,8,128,1,float16,float16,0,0.20879039764404297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,40,4,128,1,float16,float16,0,2.7439727783203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,40,2,128,1,float16,float16,0,1.1857487678527832
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,40,8,128,1,float16,float16,0,0.049476799368858335
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,40,8,128,1,float16,float16,0,3.362179183959961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,40,1,128,1,float16,float16,0,0.6020607948303223
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,40,4,128,1,float16,float16,0,1.411248016357422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,40,4,128,1,float16,float16,0,0.6589488029479981
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,40,2,128,1,float16,float16,0,0.6465456008911132
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,40,40,128,1,float16,float16,0,4.095756912231446
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,40,2,128,1,float16,float16,0,0.28192479610443116
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,40,1,128,1,float16,float16,0,0.2923487901687622
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,40,8,128,1,float16,float16,0,1.7251760482788085
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,40,4,128,1,float16,float16,0,0.3258224010467529
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,40,8,128,1,float16,float16,0,0.8758399963378907
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,40,40,128,1,float16,float16,0,2.0619184494018556
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,40,1,128,1,float16,float16,0,0.09903519749641418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,40,40,128,1,float16,float16,0,1.0511584281921387
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,40,40,128,1,float16,float16,0,0.5155824184417724
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,40,8,128,1,float16,float16,0,0.1598639965057373
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,40,2,128,1,float16,float16,0,0.10085439682006836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,40,8,128,1,float16,float16,0,0.4218416213989258
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,40,4,128,1,float16,float16,0,0.112609601020813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,40,2,128,1,float16,float16,0,0.054555201530456544
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,40,4,128,1,float16,float16,0,0.0554256021976471
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,40,8,128,1,float16,float16,0,0.05695199966430664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,40,40,128,1,float16,float16,0,0.04056319892406464
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,40,40,128,1,float16,float16,0,0.16212480068206786
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,40,1,128,1,float16,float16,0,0.031780800223350524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,40,4,128,1,float16,float16,0,0.0330704003572464
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,40,8,128,1,float16,float16,0,0.03364959955215454
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,40,1,128,1,float16,float16,0,0.02380640059709549
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,40,2,128,1,float16,float16,0,0.03203679919242859
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,40,40,128,1,float16,float16,0,0.028036800026893616
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,40,2,128,1,float16,float16,0,0.02419999986886978
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,40,4,128,1,float16,float16,0,0.024347199499607085
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,40,40,128,1,float16,float16,0,0.018771199882030486
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,40,1,128,1,float16,float16,0,0.01639840006828308
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,40,2,128,1,float16,float16,0,0.016444799304008485
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,40,4,128,1,float16,float16,0,0.01656000018119812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,40,8,128,1,float16,float16,0,0.016896000504493712
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,40,1,128,1,float16,float16,0,0.5467152118682861
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,40,1,128,1,float16,float16,0,1.1343744277954102
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,40,2,128,1,float16,float16,0,1.2252464294433594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,40,4,128,1,float16,float16,0,1.433471965789795
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,40,2,128,1,float16,float16,0,0.5833663940429688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,40,8,128,1,float16,float16,0,1.6937744140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,40,40,128,1,float16,float16,0,2.061599922180176
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,40,1,128,1,float16,float16,0,0.2692591905593872
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,40,8,128,1,float16,float16,0,0.8437552452087402
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,40,4,128,1,float16,float16,0,0.7084815979003907
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,40,1,128,1,float16,float16,0,0.05464640259742737
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,40,2,128,1,float16,float16,0,0.29587359428405763
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,40,1,128,1,float16,float16,0,0.07656159996986389
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,40,8,128,1,float16,float16,0,0.3903183937072754
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,40,40,128,1,float16,float16,0,1.0496848106384278
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,40,2,128,1,float16,float16,0,0.07879679799079894
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,40,4,128,1,float16,float16,0,0.09109600186347962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,40,40,128,1,float16,float16,0,0.5080239772796631
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,40,4,128,1,float16,float16,0,0.3254528045654297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,40,8,128,1,float16,float16,0,0.15355360507965088
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,40,2,128,1,float16,float16,0,0.04328640103340149
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,40,4,128,1,float16,float16,0,0.044388800859451294
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,40,8,128,1,float16,float16,0,0.02481119930744171
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,40,8,128,1,float16,float16,0,0.045691201090812684
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,40,1,128,1,float16,float16,0,0.027608001232147218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,40,40,128,1,float16,float16,0,0.14949120283126832
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,40,2,128,1,float16,float16,0,0.027775999903678895
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,40,4,128,1,float16,float16,0,0.028327998518943787
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,40,8,128,1,float16,float16,0,0.029067200422286988
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,40,1,128,1,float16,float16,0,0.02003040015697479
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,40,2,128,1,float16,float16,0,0.02006240040063858
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,40,4,128,1,float16,float16,0,0.02038719952106476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,40,8,128,1,float16,float16,0,0.020761600136756896
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,40,40,128,1,float16,float16,0,0.014910399913787842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,40,1,128,1,float16,float16,0,0.013006399571895599
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,40,2,128,1,float16,float16,0,0.01289760023355484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,40,4,128,1,float16,float16,0,0.012838399410247803
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,40,8,128,1,float16,float16,0,0.013128000497817992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,40,40,128,1,float16,float16,0,0.013203200697898865
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,40,1,128,1,float16,float16,0,0.012198399752378464
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,40,2,128,1,float16,float16,0,0.01210239976644516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,40,4,128,1,float16,float16,0,0.012188799679279327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,40,8,128,1,float16,float16,0,0.01228479966521263
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,40,1,128,1,float16,float16,0,0.5490447998046875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,40,2,128,1,float16,float16,0,0.5887167930603028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,40,4,128,1,float16,float16,0,0.6623807907104492
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,40,8,128,1,float16,float16,0,0.8618880271911621
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,40,1,128,1,float16,float16,0,0.26813759803771975
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,40,2,128,1,float16,float16,0,0.2824719905853271
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,40,4,128,1,float16,float16,0,0.3163503885269165
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,40,40,128,1,float16,float16,0,1.0413663864135743
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,40,2,128,1,float16,float16,0,0.07715680003166199
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,40,40,128,1,float16,float16,0,0.506767988204956
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,40,4,128,1,float16,float16,0,0.08352959752082825
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,40,8,128,1,float16,float16,0,0.1444383978843689
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,40,1,128,1,float16,float16,0,0.04324800074100495
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,40,1,128,1,float16,float16,0,0.04206880033016205
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,40,2,128,1,float16,float16,0,0.04229280054569244
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,40,40,128,1,float16,float16,0,0.03605599999427796
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,40,8,128,1,float16,float16,0,0.05005279779434204
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,40,4,128,1,float16,float16,0,0.04302240014076233
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,40,40,128,1,float16,float16,0,0.024391999840736388
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,40,1,128,1,float16,float16,0,0.027086400985717775
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,40,2,128,1,float16,float16,0,0.027164798974990845
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,40,4,128,1,float16,float16,0,0.027908799052238465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,40,8,128,1,float16,float16,0,0.028622400760650635
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,40,40,128,1,float16,float16,0,0.028591999411582948
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,40,2,128,1,float16,float16,0,0.019542400538921357
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,40,4,128,1,float16,float16,0,0.019726400077342988
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,40,1,128,1,float16,float16,0,0.01992959976196289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,40,8,128,1,float16,float16,0,0.02033119946718216
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,40,40,128,1,float16,float16,0,0.019823999702930452
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,40,1,128,1,float16,float16,0,0.012932799756526947
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,40,2,128,1,float16,float16,0,0.01268640011548996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,40,4,128,1,float16,float16,0,0.012574400007724761
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,40,40,128,1,float16,float16,0,0.012779200077056884
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,40,1,128,1,float16,float16,0,0.011934400349855424
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,40,8,128,1,float16,float16,0,0.013232000172138214
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,40,2,128,1,float16,float16,0,0.01188800036907196
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,40,4,128,1,float16,float16,0,0.011987199634313583
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,40,40,128,1,float16,float16,0,0.012201599776744843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,40,1,128,1,float16,float16,0,0.01170400008559227
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,40,4,128,1,float16,float16,0,0.011779200285673141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,40,2,128,1,float16,float16,0,0.011785600334405899
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,40,8,128,1,float16,float16,0,0.011883199959993363
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,40,1,128,1,float16,float16,0,0.27208800315856935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,40,2,128,1,float16,float16,0,0.2847007989883423
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,40,1,128,1,float16,float16,0,0.07449920177459717
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,40,4,128,1,float16,float16,0,0.31065120697021487
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,40,8,128,1,float16,float16,0,0.39278080463409426
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,40,8,128,1,float16,float16,0,0.37033278942108155
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,40,1,128,1,float16,float16,0,0.07413280010223389
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,40,2,128,1,float16,float16,0,0.07650880217552185
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,40,40,128,1,float16,float16,0,0.4719376087188721
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,40,4,128,1,float16,float16,0,0.09503679871559143
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,40,40,128,1,float16,float16,0,0.1663264036178589
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,40,8,128,1,float16,float16,0,0.16192640066146852
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,40,1,128,1,float16,float16,0,0.04200960099697113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,40,2,128,1,float16,float16,0,0.042238399386405945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,40,4,128,1,float16,float16,0,0.04812000095844269
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,40,8,128,1,float16,float16,0,0.06000639796257019
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,40,40,128,1,float16,float16,0,0.07240319848060608
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,40,1,128,1,float16,float16,0,0.027161601185798644
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,40,2,128,1,float16,float16,0,0.02717599868774414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,40,8,128,1,float16,float16,0,0.03371999859809875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,40,40,128,1,float16,float16,0,0.04501760005950928
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,40,40,128,1,float16,float16,0,0.039268800616264345
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,40,1,128,1,float16,float16,0,0.019598400592803954
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,40,2,128,1,float16,float16,0,0.019531199336051942
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,40,4,128,1,float16,float16,0,0.019491200149059296
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,40,8,128,1,float16,float16,0,0.019865599274635316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,40,1,128,1,float16,float16,0,0.012540799379348756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,40,2,128,1,float16,float16,0,0.012513600289821625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,40,4,128,1,float16,float16,0,0.01255040019750595
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,40,8,128,1,float16,float16,0,0.012585599720478059
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,40,40,128,1,float16,float16,0,0.018089599907398224
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,40,1,128,1,float16,float16,0,0.011715199798345566
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,40,4,128,1,float16,float16,0,0.011832000315189361
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,40,8,128,1,float16,float16,0,0.012036799639463424
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,40,40,128,1,float16,float16,0,0.011980800330638886
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,40,1,128,1,float16,float16,0,0.011633600294589996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,40,8,128,1,float16,float16,0,0.012439999729394913
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,40,2,128,1,float16,float16,0,0.011707200109958649
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,40,8,128,1,float16,float16,0,0.011777599900960922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,40,40,128,1,float16,float16,0,0.011751999706029892
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,40,4,128,1,float16,float16,0,0.012035199999809265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,40,1,128,1,float16,float16,0,0.011606399714946748
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,40,4,128,1,float16,float16,0,0.011644800007343293
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,40,2,128,1,float16,float16,0,0.011984000355005265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,40,8,128,1,float16,float16,0,0.01165279969573021
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,40,4,128,1,float16,float16,0,0.11245280504226685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,40,1,128,1,float16,float16,0,0.07568640112876893
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,40,8,128,1,float16,float16,0,0.16724159717559814
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,40,1,128,1,float16,float16,0,0.04221439957618713
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,40,40,128,1,float16,float16,0,0.32527999877929686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,40,2,128,1,float16,float16,0,0.04750880002975464
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,40,40,128,1,float16,float16,0,0.21649599075317383
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,40,4,128,1,float16,float16,0,0.05902079939842224
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,40,8,128,1,float16,float16,0,0.06074720025062561
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,40,40,128,1,float16,float16,0,0.12469439506530762
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,40,4,128,1,float16,float16,0,0.02757599949836731
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,40,2,128,1,float16,float16,0,0.0271807998418808
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,40,1,128,1,float16,float16,0,0.02783679962158203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,40,4,128,1,float16,float16,0,0.03336000144481659
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,40,8,128,1,float16,float16,0,0.03383359909057617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,40,40,128,1,float16,float16,0,0.025216001272201537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,40,40,128,1,float16,float16,0,0.06755040287971496
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,40,1,128,1,float16,float16,0,0.01996160000562668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,40,2,128,1,float16,float16,0,0.019724799692630766
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,40,4,128,1,float16,float16,0,0.019644799828529357
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,40,40,128,1,float16,float16,0,0.036575999855995175
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,40,8,128,1,float16,float16,0,0.020187200605869295
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,40,1,128,1,float16,float16,0,0.012804800271987915
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,40,4,128,1,float16,float16,0,0.012468799948692322
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,40,8,128,1,float16,float16,0,0.012479999661445617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,40,2,128,1,float16,float16,0,0.011751999706029892
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,40,2,128,1,float16,float16,0,0.0117296002805233
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,40,1,128,1,float16,float16,0,0.011847999691963196
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,40,8,128,1,float16,float16,0,0.011804799735546111
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,40,40,128,1,float16,float16,0,0.017608000338077544
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,40,1,128,1,float16,float16,0,0.011737599968910217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,40,4,128,1,float16,float16,0,0.011694400012493134
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,40,2,128,1,float16,float16,0,0.011796800047159195
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,40,40,128,1,float16,float16,0,0.011870399862527848
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,40,2,128,1,float16,float16,0,0.011604800075292587
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,40,1,128,1,float16,float16,0,0.011776000261306763
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,40,8,128,1,float16,float16,0,0.011446399986743927
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,40,40,128,1,float16,float16,0,0.011884800344705581
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,40,1,128,1,float16,float16,0,0.01167680025100708
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,40,2,128,1,float16,float16,0,0.011691199988126755
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,40,2,128,1,float16,float16,0,0.0835536003112793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,40,4,128,1,float16,float16,0,0.011763200163841248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,40,8,128,1,float16,float16,0,0.011336000263690948
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,40,1,128,1,float16,float16,0,0.02000479996204376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,40,2,128,1,float16,float16,0,0.03172959983348846
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,40,4,128,1,float16,float16,0,0.055313599109649655
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,40,8,128,1,float16,float16,0,0.09707840085029602
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,40,1,128,1,float16,float16,0,0.012803199887275695
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,40,40,128,1,float16,float16,0,0.22191519737243653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,40,4,128,1,float16,float16,0,0.030551999807357788
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,40,8,128,1,float16,float16,0,0.05446079969406128
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,40,40,128,1,float16,float16,0,0.11673599481582642
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,40,1,128,1,float16,float16,0,0.011820799857378005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,40,2,128,1,float16,float16,0,0.012654399871826172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,40,40,128,1,float16,float16,0,0.024172799289226533
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,40,4,128,1,float16,float16,0,0.01820639967918396
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,40,40,128,1,float16,float16,0,0.06460800170898437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,40,8,128,1,float16,float16,0,0.011503999680280685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,40,8,128,1,float16,float16,0,0.03015359938144684
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,40,1,128,1,float16,float16,0,0.011841599643230439
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,40,2,128,1,float16,float16,0,0.011547199636697768
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,40,4,128,1,float16,float16,0,0.011718399822711945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,40,4,128,1,float16,float16,0,0.01194079965353012
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,40,8,128,1,float16,float16,0,0.01791519969701767
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,40,1,128,1,float16,float16,0,0.011135999858379365
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,40,2,128,1,float16,float16,0,0.011283200234174728
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,40,8,128,1,float16,float16,0,0.011430399864912033
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,40,4,128,1,float16,float16,0,0.011793600022792816
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,40,40,128,1,float16,float16,0,0.02366719990968704
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,40,1,128,1,float16,float16,0,0.011137600243091583
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,40,2,128,1,float16,float16,0,0.011814399808645248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,40,8,128,1,float16,float16,0,0.011220800131559372
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,40,4,128,1,float16,float16,0,0.01125440001487732
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,40,40,128,1,float16,float16,0,0.0177824005484581
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,40,1,128,1,float16,float16,0,0.011108800023794174
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,40,2,128,1,float16,float16,0,0.011982399970293045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,40,8,128,1,float16,float16,0,0.01125119999051094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,40,4,128,1,float16,float16,0,0.012030400335788727
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,40,40,128,1,float16,float16,0,0.012151999771595002
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,40,1,128,1,float16,float16,0,0.011187200248241425
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,40,2,128,1,float16,float16,0,0.011736000329256058
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,40,8,128,1,float16,float16,0,0.010927999764680863
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,40,40,128,1,float16,float16,0,0.011982399970293045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,40,1,128,1,float16,float16,0,0.010900799930095673
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,40,2,128,1,float16,float16,0,0.018542400002479552
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,40,4,128,1,float16,float16,0,0.010918399691581726
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,40,2,128,1,float16,float16,0,0.011798399686813354
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,40,8,128,1,float16,float16,0,0.011433599889278412
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,40,2,128,1,float16,float16,0,0.012009599804878235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,40,4,128,1,float16,float16,0,0.011788800358772278
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,40,40,128,1,float16,float16,0,0.035281598567962646
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,40,4,128,1,float16,float16,0,0.011187200248241425
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,32,1,128,1,float16,float16,0,41.877743530273435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,32,2,128,1,float16,float16,0,41.4711669921875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,32,4,128,1,float16,float16,0,42.01717834472656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,32,1,128,1,float16,float16,0,20.675289916992188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,32,2,128,1,float16,float16,0,83.63529052734376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,32,1,128,1,float16,float16,0,85.91815185546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,32,4,128,1,float16,float16,0,85.54374389648437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,32,8,128,1,float16,float16,0,43.62745361328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,32,8,128,1,float16,float16,0,89.3420166015625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,32,2,128,1,float16,float16,0,21.056132507324218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,32,1,128,1,float16,float16,0,10.246844482421874
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,32,2,128,1,float16,float16,0,10.49957275390625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,32,4,128,1,float16,float16,0,21.155311584472656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,32,4,128,1,float16,float16,0,10.56195526123047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,32,8,128,1,float16,float16,0,21.759379577636718
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,32,8,128,1,float16,float16,0,10.547881317138671
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,32,1,128,1,float16,float16,0,49.4565673828125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,32,2,128,1,float16,float16,0,48.01676940917969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,32,4,128,1,float16,float16,0,49.1395751953125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,32,8,128,1,float16,float16,0,51.10338134765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,32,1,128,1,float16,float16,0,24.913656616210936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,32,2,128,1,float16,float16,0,24.008592224121095
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,32,4,128,1,float16,float16,0,24.72509002685547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,32,8,128,1,float16,float16,0,25.217291259765624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,32,1,128,1,float16,float16,0,11.787750244140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,32,2,128,1,float16,float16,0,11.771334075927735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,32,4,128,1,float16,float16,0,12.568692779541015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,32,8,128,1,float16,float16,0,12.424034881591798
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,32,1,128,1,float16,float16,0,5.909849548339844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,32,2,128,1,float16,float16,0,5.71118392944336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,32,4,128,1,float16,float16,0,6.070473480224609
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,32,8,128,1,float16,float16,0,6.050993728637695
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,32,1,128,1,float16,float16,0,34.78251342773437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,32,2,128,1,float16,float16,0,34.040341186523435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,32,32,128,1,float16,float16,0,10.797036743164062
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,32,32,128,1,float16,float16,0,6.261878585815429
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,32,32,128,1,float16,float16,0,13.241017150878907
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,32,4,128,1,float16,float16,0,35.69153747558594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,32,8,128,1,float16,float16,0,34.864990234375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,32,32,128,1,float16,float16,0,22.04249725341797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,32,32,128,1,float16,float16,0,25.88787841796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,32,1,128,1,float16,float16,0,16.072308349609376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,32,32,128,1,float16,float16,0,18.17852020263672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,32,32,128,1,float16,float16,0,9.041844940185547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,32,1,128,1,float16,float16,0,8.098494720458984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,32,2,128,1,float16,float16,0,16.754161071777343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,32,2,128,1,float16,float16,0,8.029496002197266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,32,4,128,1,float16,float16,0,8.271132659912109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,32,4,128,1,float16,float16,0,17.51036376953125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,32,1,128,1,float16,float16,0,4.062707138061524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,32,32,128,1,float16,float16,0,4.73076171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,32,8,128,1,float16,float16,0,17.26627197265625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,32,2,128,1,float16,float16,0,4.187713623046875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,32,4,128,1,float16,float16,0,4.165184020996094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,32,8,128,1,float16,float16,0,8.976187133789063
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,32,8,128,1,float16,float16,0,4.186475372314453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,32,32,128,1,float16,float16,0,46.220681762695314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,32,1,128,1,float16,float16,0,21.362608337402342
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,32,2,128,1,float16,float16,0,22.706346130371095
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,32,4,128,1,float16,float16,0,22.446400451660157
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,32,32,128,1,float16,float16,0,24.431292724609374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,32,2,128,1,float16,float16,0,10.57715072631836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,32,1,128,1,float16,float16,0,11.309214019775391
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,32,32,128,1,float16,float16,0,12.391841888427734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,32,2,128,1,float16,float16,0,44.119659423828125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,32,1,128,1,float16,float16,0,45.14564208984375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,32,32,128,1,float16,float16,0,6.204008102416992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,32,4,128,1,float16,float16,0,45.070611572265626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,32,8,128,1,float16,float16,0,23.290538024902343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,32,8,128,1,float16,float16,0,45.58124694824219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,32,4,128,1,float16,float16,0,11.064905548095703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,32,1,128,1,float16,float16,0,5.192779159545898
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,32,8,128,1,float16,float16,0,11.671196746826173
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,32,32,128,1,float16,float16,0,2.8683904647827148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,32,1,128,1,float16,float16,0,2.5199119567871096
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,32,2,128,1,float16,float16,0,5.555207824707031
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,32,4,128,1,float16,float16,0,5.3528800964355465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,32,8,128,1,float16,float16,0,5.344591903686523
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,32,2,128,1,float16,float16,0,2.58239688873291
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,32,4,128,1,float16,float16,0,2.6005647659301756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,32,8,128,1,float16,float16,0,2.7022815704345704
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,32,1,128,1,float16,float16,0,11.929124450683593
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,32,2,128,1,float16,float16,0,12.414542388916015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,32,4,128,1,float16,float16,0,13.087635803222657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,32,32,128,1,float16,float16,0,14.406678771972656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,32,1,128,1,float16,float16,0,6.023174285888672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,32,32,128,1,float16,float16,0,7.242950439453125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,32,2,128,1,float16,float16,0,5.922824096679688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,32,1,128,1,float16,float16,0,24.776573181152344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,32,2,128,1,float16,float16,0,24.871115112304686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,32,4,128,1,float16,float16,0,26.588409423828125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,32,8,128,1,float16,float16,0,12.790396881103515
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,32,8,128,1,float16,float16,0,27.089031982421876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,32,1,128,1,float16,float16,0,2.8425567626953123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,32,32,128,1,float16,float16,0,3.606808090209961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,32,4,128,1,float16,float16,0,6.777928161621094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,32,8,128,1,float16,float16,0,6.4934333801269535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,32,32,128,1,float16,float16,0,1.7298736572265625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,32,4,128,1,float16,float16,0,2.8650400161743166
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,32,1,128,1,float16,float16,0,1.4959936141967773
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,32,2,128,1,float16,float16,0,3.146633529663086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,32,2,128,1,float16,float16,0,1.6429183959960938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,32,8,128,1,float16,float16,0,2.9332672119140626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,32,4,128,1,float16,float16,0,1.5203632354736327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,32,8,128,1,float16,float16,0,1.5163840293884276
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,32,1,128,1,float16,float16,0,11.356185913085938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,32,2,128,1,float16,float16,0,11.585724639892579
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,32,4,128,1,float16,float16,0,11.45891876220703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,32,32,128,1,float16,float16,0,14.20533447265625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,32,32,128,1,float16,float16,0,6.661316680908203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,32,1,128,1,float16,float16,0,5.307470321655273
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,32,2,128,1,float16,float16,0,5.080310440063476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,32,1,128,1,float16,float16,0,22.934918212890626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,32,2,128,1,float16,float16,0,23.13813171386719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,32,8,128,1,float16,float16,0,12.239769744873048
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,32,32,128,1,float16,float16,0,3.3824478149414063
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,32,4,128,1,float16,float16,0,24.983702087402342
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,32,8,128,1,float16,float16,0,25.455816650390624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,32,1,128,1,float16,float16,0,2.569881629943848
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,32,2,128,1,float16,float16,0,2.6482416152954102
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,32,4,128,1,float16,float16,0,5.591175842285156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,32,8,128,1,float16,float16,0,5.798155212402344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,32,32,128,1,float16,float16,0,1.7098255157470703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,32,2,128,1,float16,float16,0,1.3675344467163086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,32,4,128,1,float16,float16,0,2.671228790283203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,32,1,128,1,float16,float16,0,1.299452781677246
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,32,32,128,1,float16,float16,0,0.9211647987365723
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,32,8,128,1,float16,float16,0,2.8801904678344727
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,32,8,128,1,float16,float16,0,1.3624688148498536
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,32,2,128,1,float16,float16,0,0.6869391918182373
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,32,1,128,1,float16,float16,0,0.6648015975952148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,32,4,128,1,float16,float16,0,0.763102388381958
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,32,4,128,1,float16,float16,0,1.434553623199463
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,32,8,128,1,float16,float16,0,0.7579535961151123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,32,1,128,1,float16,float16,0,6.5130767822265625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,32,32,128,1,float16,float16,0,8.466770935058594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,32,2,128,1,float16,float16,0,5.81495361328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,32,4,128,1,float16,float16,0,6.870320129394531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,32,1,128,1,float16,float16,0,13.173529052734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,32,1,128,1,float16,float16,0,2.9106975555419923
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,32,32,128,1,float16,float16,0,4.122691345214844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,32,2,128,1,float16,float16,0,14.185772705078126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,32,8,128,1,float16,float16,0,14.209237670898437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,32,8,128,1,float16,float16,0,6.850950622558594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,32,2,128,1,float16,float16,0,3.0861759185791016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,32,1,128,1,float16,float16,0,1.4414400100708007
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,32,4,128,1,float16,float16,0,3.317804718017578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,32,32,128,1,float16,float16,0,2.123838424682617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,32,4,128,1,float16,float16,0,1.5543984413146972
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,32,8,128,1,float16,float16,0,3.5100208282470704
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,32,2,128,1,float16,float16,0,1.7932863235473633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,32,8,128,1,float16,float16,0,1.6522319793701172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,32,32,128,1,float16,float16,0,1.1087823867797852
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,32,4,128,1,float16,float16,0,0.814691162109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,32,2,128,1,float16,float16,0,0.8076208114624024
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,32,1,128,1,float16,float16,0,0.4757472038269043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,32,1,128,1,float16,float16,0,0.8430000305175781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,32,32,128,1,float16,float16,0,0.6096848011016845
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,32,8,128,1,float16,float16,0,0.8477647781372071
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,32,2,128,1,float16,float16,0,0.40976800918579104
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,32,4,128,1,float16,float16,0,0.41892638206481936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,32,8,128,1,float16,float16,0,0.4150656223297119
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,32,4,128,1,float16,float16,0,14.218598937988281
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,32,2,128,1,float16,float16,0,5.820119857788086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,32,1,128,1,float16,float16,0,5.932758331298828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,32,32,128,1,float16,float16,0,8.461427307128906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,32,4,128,1,float16,float16,0,5.920982360839844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,32,32,128,1,float16,float16,0,4.194478225708008
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,32,1,128,1,float16,float16,0,12.569417572021484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,32,1,128,1,float16,float16,0,2.6619264602661135
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,32,4,128,1,float16,float16,0,12.60475845336914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,32,2,128,1,float16,float16,0,12.72908935546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,32,2,128,1,float16,float16,0,2.982921600341797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,32,8,128,1,float16,float16,0,6.913910675048828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,32,8,128,1,float16,float16,0,13.776899719238282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,32,4,128,1,float16,float16,0,2.8303056716918946
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,32,2,128,1,float16,float16,0,1.4314448356628418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,32,1,128,1,float16,float16,0,1.6065343856811523
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,32,4,128,1,float16,float16,0,1.507487964630127
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,32,32,128,1,float16,float16,0,2.110438346862793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,32,2,128,1,float16,float16,0,0.747214412689209
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,32,1,128,1,float16,float16,0,0.7582047939300537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,32,32,128,1,float16,float16,0,1.2069120407104492
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,32,4,128,1,float16,float16,0,0.7124415874481201
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,32,8,128,1,float16,float16,0,1.588155174255371
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,32,8,128,1,float16,float16,0,3.2601600646972657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,32,32,128,1,float16,float16,0,0.5510272026062012
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,32,8,128,1,float16,float16,0,0.8493727684020996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,32,2,128,1,float16,float16,0,0.38390560150146485
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,32,4,128,1,float16,float16,0,0.37005441188812255
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,32,8,128,1,float16,float16,0,0.37568800449371337
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,32,2,128,1,float16,float16,0,0.1961408019065857
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,32,32,128,1,float16,float16,0,0.25950400829315184
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,32,1,128,1,float16,float16,0,0.25211520195007325
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,32,8,128,1,float16,float16,0,0.22365760803222656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,32,1,128,1,float16,float16,0,0.3676016092300415
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,32,4,128,1,float16,float16,0,0.22845759391784667
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,32,32,128,1,float16,float16,0,5.460209655761719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,32,1,128,1,float16,float16,0,6.6328880310058596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,32,1,128,1,float16,float16,0,3.2218128204345704
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,32,2,128,1,float16,float16,0,7.463775634765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,32,2,128,1,float16,float16,0,3.4770511627197265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,32,4,128,1,float16,float16,0,8.343816375732422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,32,8,128,1,float16,float16,0,8.676918029785156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,32,1,128,1,float16,float16,0,1.9236543655395508
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,32,2,128,1,float16,float16,0,1.8619216918945312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,32,4,128,1,float16,float16,0,3.0947200775146486
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,32,32,128,1,float16,float16,0,2.8824575424194334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,32,4,128,1,float16,float16,0,1.7526687622070312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,32,8,128,1,float16,float16,0,3.813832092285156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,32,1,128,1,float16,float16,0,0.8762543678283692
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,32,2,128,1,float16,float16,0,0.7869631767272949
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,32,8,128,1,float16,float16,0,1.8392976760864257
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,32,4,128,1,float16,float16,0,0.8722847938537598
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,32,32,128,1,float16,float16,0,1.381537628173828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,32,2,128,1,float16,float16,0,0.45018558502197265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,32,32,128,1,float16,float16,0,0.7924352169036866
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,32,8,128,1,float16,float16,0,0.8865568161010742
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,32,1,128,1,float16,float16,0,0.48633441925048826
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,32,1,128,1,float16,float16,0,0.21729280948638915
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,32,32,128,1,float16,float16,0,0.3697376012802124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,32,2,128,1,float16,float16,0,0.23570079803466798
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,32,1,128,1,float16,float16,0,0.12452160120010376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,32,8,128,1,float16,float16,0,0.5073984146118165
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,32,32,128,1,float16,float16,0,0.15998560190200806
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,32,2,128,1,float16,float16,0,0.1409551978111267
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,32,8,128,1,float16,float16,0,0.2276832103729248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,32,4,128,1,float16,float16,0,0.2579375982284546
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,32,8,128,1,float16,float16,0,0.13374240398406984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,32,4,128,1,float16,float16,0,0.43816637992858887
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,32,1,128,1,float16,float16,0,3.226216125488281
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,32,4,128,1,float16,float16,0,0.13989919424057007
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,32,2,128,1,float16,float16,0,3.076980781555176
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,32,1,128,1,float16,float16,0,7.226248168945313
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,32,32,128,1,float16,float16,0,6.491600036621094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,32,2,128,1,float16,float16,0,7.195513916015625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,32,4,128,1,float16,float16,0,7.173153686523437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,32,4,128,1,float16,float16,0,3.367486572265625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,32,8,128,1,float16,float16,0,3.655246353149414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,32,8,128,1,float16,float16,0,8.082202911376953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,32,1,128,1,float16,float16,0,1.6349519729614257
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,32,1,128,1,float16,float16,0,0.8039263725280762
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,32,2,128,1,float16,float16,0,1.4875344276428222
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,32,4,128,1,float16,float16,0,1.890376091003418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,32,8,128,1,float16,float16,0,1.8923904418945312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,32,32,128,1,float16,float16,0,1.726705551147461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,32,2,128,1,float16,float16,0,0.8200207710266113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,32,4,128,1,float16,float16,0,0.8978256225585938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,32,8,128,1,float16,float16,0,0.9185600280761719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,32,1,128,1,float16,float16,0,0.41777920722961426
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,32,4,128,1,float16,float16,0,0.5070752143859864
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,32,32,128,1,float16,float16,0,0.8545568466186524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,32,2,128,1,float16,float16,0,0.4314176082611084
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,32,8,128,1,float16,float16,0,0.46787362098693847
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,32,1,128,1,float16,float16,0,0.20167040824890137
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,32,4,128,1,float16,float16,0,0.23539040088653565
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,32,32,128,1,float16,float16,0,0.15209920406341554
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,32,8,128,1,float16,float16,0,0.23375840187072755
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,32,32,128,1,float16,float16,0,0.4385200023651123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,32,1,128,1,float16,float16,0,0.11081759929656983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,32,4,128,1,float16,float16,0,0.1280336022377014
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,32,8,128,1,float16,float16,0,0.12368320226669312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,32,2,128,1,float16,float16,0,0.06683200001716613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,32,1,128,1,float16,float16,0,0.06583520174026489
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,32,2,128,1,float16,float16,0,0.11531840562820435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,32,4,128,1,float16,float16,0,0.08562719821929932
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,32,8,128,1,float16,float16,0,0.08009600043296813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,32,32,128,1,float16,float16,0,3.2913169860839844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,32,2,128,1,float16,float16,0,0.21090879440307617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,32,32,128,1,float16,float16,0,0.09528639912605286
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,32,1,128,1,float16,float16,0,4.186544036865234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,32,2,128,1,float16,float16,0,4.3951679229736325
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,32,4,128,1,float16,float16,0,4.655820846557617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,32,1,128,1,float16,float16,0,1.8144351959228515
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,32,2,128,1,float16,float16,0,1.9040031433105469
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,32,4,128,1,float16,float16,0,2.234222412109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,32,2,128,1,float16,float16,0,0.9685664176940918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,32,1,128,1,float16,float16,0,1.0417712211608887
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,32,32,128,1,float16,float16,0,6.4513694763183596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,32,1,128,1,float16,float16,0,0.5215248107910156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,32,8,128,1,float16,float16,0,2.941044807434082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,32,8,128,1,float16,float16,0,1.4136063575744628
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,32,2,128,1,float16,float16,0,0.5051760196685791
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,32,32,128,1,float16,float16,0,1.665166473388672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,32,32,128,1,float16,float16,0,3.326935958862305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,32,8,128,1,float16,float16,0,0.736078405380249
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,32,1,128,1,float16,float16,0,0.25799200534820554
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,32,32,128,1,float16,float16,0,0.8718048095703125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,32,2,128,1,float16,float16,0,0.2703711986541748
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,32,4,128,1,float16,float16,0,0.2958415985107422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,32,32,128,1,float16,float16,0,0.3887631893157959
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,32,1,128,1,float16,float16,0,0.11315040588378907
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,32,8,128,1,float16,float16,0,0.3612112045288086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,32,2,128,1,float16,float16,0,0.1183359980583191
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,32,4,128,1,float16,float16,0,0.11952159404754639
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,32,8,128,1,float16,float16,0,0.1351696014404297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,32,8,128,1,float16,float16,0,5.734382247924804
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,32,2,128,1,float16,float16,0,0.0686303973197937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,32,4,128,1,float16,float16,0,0.0750111997127533
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,32,1,128,1,float16,float16,0,0.06576319932937622
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,32,32,128,1,float16,float16,0,0.046588799357414244
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,32,1,128,1,float16,float16,0,0.03961600065231323
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,32,2,128,1,float16,float16,0,0.0421424001455307
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,32,4,128,1,float16,float16,0,0.04071359932422638
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,32,8,128,1,float16,float16,0,0.041259199380874634
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,32,1,128,1,float16,float16,0,0.024907200038433074
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,32,2,128,1,float16,float16,0,0.0247856006026268
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,32,8,128,1,float16,float16,0,0.07829440236091614
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,32,4,128,1,float16,float16,0,1.1342543601989745
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,32,4,128,1,float16,float16,0,0.025142401456832886
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,32,8,128,1,float16,float16,0,0.025273600220680238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,32,4,128,1,float16,float16,0,0.560758399963379
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,32,1,128,1,float16,float16,0,1.9629056930541993
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,32,2,128,1,float16,float16,0,2.012148857116699
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,32,4,128,1,float16,float16,0,2.355232048034668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,32,32,128,1,float16,float16,0,0.10074399709701538
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,32,8,128,1,float16,float16,0,2.938369560241699
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,32,32,128,1,float16,float16,0,0.02802560031414032
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,32,1,128,1,float16,float16,0,0.9336992263793945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,32,2,128,1,float16,float16,0,1.000883197784424
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,32,1,128,1,float16,float16,0,0.4406447887420654
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,32,4,128,1,float16,float16,0,1.2086079597473145
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,32,2,128,1,float16,float16,0,0.5167952060699463
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,32,8,128,1,float16,float16,0,1.4751952171325684
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,32,32,128,1,float16,float16,0,3.2885822296142577
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,32,1,128,1,float16,float16,0,0.21330881118774414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,32,8,128,1,float16,float16,0,0.7358128070831299
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,32,2,128,1,float16,float16,0,0.22765600681304932
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,32,32,128,1,float16,float16,0,1.6541536331176758
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,32,8,128,1,float16,float16,0,0.33027360439300535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,32,32,128,1,float16,float16,0,0.8441328048706055
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,32,4,128,1,float16,float16,0,0.2749824047088623
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,32,32,128,1,float16,float16,0,0.36266560554504396
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,32,4,128,1,float16,float16,0,0.08288639783859253
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,32,8,128,1,float16,float16,0,0.09558719992637635
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,32,1,128,1,float16,float16,0,0.07963359951972962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,32,32,128,1,float16,float16,0,0.08752480149269104
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,32,2,128,1,float16,float16,0,0.046191999316215517
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,32,2,128,1,float16,float16,0,0.0811344027519226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,32,8,128,1,float16,float16,0,0.048681598901748654
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,32,1,128,1,float16,float16,0,0.028457599878311157
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,32,4,128,1,float16,float16,0,0.028830400109291075
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,32,32,128,1,float16,float16,0,0.034971201419830324
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,32,8,128,1,float16,float16,0,0.02985279858112335
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,32,2,128,1,float16,float16,0,0.01724800020456314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,32,4,128,1,float16,float16,0,0.017401599884033205
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,32,8,128,1,float16,float16,0,0.017894400656223296
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,32,32,128,1,float16,float16,0,0.01775840073823929
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,32,1,128,1,float16,float16,0,0.016196799278259278
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,32,2,128,1,float16,float16,0,0.016044799983501435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,32,4,128,1,float16,float16,0,0.016246399283409117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,32,8,128,1,float16,float16,0,0.0164560005068779
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,32,4,128,1,float16,float16,0,0.5513440132141113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,32,1,128,1,float16,float16,0,0.9115695953369141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,32,2,128,1,float16,float16,0,0.9960607528686524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,32,4,128,1,float16,float16,0,0.04708000123500824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,32,1,128,1,float16,float16,0,0.045423999428749084
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,32,2,128,1,float16,float16,0,0.02837119996547699
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,32,32,128,1,float16,float16,0,0.020449599623680113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,32,4,128,1,float16,float16,0,1.1863295555114746
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,32,2,128,1,float16,float16,0,0.4975696086883545
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,32,8,128,1,float16,float16,0,1.4643376350402832
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,32,4,128,1,float16,float16,0,0.5540639877319335
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,32,1,128,1,float16,float16,0,0.20833919048309327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,32,8,128,1,float16,float16,0,0.7191167831420898
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,32,2,128,1,float16,float16,0,0.2352368116378784
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,32,4,128,1,float16,float16,0,0.2573024034500122
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,32,8,128,1,float16,float16,0,0.3255808115005493
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,32,32,128,1,float16,float16,0,0.8301823616027832
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,32,2,128,1,float16,float16,0,0.06352319717407226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,32,4,128,1,float16,float16,0,0.06592479944229127
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,32,1,128,1,float16,float16,0,0.0628063976764679
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,32,32,128,1,float16,float16,0,1.646023941040039
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,32,2,128,1,float16,float16,0,0.036620798707008365
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,32,8,128,1,float16,float16,0,0.08141599893569947
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,32,8,128,1,float16,float16,0,0.03914400041103363
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,32,32,128,1,float16,float16,0,0.02847520112991333
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,32,1,128,1,float16,float16,0,0.02218399941921234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,32,4,128,1,float16,float16,0,0.037412801384925844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,32,2,128,1,float16,float16,0,0.022043199837207796
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,32,4,128,1,float16,float16,0,0.022644799947738648
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,32,8,128,1,float16,float16,0,0.023537600040435792
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,32,1,128,1,float16,float16,0,0.01377439945936203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,32,2,128,1,float16,float16,0,0.01372160017490387
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,32,4,128,1,float16,float16,0,0.013868799805641175
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,32,8,128,1,float16,float16,0,0.01432960033416748
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,32,32,128,1,float16,float16,0,0.013872000575065612
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,32,1,128,1,float16,float16,0,0.0123648002743721
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,32,2,128,1,float16,float16,0,0.012412799894809723
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,32,4,128,1,float16,float16,0,0.01239520013332367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,32,8,128,1,float16,float16,0,0.012564800679683685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,32,32,128,1,float16,float16,0,0.012779200077056884
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,32,1,128,1,float16,float16,0,0.011939200013875962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,32,2,128,1,float16,float16,0,0.012036799639463424
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,32,4,128,1,float16,float16,0,0.012064000219106674
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,32,1,128,1,float16,float16,0,0.017360000312328337
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,32,8,128,1,float16,float16,0,0.012121599912643433
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,32,1,128,1,float16,float16,0,0.44826879501342776
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,32,1,128,1,float16,float16,0,0.4462319850921631
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,32,2,128,1,float16,float16,0,0.48220000267028806
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,32,4,128,1,float16,float16,0,0.5541744232177734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,32,1,128,1,float16,float16,0,0.20984480381011963
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,32,32,128,1,float16,float16,0,0.3514111995697021
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,32,32,128,1,float16,float16,0,0.05538079738616943
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,32,8,128,1,float16,float16,0,0.7364528179168701
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,32,1,128,1,float16,float16,0,0.03618240058422088
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,32,2,128,1,float16,float16,0,0.22333118915557862
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,32,32,128,1,float16,float16,0,0.8170576095581055
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,32,1,128,1,float16,float16,0,0.06150720119476318
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,32,4,128,1,float16,float16,0,0.25555360317230225
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,32,8,128,1,float16,float16,0,0.3275023937225342
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,32,32,128,1,float16,float16,0,0.35167200565338136
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,32,8,128,1,float16,float16,0,0.07472320199012757
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,32,4,128,1,float16,float16,0,0.0637391984462738
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,32,32,128,1,float16,float16,0,0.0814415991306305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,32,1,128,1,float16,float16,0,0.03560799956321716
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,32,4,128,1,float16,float16,0,0.03629119992256165
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,32,8,128,1,float16,float16,0,0.03785600066184998
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,32,2,128,1,float16,float16,0,0.0355103999376297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,32,32,128,1,float16,float16,0,0.037478399276733396
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,32,2,128,1,float16,float16,0,0.02090719938278198
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,32,1,128,1,float16,float16,0,0.021076799929142
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,32,32,128,1,float16,float16,0,0.021875199675559998
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,32,1,128,1,float16,float16,0,0.013331200182437896
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,32,2,128,1,float16,float16,0,0.013521599769592284
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,32,4,128,1,float16,float16,0,0.013707199692726135
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,32,8,128,1,float16,float16,0,0.013928000628948212
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,32,32,128,1,float16,float16,0,0.013822400569915771
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,32,2,128,1,float16,float16,0,0.012174399942159653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,32,1,128,1,float16,float16,0,0.012272000312805176
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,32,4,128,1,float16,float16,0,0.012403199821710587
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,32,8,128,1,float16,float16,0,0.012368000298738479
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,32,32,128,1,float16,float16,0,0.012457600235939026
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,32,1,128,1,float16,float16,0,0.011910399794578553
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,32,4,128,1,float16,float16,0,0.012049599736928939
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,32,8,128,1,float16,float16,0,0.012095999717712403
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,32,32,128,1,float16,float16,0,0.012187200039625168
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,32,1,128,1,float16,float16,0,0.01191679984331131
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,32,2,128,1,float16,float16,0,0.011977600306272507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,32,4,128,1,float16,float16,0,0.011974400281906128
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,32,8,128,1,float16,float16,0,0.012004800140857697
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,32,1,128,1,float16,float16,0,0.21421918869018555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,32,2,128,1,float16,float16,0,0.2278656005859375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,32,4,128,1,float16,float16,0,0.2556688070297241
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,32,8,128,1,float16,float16,0,0.31691200733184816
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,32,1,128,1,float16,float16,0,0.06172320246696472
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,32,2,128,1,float16,float16,0,0.06278560161590577
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,32,32,128,1,float16,float16,0,0.3718640089035034
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,32,2,128,1,float16,float16,0,0.06256319880485535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,32,32,128,1,float16,float16,0,0.016812799870967864
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,32,4,128,1,float16,float16,0,0.06465920209884643
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,32,8,128,1,float16,float16,0,0.08225280046463013
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,32,1,128,1,float16,float16,0,0.03537600040435791
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,32,32,128,1,float16,float16,0,0.11459200382232666
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,32,2,128,1,float16,float16,0,0.03565919995307922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,32,8,128,1,float16,float16,0,0.03767200112342835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,32,32,128,1,float16,float16,0,0.05940319895744324
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,32,1,128,1,float16,float16,0,0.02131199985742569
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,32,2,128,1,float16,float16,0,0.020785599946975708
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,32,4,128,1,float16,float16,0,0.021583999693393707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,32,8,128,1,float16,float16,0,0.022361600399017335
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,32,4,128,1,float16,float16,0,0.021345600485801697
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,32,8,128,1,float16,float16,0,0.02229280024766922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,32,32,128,1,float16,float16,0,0.032753598690032956
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,32,1,128,1,float16,float16,0,0.013238400220870972
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,32,2,128,1,float16,float16,0,0.013116799294948578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,32,4,128,1,float16,float16,0,0.013447999954223633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,32,8,128,1,float16,float16,0,0.01408800035715103
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,32,4,128,1,float16,float16,0,0.012140800058841706
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,32,32,128,1,float16,float16,0,0.012275200337171555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,32,8,128,1,float16,float16,0,0.0125231996178627
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,32,2,128,1,float16,float16,0,0.011761599779129028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,32,4,128,1,float16,float16,0,0.011894399672746659
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,32,2,128,1,float16,float16,0,0.011915200203657151
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,32,8,128,1,float16,float16,0,0.012027200311422348
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,32,32,128,1,float16,float16,0,0.011955200135707856
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,32,1,128,1,float16,float16,0,0.011823999881744384
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,32,2,128,1,float16,float16,0,0.011873599886894227
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,32,4,128,1,float16,float16,0,0.011819200217723846
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,32,8,128,1,float16,float16,0,0.011908800154924393
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,32,32,128,1,float16,float16,0,0.011736000329256058
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,32,1,128,1,float16,float16,0,0.011843200027942657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,32,2,128,1,float16,float16,0,0.011800000071525573
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,32,4,128,1,float16,float16,0,0.011857599765062333
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,32,8,128,1,float16,float16,0,0.011716800183057785
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,32,1,128,1,float16,float16,0,0.06270880103111268
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,32,4,128,1,float16,float16,0,0.06506720185279846
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,32,8,128,1,float16,float16,0,0.12303040027618409
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,32,32,128,1,float16,float16,0,0.21574718952178956
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,32,1,128,1,float16,float16,0,0.035841599106788635
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,32,2,128,1,float16,float16,0,0.035872000455856326
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,32,4,128,1,float16,float16,0,0.036575999855995175
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,32,4,128,1,float16,float16,0,0.03614720106124878
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,32,8,128,1,float16,float16,0,0.05950400233268738
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,32,1,128,1,float16,float16,0,0.02136639952659607
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,32,32,128,1,float16,float16,0,0.09995520114898682
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,32,2,128,1,float16,float16,0,0.021238400042057036
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,32,4,128,1,float16,float16,0,0.02162559926509857
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,32,8,128,1,float16,float16,0,0.032950401306152344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,32,32,128,1,float16,float16,0,0.055606400966644286
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,32,1,128,1,float16,float16,0,0.012027200311422348
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,32,32,128,1,float16,float16,0,0.019465599954128266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,32,1,128,1,float16,float16,0,0.013531200587749481
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,32,2,128,1,float16,float16,0,0.013465599715709686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,32,32,128,1,float16,float16,0,0.030399999022483824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,32,4,128,1,float16,float16,0,0.013921600580215455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,32,1,128,1,float16,float16,0,0.011931200325489045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,32,1,128,1,float16,float16,0,0.011664000153541566
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,32,2,128,1,float16,float16,0,0.01210239976644516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,32,8,128,1,float16,float16,0,0.012118399888277055
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,32,32,128,1,float16,float16,0,0.017903999984264375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,32,1,128,1,float16,float16,0,0.011638399958610535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,32,2,128,1,float16,float16,0,0.011684799939393998
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,32,4,128,1,float16,float16,0,0.011697600036859513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,32,32,128,1,float16,float16,0,0.011905600130558015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,32,1,128,1,float16,float16,0,0.011593600362539291
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,32,2,128,1,float16,float16,0,0.011608000099658965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,32,4,128,1,float16,float16,0,0.011649599671363831
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,32,8,128,1,float16,float16,0,0.011240000277757645
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,32,32,128,1,float16,float16,0,0.011627200245857238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,32,1,128,1,float16,float16,0,0.011587200313806533
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,32,2,128,1,float16,float16,0,0.011606399714946748
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,32,4,128,1,float16,float16,0,0.011644800007343293
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,32,8,128,1,float16,float16,0,0.011215999722480774
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,32,32,128,1,float16,float16,0,0.011505600064992905
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,32,1,128,1,float16,float16,0,0.011523199826478958
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,32,2,128,1,float16,float16,0,0.011556799709796905
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,32,4,128,1,float16,float16,0,0.01162080019712448
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,32,8,128,1,float16,float16,0,0.011150400340557098
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,32,1,128,1,float16,float16,0,0.019648000597953796
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,32,2,128,1,float16,float16,0,0.03147040009498596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,32,2,128,1,float16,float16,0,0.06392480134963989
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,32,4,128,1,float16,float16,0,0.055150401592254636
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,32,8,128,1,float16,float16,0,0.09630879759788513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,32,1,128,1,float16,float16,0,0.012587200105190276
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,32,2,128,1,float16,float16,0,0.01851679980754852
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,32,32,128,1,float16,float16,0,0.17695200443267822
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,32,4,128,1,float16,float16,0,0.0303600013256073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,32,8,128,1,float16,float16,0,0.05392799973487854
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,32,32,128,1,float16,float16,0,0.09423840045928955
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,32,2,128,1,float16,float16,0,0.011992000043392181
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,32,4,128,1,float16,float16,0,0.017926399409770966
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,32,8,128,1,float16,float16,0,0.029684799909591674
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,32,32,128,1,float16,float16,0,0.052960002422332765
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,32,2,128,1,float16,float16,0,0.011417599767446518
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,32,4,128,1,float16,float16,0,0.01162559986114502
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,32,8,128,1,float16,float16,0,0.017539200186729432
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,32,32,128,1,float16,float16,0,0.029281601309776306
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,32,1,128,1,float16,float16,0,0.010958400368690491
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,32,2,128,1,float16,float16,0,0.010977599769830704
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,32,4,128,1,float16,float16,0,0.011083199828863143
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,32,2,128,1,float16,float16,0,0.012121599912643433
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,32,32,128,1,float16,float16,0,0.0178384006023407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,32,8,128,1,float16,float16,0,0.019704000651836397
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,32,1,128,1,float16,float16,0,0.011231999844312668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,32,4,128,1,float16,float16,0,0.012409599870443344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,32,4,128,1,float16,float16,0,0.01125119999051094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,32,8,128,1,float16,float16,0,0.011475200206041336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,32,8,128,1,float16,float16,0,0.011313600093126297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,32,1,128,1,float16,float16,0,0.010838399827480315
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,32,32,128,1,float16,float16,0,0.012083200365304947
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,32,4,128,1,float16,float16,0,0.010889600217342376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,32,2,128,1,float16,float16,0,0.011163199692964554
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,32,32,128,1,float16,float16,0,0.011531200259923935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,32,8,128,1,float16,float16,0,0.011619199812412263
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,32,2,128,1,float16,float16,0,0.010902400314807891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,32,1,128,1,float16,float16,0,0.011043199896812439
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,32,8,128,1,float16,float16,0,0.010894399881362916
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,32,4,128,1,float16,float16,0,0.011273600161075592
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,32,32,128,1,float16,float16,0,0.0117296002805233
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,32,2,128,1,float16,float16,0,0.011502400040626526
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,32,4,128,1,float16,float16,0,0.011140800267457961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,32,8,128,1,float16,float16,0,0.011195199936628342
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,32,1,128,1,float16,float16,0,0.011767999827861786
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,32,1,128,1,float16,float16,0,0.011313600093126297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,32,8,128,1,float16,float16,0,0.011371199786663056
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,32,2,128,1,float16,float16,0,0.011169599741697312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,32,1,128,1,float16,float16,0,0.010812799632549285
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,24,1,128,1,float16,float16,0,30.783889770507812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,24,2,128,1,float16,float16,0,31.70635070800781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,24,4,128,1,float16,float16,0,31.165530395507812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,24,1,128,1,float16,float16,0,16.215902709960936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,24,1,128,1,float16,float16,0,63.480767822265626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,24,4,128,1,float16,float16,0,64.21924438476563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,24,2,128,1,float16,float16,0,64.74091796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,24,8,128,1,float16,float16,0,65.28103637695312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,24,2,128,1,float16,float16,0,15.315632629394532
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,24,8,128,1,float16,float16,0,33.416586303710936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,24,1,128,1,float16,float16,0,7.540914916992188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,24,2,128,1,float16,float16,0,7.61466064453125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,24,4,128,1,float16,float16,0,16.528326416015624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,24,8,128,1,float16,float16,0,16.4905517578125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,24,4,128,1,float16,float16,0,8.195139312744141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,24,8,128,1,float16,float16,0,8.293534088134766
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,24,1,128,1,float16,float16,0,35.494384765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,24,2,128,1,float16,float16,0,36.620034790039064
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,24,4,128,1,float16,float16,0,36.99947509765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,24,8,128,1,float16,float16,0,37.942535400390625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,24,1,128,1,float16,float16,0,18.697390747070312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,24,2,128,1,float16,float16,0,18.460173034667967
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,24,4,128,1,float16,float16,0,18.48754119873047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,24,8,128,1,float16,float16,0,18.814593505859374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,24,1,128,1,float16,float16,0,8.88329620361328
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,24,2,128,1,float16,float16,0,8.678559875488281
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,24,4,128,1,float16,float16,0,9.029817962646485
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,24,8,128,1,float16,float16,0,9.787510681152344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,24,1,128,1,float16,float16,0,4.276358413696289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,24,2,128,1,float16,float16,0,4.414630508422851
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,24,4,128,1,float16,float16,0,4.620956802368164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,24,8,128,1,float16,float16,0,4.676903915405274
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,24,1,128,1,float16,float16,0,25.223130798339845
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,24,2,128,1,float16,float16,0,25.835751342773438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,24,4,128,1,float16,float16,0,26.233197021484376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,24,8,128,1,float16,float16,0,27.185763549804687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,24,1,128,1,float16,float16,0,12.868630981445312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,24,2,128,1,float16,float16,0,12.752433776855469
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,24,4,128,1,float16,float16,0,13.279344177246093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,24,24,128,1,float16,float16,0,4.460750579833984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,24,8,128,1,float16,float16,0,13.627389526367187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,24,24,128,1,float16,float16,0,9.305897521972657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,24,24,128,1,float16,float16,0,8.181966400146484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,24,24,128,1,float16,float16,0,6.716305541992187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,24,24,128,1,float16,float16,0,13.917640686035156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,24,1,128,1,float16,float16,0,6.19330062866211
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,24,24,128,1,float16,float16,0,3.0795215606689452
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,24,2,128,1,float16,float16,0,6.093423843383789
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,24,4,128,1,float16,float16,0,6.294038391113281
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,24,8,128,1,float16,float16,0,6.261360168457031
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,24,1,128,1,float16,float16,0,2.9149999618530273
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,24,2,128,1,float16,float16,0,2.920894432067871
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,24,4,128,1,float16,float16,0,2.9586463928222657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,24,8,128,1,float16,float16,0,3.0878400802612305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,24,24,128,1,float16,float16,0,19.879017639160157
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,24,24,128,1,float16,float16,0,16.861004638671876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,24,24,128,1,float16,float16,0,33.546810913085935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,24,1,128,1,float16,float16,0,15.734805297851562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,24,24,128,1,float16,float16,0,18.945883178710936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,24,2,128,1,float16,float16,0,16.337464904785158
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,24,4,128,1,float16,float16,0,17.299763488769532
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,24,1,128,1,float16,float16,0,8.529827117919922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,24,24,128,1,float16,float16,0,9.699273681640625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,24,1,128,1,float16,float16,0,32.709912109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,24,2,128,1,float16,float16,0,34.08926086425781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,24,4,128,1,float16,float16,0,34.84566650390625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,24,24,128,1,float16,float16,0,4.387814331054687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,24,8,128,1,float16,float16,0,17.8503662109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,24,1,128,1,float16,float16,0,3.670310211181641
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,24,8,128,1,float16,float16,0,36.12675476074219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,24,2,128,1,float16,float16,0,7.811682891845703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,24,4,128,1,float16,float16,0,8.694601440429688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,24,24,128,1,float16,float16,0,2.1102848052978516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,24,1,128,1,float16,float16,0,2.1068607330322267
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,24,8,128,1,float16,float16,0,8.732222747802734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,24,2,128,1,float16,float16,0,1.9437471389770509
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,24,2,128,1,float16,float16,0,3.5932350158691406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,24,4,128,1,float16,float16,0,3.7414207458496094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,24,8,128,1,float16,float16,0,4.016223907470703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,24,4,128,1,float16,float16,0,2.083273506164551
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,24,8,128,1,float16,float16,0,1.9414207458496093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,24,1,128,1,float16,float16,0,8.954251098632813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,24,2,128,1,float16,float16,0,9.365289306640625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,24,4,128,1,float16,float16,0,9.801811218261719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,24,24,128,1,float16,float16,0,10.711965179443359
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,24,1,128,1,float16,float16,0,4.418288040161133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,24,2,128,1,float16,float16,0,4.468700790405274
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,24,24,128,1,float16,float16,0,5.238353729248047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,24,1,128,1,float16,float16,0,18.854656982421876
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,24,2,128,1,float16,float16,0,19.04557647705078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,24,24,128,1,float16,float16,0,2.5638656616210938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,24,4,128,1,float16,float16,0,20.132571411132812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,24,8,128,1,float16,float16,0,9.776132965087891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,24,8,128,1,float16,float16,0,20.02568817138672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,24,4,128,1,float16,float16,0,4.797996902465821
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,24,1,128,1,float16,float16,0,2.1063791275024415
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,24,2,128,1,float16,float16,0,2.2419151306152343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,24,8,128,1,float16,float16,0,4.742153549194336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,24,24,128,1,float16,float16,0,1.294542407989502
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,24,1,128,1,float16,float16,0,1.0901408195495605
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,24,4,128,1,float16,float16,0,2.098512077331543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,24,4,128,1,float16,float16,0,1.123249626159668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,24,8,128,1,float16,float16,0,2.3259695053100584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,24,2,128,1,float16,float16,0,1.1626591682434082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,24,8,128,1,float16,float16,0,1.2415679931640624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,24,1,128,1,float16,float16,0,8.260540771484376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,24,2,128,1,float16,float16,0,8.868958282470704
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,24,4,128,1,float16,float16,0,9.371353912353516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,24,24,128,1,float16,float16,0,10.819380950927734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,24,1,128,1,float16,float16,0,4.168246459960938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,24,24,128,1,float16,float16,0,5.078497695922851
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,24,2,128,1,float16,float16,0,4.269121551513672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,24,1,128,1,float16,float16,0,17.553904724121093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,24,2,128,1,float16,float16,0,17.77226867675781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,24,4,128,1,float16,float16,0,19.063414001464842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,24,8,128,1,float16,float16,0,9.136563110351563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,24,4,128,1,float16,float16,0,4.2281230926513675
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,24,8,128,1,float16,float16,0,19.690956115722656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,24,1,128,1,float16,float16,0,2.0686431884765626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,24,24,128,1,float16,float16,0,2.4738143920898437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,24,2,128,1,float16,float16,0,1.9322383880615235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,24,4,128,1,float16,float16,0,1.913528060913086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,24,1,128,1,float16,float16,0,1.0111616134643555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,24,2,128,1,float16,float16,0,0.9664383888244629
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,24,8,128,1,float16,float16,0,4.530774307250977
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,24,24,128,1,float16,float16,0,1.283796787261963
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,24,1,128,1,float16,float16,0,0.5015744209289551
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,24,24,128,1,float16,float16,0,0.662559986114502
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,24,8,128,1,float16,float16,0,2.2017391204833983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,24,4,128,1,float16,float16,0,1.0487520217895507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,24,8,128,1,float16,float16,0,1.0548895835876464
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,24,2,128,1,float16,float16,0,0.6344799995422363
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,24,4,128,1,float16,float16,0,0.5388944149017334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,24,8,128,1,float16,float16,0,0.5084928035736084
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,24,2,128,1,float16,float16,0,4.181577682495117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,24,1,128,1,float16,float16,0,4.806568145751953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,24,24,128,1,float16,float16,0,6.242529678344726
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,24,4,128,1,float16,float16,0,4.76692008972168
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,24,24,128,1,float16,float16,0,3.0505903244018553
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,24,1,128,1,float16,float16,0,9.656028747558594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,24,1,128,1,float16,float16,0,2.1204511642456056
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,24,2,128,1,float16,float16,0,10.326952362060547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,24,8,128,1,float16,float16,0,5.207235336303711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,24,4,128,1,float16,float16,0,10.443408203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,24,2,128,1,float16,float16,0,2.2969152450561525
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,24,8,128,1,float16,float16,0,11.091732788085938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,24,4,128,1,float16,float16,0,2.2714111328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,24,24,128,1,float16,float16,0,1.5878239631652833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,24,2,128,1,float16,float16,0,1.1532992362976073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,24,1,128,1,float16,float16,0,1.2293487548828126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,24,4,128,1,float16,float16,0,1.1757391929626464
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,24,8,128,1,float16,float16,0,2.6307727813720705
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,24,8,128,1,float16,float16,0,1.3334239959716796
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,24,1,128,1,float16,float16,0,0.6905248165130615
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,24,2,128,1,float16,float16,0,0.5482751846313476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,24,24,128,1,float16,float16,0,0.8581744194030761
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,24,4,128,1,float16,float16,0,0.6079967975616455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,24,8,128,1,float16,float16,0,0.3193376064300537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,24,4,128,1,float16,float16,0,0.3495887994766235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,24,24,128,1,float16,float16,0,0.4281280040740967
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,24,8,128,1,float16,float16,0,0.694652795791626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,24,1,128,1,float16,float16,0,0.36192800998687746
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,24,2,128,1,float16,float16,0,0.33965120315551756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,24,1,128,1,float16,float16,0,4.241680145263672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,24,2,128,1,float16,float16,0,4.190142440795898
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,24,4,128,1,float16,float16,0,4.781059265136719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,24,24,128,1,float16,float16,0,6.312031936645508
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,24,1,128,1,float16,float16,0,2.153001594543457
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,24,24,128,1,float16,float16,0,3.166329574584961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,24,1,128,1,float16,float16,0,8.891107177734375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,24,2,128,1,float16,float16,0,9.284587097167968
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,24,2,128,1,float16,float16,0,2.044772720336914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,24,4,128,1,float16,float16,0,10.200078582763672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,24,8,128,1,float16,float16,0,5.2541553497314455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,24,4,128,1,float16,float16,0,2.2604976654052735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,24,24,128,1,float16,float16,0,1.7071727752685546
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,24,2,128,1,float16,float16,0,1.1525584220886231
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,24,1,128,1,float16,float16,0,1.0493200302124024
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,24,8,128,1,float16,float16,0,2.2926656723022463
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,24,24,128,1,float16,float16,0,0.851689624786377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,24,4,128,1,float16,float16,0,0.5543439865112305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,24,1,128,1,float16,float16,0,0.5624752044677734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,24,4,128,1,float16,float16,0,1.0996607780456542
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,24,8,128,1,float16,float16,0,1.2256671905517578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,24,24,128,1,float16,float16,0,0.42671680450439453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,24,2,128,1,float16,float16,0,0.2904112100601196
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,24,1,128,1,float16,float16,0,0.27822558879852294
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,24,8,128,1,float16,float16,0,0.6212351799011231
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,24,1,128,1,float16,float16,0,0.1701151967048645
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,24,2,128,1,float16,float16,0,0.16949599981307983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,24,4,128,1,float16,float16,0,0.1843503952026367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,24,4,128,1,float16,float16,0,0.3134912014007568
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,24,24,128,1,float16,float16,0,0.19685280323028564
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,24,8,128,1,float16,float16,0,0.3150464057922363
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,24,8,128,1,float16,float16,0,0.17501120567321776
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,24,8,128,1,float16,float16,0,10.93605728149414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,24,2,128,1,float16,float16,0,0.590123176574707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,24,1,128,1,float16,float16,0,2.261895942687988
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,24,24,128,1,float16,float16,0,4.0819343566894535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,24,2,128,1,float16,float16,0,2.2338911056518556
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,24,1,128,1,float16,float16,0,5.289023971557617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,24,4,128,1,float16,float16,0,2.461419105529785
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,24,4,128,1,float16,float16,0,5.842705535888672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,24,8,128,1,float16,float16,0,2.8891632080078127
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,24,8,128,1,float16,float16,0,5.907527923583984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,24,1,128,1,float16,float16,0,1.236398410797119
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,24,2,128,1,float16,float16,0,1.335916805267334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,24,24,128,1,float16,float16,0,2.0460512161254885
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,24,2,128,1,float16,float16,0,0.683678388595581
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,24,24,128,1,float16,float16,0,1.0508864402770997
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,24,1,128,1,float16,float16,0,0.6732768058776856
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,24,8,128,1,float16,float16,0,1.4815296173095702
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,24,4,128,1,float16,float16,0,0.7274767875671386
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,24,24,128,1,float16,float16,0,0.5251967906951904
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,24,8,128,1,float16,float16,0,0.7043471813201905
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,24,1,128,1,float16,float16,0,0.3067807912826538
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,24,8,128,1,float16,float16,0,0.3738640069961548
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,24,4,128,1,float16,float16,0,0.3574368000030518
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,24,24,128,1,float16,float16,0,0.26072158813476565
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,24,2,128,1,float16,float16,0,0.3350608110427856
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,24,2,128,1,float16,float16,0,0.17666560411453247
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,24,24,128,1,float16,float16,0,0.1387104034423828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,24,8,128,1,float16,float16,0,0.18930399417877197
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,24,4,128,1,float16,float16,0,0.1989680051803589
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,24,1,128,1,float16,float16,0,0.11191519498825073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,24,2,128,1,float16,float16,0,0.11698880195617675
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,24,2,128,1,float16,float16,0,5.411108779907226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,24,8,128,1,float16,float16,0,0.13264960050582886
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,24,4,128,1,float16,float16,0,1.3686047554016114
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,24,1,128,1,float16,float16,0,0.17758079767227172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,24,4,128,1,float16,float16,0,0.11318559646606445
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,24,1,128,1,float16,float16,0,5.183644866943359
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,24,2,128,1,float16,float16,0,5.136334228515625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,24,1,128,1,float16,float16,0,2.220857620239258
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,24,4,128,1,float16,float16,0,5.502406311035156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,24,2,128,1,float16,float16,0,2.2020816802978516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,24,8,128,1,float16,float16,0,6.237228775024414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,24,24,128,1,float16,float16,0,4.8748016357421875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,24,1,128,1,float16,float16,0,1.1913408279418944
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,24,2,128,1,float16,float16,0,1.1899359703063965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,24,4,128,1,float16,float16,0,2.7757440567016602
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,24,4,128,1,float16,float16,0,1.296012783050537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,24,24,128,1,float16,float16,0,2.5385791778564455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,24,2,128,1,float16,float16,0,0.5921296119689942
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,24,1,128,1,float16,float16,0,0.6570623874664306
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,24,8,128,1,float16,float16,0,3.0228191375732423
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,24,24,128,1,float16,float16,0,1.2645759582519531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,24,8,128,1,float16,float16,0,1.4602255821228027
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,24,1,128,1,float16,float16,0,0.3070847988128662
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,24,2,128,1,float16,float16,0,0.34203999042510985
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,24,4,128,1,float16,float16,0,0.6877711772918701
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,24,1,128,1,float16,float16,0,0.153331196308136
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,24,2,128,1,float16,float16,0,0.16831040382385254
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,24,24,128,1,float16,float16,0,0.6861616134643554
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,24,8,128,1,float16,float16,0,0.7808784008026123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,24,24,128,1,float16,float16,0,0.305347204208374
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,24,4,128,1,float16,float16,0,0.18209600448608398
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,24,24,128,1,float16,float16,0,0.11528639793395996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,24,4,128,1,float16,float16,0,0.36793758869171145
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,24,4,128,1,float16,float16,0,0.10581760406494141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,24,1,128,1,float16,float16,0,0.09280480146408081
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,24,24,128,1,float16,float16,0,0.06845920085906983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,24,8,128,1,float16,float16,0,0.11598720550537109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,24,1,128,1,float16,float16,0,0.062319999933242796
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,24,2,128,1,float16,float16,0,0.058534401655197146
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,24,4,128,1,float16,float16,0,0.0574400007724762
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,24,8,128,1,float16,float16,0,0.06046720147132874
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,24,8,128,1,float16,float16,0,0.3853456020355225
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,24,8,128,1,float16,float16,0,0.16441760063171387
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,24,2,128,1,float16,float16,0,0.10210720300674439
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,24,1,128,1,float16,float16,0,3.5668750762939454
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,24,2,128,1,float16,float16,0,3.5673614501953126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,24,4,128,1,float16,float16,0,4.142633438110352
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,24,1,128,1,float16,float16,0,1.6242944717407226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,24,2,128,1,float16,float16,0,1.7724735260009765
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,24,24,128,1,float16,float16,0,5.203168106079102
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,24,8,128,1,float16,float16,0,5.3935089111328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,24,4,128,1,float16,float16,0,2.0620832443237305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,24,1,128,1,float16,float16,0,0.8361743927001953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,24,2,128,1,float16,float16,0,0.882356834411621
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,24,4,128,1,float16,float16,0,1.0409664154052733
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,24,1,128,1,float16,float16,0,0.43674240112304685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,24,8,128,1,float16,float16,0,2.7034656524658205
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,24,2,128,1,float16,float16,0,0.4518896102905273
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,24,4,128,1,float16,float16,0,0.5089568138122559
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,24,24,128,1,float16,float16,0,2.739681625366211
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,24,8,128,1,float16,float16,0,1.3727519989013672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,24,24,128,1,float16,float16,0,1.3902400016784668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,24,2,128,1,float16,float16,0,0.2415695905685425
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,24,8,128,1,float16,float16,0,0.7019120216369629
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,24,24,128,1,float16,float16,0,0.7264416217803955
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,24,1,128,1,float16,float16,0,0.23632960319519042
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,24,24,128,1,float16,float16,0,0.2990351915359497
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,24,1,128,1,float16,float16,0,0.10697439908981324
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,24,8,128,1,float16,float16,0,0.3278975963592529
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,24,2,128,1,float16,float16,0,0.10973919630050659
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,24,24,128,1,float16,float16,0,0.09295520186424255
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,24,8,128,1,float16,float16,0,0.12244800329208375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,24,4,128,1,float16,float16,0,0.11231839656829834
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,24,2,128,1,float16,float16,0,0.06182240247726441
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,24,1,128,1,float16,float16,0,0.06174880266189575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,24,4,128,1,float16,float16,0,0.06299999952316285
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,24,24,128,1,float16,float16,0,0.056576001644134524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,24,8,128,1,float16,float16,0,0.06552479863166809
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,24,1,128,1,float16,float16,0,0.03561759889125824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,24,4,128,1,float16,float16,0,0.036353600025177
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,24,8,128,1,float16,float16,0,0.03782880008220672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,24,2,128,1,float16,float16,0,0.03636960089206696
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,24,24,128,1,float16,float16,0,0.04549280107021332
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,24,1,128,1,float16,float16,0,0.036134400963783266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,24,2,128,1,float16,float16,0,0.03824639916419983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,24,4,128,1,float16,float16,0,0.03901279866695404
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,24,1,128,1,float16,float16,0,1.5428624153137207
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,24,2,128,1,float16,float16,0,1.6977951049804687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,24,4,128,1,float16,float16,0,0.26353600025177004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,24,4,128,1,float16,float16,0,2.0761951446533202
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,24,24,128,1,float16,float16,0,2.4847391128540037
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,24,8,128,1,float16,float16,0,2.6940431594848633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,24,2,128,1,float16,float16,0,0.8436191558837891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,24,8,128,1,float16,float16,0,0.043172800540924074
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,24,4,128,1,float16,float16,0,0.9669232368469238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,24,1,128,1,float16,float16,0,0.3736943960189819
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,24,2,128,1,float16,float16,0,0.37083361148834226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,24,8,128,1,float16,float16,0,1.3625167846679687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,24,4,128,1,float16,float16,0,0.4495984077453613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,24,1,128,1,float16,float16,0,0.1442752003669739
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,24,8,128,1,float16,float16,0,0.619271993637085
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,24,24,128,1,float16,float16,0,1.2448975563049316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,24,4,128,1,float16,float16,0,0.1894927978515625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,24,2,128,1,float16,float16,0,0.17812800407409668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,24,8,128,1,float16,float16,0,0.26219360828399657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,24,2,128,1,float16,float16,0,0.06344159841537475
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,24,8,128,1,float16,float16,0,0.07007200121879578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,24,24,128,1,float16,float16,0,0.6697440147399902
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,24,1,128,1,float16,float16,0,0.0628383994102478
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,24,4,128,1,float16,float16,0,0.06680960059165955
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,24,4,128,1,float16,float16,0,0.03800320029258728
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,24,1,128,1,float16,float16,0,0.03752639889717102
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,24,8,128,1,float16,float16,0,0.039996799826622007
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,24,24,128,1,float16,float16,0,0.029284799098968507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,24,1,128,1,float16,float16,0,0.024379199743270873
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,24,4,128,1,float16,float16,0,0.025723201036453248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,24,8,128,1,float16,float16,0,0.026214399933815004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,24,2,128,1,float16,float16,0,0.02475520074367523
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,24,24,128,1,float16,float16,0,0.019166399538517
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,24,1,128,1,float16,float16,0,0.016686399281024934
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,24,4,128,1,float16,float16,0,0.017003199458122252
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,24,8,128,1,float16,float16,0,0.017382399737834932
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,24,24,128,1,float16,float16,0,0.017323200404644013
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,24,2,128,1,float16,float16,0,0.016732800006866454
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,24,1,128,1,float16,float16,0,0.016176000237464905
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,24,2,128,1,float16,float16,0,0.016176000237464905
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,24,8,128,1,float16,float16,0,0.016353599727153778
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,24,4,128,1,float16,float16,0,0.016131199896335602
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,24,1,128,1,float16,float16,0,0.7574607849121093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,24,1,128,1,float16,float16,0,0.6951119899749756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,24,2,128,1,float16,float16,0,0.7667312145233154
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,24,4,128,1,float16,float16,0,0.9522944450378418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,24,1,128,1,float16,float16,0,0.33797440528869627
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,24,8,128,1,float16,float16,0,1.2538576126098633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,24,2,128,1,float16,float16,0,0.371943998336792
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,24,24,128,1,float16,float16,0,0.20524320602416993
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,24,4,128,1,float16,float16,0,0.4473135948181152
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,24,1,128,1,float16,float16,0,0.14852800369262695
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,24,2,128,1,float16,float16,0,0.16139039993286133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,24,8,128,1,float16,float16,0,0.6265679836273194
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,24,4,128,1,float16,float16,0,0.1889888048171997
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,24,2,128,1,float16,float16,0,0.0368800014257431
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,24,24,128,1,float16,float16,0,0.6181600093841553
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,24,8,128,1,float16,float16,0,0.25451838970184326
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,24,2,128,1,float16,float16,0,0.05124160051345825
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,24,24,128,1,float16,float16,0,0.20320639610290528
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,24,4,128,1,float16,float16,0,0.0528223991394043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,24,2,128,1,float16,float16,0,0.029967999458312987
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,24,1,128,1,float16,float16,0,0.050209599733352664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,24,4,128,1,float16,float16,0,0.03097119927406311
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,24,8,128,1,float16,float16,0,0.032304000854492185
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,24,24,128,1,float16,float16,0,0.025924798846244813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,24,1,128,1,float16,float16,0,0.0208624005317688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,24,2,128,1,float16,float16,0,0.020761600136756896
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,24,4,128,1,float16,float16,0,0.021620799601078034
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,24,8,128,1,float16,float16,0,0.02218399941921234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,24,24,128,1,float16,float16,0,0.015475200116634369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,24,1,128,1,float16,float16,0,0.01313440054655075
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,24,2,128,1,float16,float16,0,0.01311199963092804
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,24,4,128,1,float16,float16,0,0.013254399597644805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,24,8,128,1,float16,float16,0,0.01383039951324463
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,24,24,128,1,float16,float16,0,0.013467200100421906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,24,1,128,1,float16,float16,0,0.012163200229406358
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,24,2,128,1,float16,float16,0,0.012244799733161926
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,24,4,128,1,float16,float16,0,0.012345600128173827
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,24,8,128,1,float16,float16,0,0.012521600723266602
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,24,24,128,1,float16,float16,0,0.012620800733566284
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,24,1,128,1,float16,float16,0,0.011932799965143204
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,24,4,128,1,float16,float16,0,0.012015999853610992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,24,2,128,1,float16,float16,0,0.011964800208806992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,24,1,128,1,float16,float16,0,0.3401968002319336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,24,2,128,1,float16,float16,0,0.3729664087295532
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,24,4,128,1,float16,float16,0,0.4490960121154785
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,24,8,128,1,float16,float16,0,0.635811185836792
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,24,24,128,1,float16,float16,0,0.045870399475097655
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,24,24,128,1,float16,float16,0,1.2402671813964843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,24,1,128,1,float16,float16,0,0.1331007957458496
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,24,2,128,1,float16,float16,0,0.16403199434280397
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,24,24,128,1,float16,float16,0,0.6115600109100342
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,24,8,128,1,float16,float16,0,0.25764479637146
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,24,24,128,1,float16,float16,0,0.21841599941253662
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,24,1,128,1,float16,float16,0,0.049025601148605345
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,24,2,128,1,float16,float16,0,0.050254398584365846
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,24,8,128,1,float16,float16,0,0.06542239785194397
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,24,4,128,1,float16,float16,0,0.05204960107803345
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,24,24,128,1,float16,float16,0,0.05341759920120239
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,24,1,128,1,float16,float16,0,0.028910401463508605
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,24,2,128,1,float16,float16,0,0.028889599442481994
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,24,8,128,1,float16,float16,0,0.03647040128707886
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,24,24,128,1,float16,float16,0,0.029785600304603577
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,24,1,128,1,float16,float16,0,0.020129600167274476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,24,8,128,1,float16,float16,0,0.012078399956226348
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,24,2,128,1,float16,float16,0,0.020316800475120543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,24,4,128,1,float16,float16,0,0.0210207998752594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,24,8,128,1,float16,float16,0,0.021359999477863312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,24,24,128,1,float16,float16,0,0.02078399956226349
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,24,1,128,1,float16,float16,0,0.01279519945383072
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,24,4,128,1,float16,float16,0,0.013020800054073333
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,24,2,128,1,float16,float16,0,0.013121600449085235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,24,8,128,1,float16,float16,0,0.013476799428462981
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,24,24,128,1,float16,float16,0,0.013155199587345123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,24,1,128,1,float16,float16,0,0.011926399916410447
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,24,2,128,1,float16,float16,0,0.0122079998254776
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,24,4,128,1,float16,float16,0,0.012065599858760833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,24,24,128,1,float16,float16,0,0.01202400028705597
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,24,8,128,1,float16,float16,0,0.0124719999730587
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,24,1,128,1,float16,float16,0,0.011779200285673141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,24,2,128,1,float16,float16,0,0.011966399848461151
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,24,4,128,1,float16,float16,0,0.011820799857378005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,24,24,128,1,float16,float16,0,0.011671999841928482
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,24,8,128,1,float16,float16,0,0.012100800126791
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,24,1,128,1,float16,float16,0,0.011683200299739838
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,24,2,128,1,float16,float16,0,0.011947199702262878
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,24,4,128,1,float16,float16,0,0.011767999827861786
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,24,8,128,1,float16,float16,0,0.012015999853610992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,24,1,128,1,float16,float16,0,0.12954879999160768
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,24,2,128,1,float16,float16,0,0.16706399917602538
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,24,4,128,1,float16,float16,0,0.1883247971534729
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,24,4,128,1,float16,float16,0,0.192630398273468
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,24,1,128,1,float16,float16,0,0.029497599601745604
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,24,1,128,1,float16,float16,0,0.049262401461601255
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,24,2,128,1,float16,float16,0,0.050134402513504026
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,24,8,128,1,float16,float16,0,0.24532480239868165
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,24,4,128,1,float16,float16,0,0.06188480257987976
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,24,24,128,1,float16,float16,0,0.2640543937683105
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,24,8,128,1,float16,float16,0,0.06518560051918029
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,24,1,128,1,float16,float16,0,0.028483200073242187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,24,24,128,1,float16,float16,0,0.08176640272140503
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,24,4,128,1,float16,float16,0,0.03494080007076263
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,24,4,128,1,float16,float16,0,0.030184000730514526
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,24,2,128,1,float16,float16,0,0.02913439869880676
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,24,8,128,1,float16,float16,0,0.03626079857349396
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,24,24,128,1,float16,float16,0,0.04642080068588257
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,24,1,128,1,float16,float16,0,0.020151999592781068
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,24,8,128,1,float16,float16,0,0.02113119959831238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,24,4,128,1,float16,float16,0,0.020788800716400147
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,24,1,128,1,float16,float16,0,0.012705600261688233
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,24,2,128,1,float16,float16,0,0.020745599269866945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,24,2,128,1,float16,float16,0,0.012827199697494508
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,24,4,128,1,float16,float16,0,0.013132800161838532
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,24,8,128,1,float16,float16,0,0.013398399949073792
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,24,1,128,1,float16,float16,0,0.011838400363922119
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,24,24,128,1,float16,float16,0,0.018807999789714813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,24,4,128,1,float16,float16,0,0.012044800072908401
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,24,24,128,1,float16,float16,0,0.012171199917793274
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,24,1,128,1,float16,float16,0,0.011671999841928482
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,24,4,128,1,float16,float16,0,0.011798399686813354
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,24,8,128,1,float16,float16,0,0.011689600348472596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,24,2,128,1,float16,float16,0,0.012003199756145477
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,24,24,128,1,float16,float16,0,0.011528000235557556
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,24,2,128,1,float16,float16,0,0.011657600104808808
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,24,4,128,1,float16,float16,0,0.011668799817562104
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,24,8,128,1,float16,float16,0,0.011481600254774094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,24,24,128,1,float16,float16,0,0.011452800035476685
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,24,1,128,1,float16,float16,0,0.011633600294589996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,24,2,128,1,float16,float16,0,0.011686400324106217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,24,4,128,1,float16,float16,0,0.011718399822711945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,24,8,128,1,float16,float16,0,0.011433599889278412
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,24,1,128,1,float16,float16,0,0.05019999742507934
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,24,2,128,1,float16,float16,0,0.060996800661087036
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,24,8,128,1,float16,float16,0,0.05606080293655395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,24,24,128,1,float16,float16,0,0.039350399374961854
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,24,4,128,1,float16,float16,0,0.06268960237503052
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,24,8,128,1,float16,float16,0,0.10528800487518311
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,24,1,128,1,float16,float16,0,0.028838399052619933
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,24,24,128,1,float16,float16,0,0.1457856059074402
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,24,4,128,1,float16,float16,0,0.03497920036315918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,24,2,128,1,float16,float16,0,0.03450399935245514
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,24,8,128,1,float16,float16,0,0.058590400218963626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,24,24,128,1,float16,float16,0,0.07573599815368652
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,24,1,128,1,float16,float16,0,0.020503999292850496
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,24,4,128,1,float16,float16,0,0.02099200040102005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,24,2,128,1,float16,float16,0,0.020611199736595153
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,24,8,128,1,float16,float16,0,0.0325408011674881
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,24,24,128,1,float16,float16,0,0.04333440065383911
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,24,1,128,1,float16,float16,0,0.013206399977207184
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,24,2,128,1,float16,float16,0,0.01310880035161972
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,24,24,128,1,float16,float16,0,0.025905600190162657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,24,4,128,1,float16,float16,0,0.013055999577045441
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,24,8,128,1,float16,float16,0,0.019208000600337984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,24,1,128,1,float16,float16,0,0.011726400256156922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,24,2,128,1,float16,float16,0,0.011843200027942657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,24,24,128,1,float16,float16,0,0.024369600415229797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,24,8,128,1,float16,float16,0,0.011899200081825257
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,24,2,128,1,float16,float16,0,0.012091200053691863
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,24,1,128,1,float16,float16,0,0.011721599847078323
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,24,2,128,1,float16,float16,0,0.011791999638080596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,24,4,128,1,float16,float16,0,0.011668799817562104
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,24,8,128,1,float16,float16,0,0.011531200259923935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,24,24,128,1,float16,float16,0,0.011761599779129028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,24,1,128,1,float16,float16,0,0.011767999827861786
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,24,2,128,1,float16,float16,0,0.011740799993276596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,24,1,128,1,float16,float16,0,0.011961600184440613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,24,4,128,1,float16,float16,0,0.01157120019197464
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,24,8,128,1,float16,float16,0,0.011268799751996994
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,24,1,128,1,float16,float16,0,0.01170400008559227
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,24,24,128,1,float16,float16,0,0.011763200163841248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,24,2,128,1,float16,float16,0,0.011710400134325028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,24,4,128,1,float16,float16,0,0.011596799641847611
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,24,24,128,1,float16,float16,0,0.011483199894428253
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,24,8,128,1,float16,float16,0,0.011321599781513213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,24,1,128,1,float16,float16,0,0.011737599968910217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,24,2,128,1,float16,float16,0,0.01173119992017746
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,24,4,128,1,float16,float16,0,0.011662399768829346
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,24,8,128,1,float16,float16,0,0.011219199746847153
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,24,1,128,1,float16,float16,0,0.019713599979877473
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,24,4,128,1,float16,float16,0,0.05548319816589355
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,24,8,128,1,float16,float16,0,0.09625120162963867
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,24,24,128,1,float16,float16,0,0.13623679876327516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,24,2,128,1,float16,float16,0,0.018456000089645385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,24,4,128,1,float16,float16,0,0.030372801423072814
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,24,8,128,1,float16,float16,0,0.05395519733428955
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,24,24,128,1,float16,float16,0,0.07098399996757507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,24,1,128,1,float16,float16,0,0.011736000329256058
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,24,2,128,1,float16,float16,0,0.011950399726629257
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,24,4,128,1,float16,float16,0,0.0178399994969368
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,24,8,128,1,float16,float16,0,0.02964160144329071
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,24,24,128,1,float16,float16,0,0.04105759859085083
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,24,1,128,1,float16,float16,0,0.01130400002002716
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,24,2,128,1,float16,float16,0,0.011353600025177001
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,24,4,128,1,float16,float16,0,0.011601600050926208
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,24,8,128,1,float16,float16,0,0.017399999499320983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,24,24,128,1,float16,float16,0,0.023315200209617616
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,24,1,128,1,float16,float16,0,0.010939200222492219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,24,2,128,1,float16,float16,0,0.010956799983978272
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,24,4,128,1,float16,float16,0,0.0110384002327919
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,24,8,128,1,float16,float16,0,0.01122720018029213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,24,24,128,1,float16,float16,0,0.01714400053024292
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,24,1,128,1,float16,float16,0,0.01074720025062561
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,24,2,128,1,float16,float16,0,0.010835199803113937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,24,4,128,1,float16,float16,0,0.010903999954462052
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,24,8,128,1,float16,float16,0,0.011025600135326385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,24,24,128,1,float16,float16,0,0.011456000059843064
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,24,1,128,1,float16,float16,0,0.010824000090360641
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,24,2,128,1,float16,float16,0,0.010831999778747558
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,24,4,128,1,float16,float16,0,0.010864000022411346
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,24,4,128,1,float16,float16,0,0.011921600252389909
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,24,24,128,1,float16,float16,0,0.011289600282907486
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,24,2,128,1,float16,float16,0,0.010755199939012527
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,24,1,128,1,float16,float16,0,0.01218400001525879
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,24,4,128,1,float16,float16,0,0.010847999900579452
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,24,24,128,1,float16,float16,0,0.011191999912261963
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,24,1,128,1,float16,float16,0,0.010726399719715118
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,24,2,128,1,float16,float16,0,0.010755199939012527
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,24,4,128,1,float16,float16,0,0.010777600109577179
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,24,8,128,1,float16,float16,0,0.010782399773597717
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,24,2,128,1,float16,float16,0,0.03182719945907593
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,24,1,128,1,float16,float16,0,0.012807999551296235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,24,8,128,1,float16,float16,0,0.010897599905729295
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,24,8,128,1,float16,float16,0,0.011137600243091583
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,24,24,128,1,float16,float16,0,0.01759359985589981
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,24,8,128,1,float16,float16,0,0.012035199999809265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,16,1,128,1,float16,float16,0,20.139863586425783
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,16,2,128,1,float16,float16,0,20.62281494140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,16,4,128,1,float16,float16,0,20.610025024414064
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,16,1,128,1,float16,float16,0,10.098529815673828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,16,1,128,1,float16,float16,0,42.047750854492186
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,16,2,128,1,float16,float16,0,42.18571166992187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,16,4,128,1,float16,float16,0,42.76128234863281
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,16,8,128,1,float16,float16,0,44.53385620117187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,16,8,128,1,float16,float16,0,22.193910217285158
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,16,2,128,1,float16,float16,0,10.65226058959961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,16,1,128,1,float16,float16,0,5.022574234008789
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,16,4,128,1,float16,float16,0,11.158814239501954
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,16,2,128,1,float16,float16,0,5.1662750244140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,16,8,128,1,float16,float16,0,11.432387542724609
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,16,4,128,1,float16,float16,0,5.243865585327148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,16,8,128,1,float16,float16,0,5.319772720336914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,16,1,128,1,float16,float16,0,24.16303405761719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,16,2,128,1,float16,float16,0,23.823841857910157
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,16,4,128,1,float16,float16,0,24.803135681152344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,16,8,128,1,float16,float16,0,24.740846252441408
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,16,1,128,1,float16,float16,0,12.478691101074219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,16,2,128,1,float16,float16,0,12.486949157714843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,16,4,128,1,float16,float16,0,12.874111938476563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,16,8,128,1,float16,float16,0,12.703412628173828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,16,1,128,1,float16,float16,0,5.7549999237060545
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,16,2,128,1,float16,float16,0,5.830244827270508
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,16,4,128,1,float16,float16,0,5.805420684814453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,16,8,128,1,float16,float16,0,6.142307281494141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,16,1,128,1,float16,float16,0,2.8307504653930664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,16,2,128,1,float16,float16,0,2.831046485900879
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,16,4,128,1,float16,float16,0,2.8552608489990234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,16,8,128,1,float16,float16,0,2.9833423614501955
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,16,1,128,1,float16,float16,0,16.695332336425782
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,16,2,128,1,float16,float16,0,16.481657409667967
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,16,4,128,1,float16,float16,0,17.650180053710937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,16,8,128,1,float16,float16,0,17.696246337890624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,16,1,128,1,float16,float16,0,8.574208068847657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,16,2,128,1,float16,float16,0,8.26109619140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,16,4,128,1,float16,float16,0,8.33475341796875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,16,8,128,1,float16,float16,0,8.680129241943359
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,16,16,128,1,float16,float16,0,4.232780838012696
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,16,16,128,1,float16,float16,0,3.001156806945801
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,16,16,128,1,float16,float16,0,6.135292816162109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,16,1,128,1,float16,float16,0,3.7833423614501953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,16,2,128,1,float16,float16,0,4.060307312011719
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,16,16,128,1,float16,float16,0,5.209750366210938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,16,4,128,1,float16,float16,0,4.142340850830078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,16,16,128,1,float16,float16,0,2.3060272216796873
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,16,1,128,1,float16,float16,0,1.8508384704589844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,16,16,128,1,float16,float16,0,12.818229675292969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,16,8,128,1,float16,float16,0,4.219697570800781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,16,2,128,1,float16,float16,0,2.034187126159668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,16,4,128,1,float16,float16,0,2.0304336547851562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,16,8,128,1,float16,float16,0,2.0520095825195312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,16,16,128,1,float16,float16,0,11.010342407226563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,16,16,128,1,float16,float16,0,8.73068618774414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,16,16,128,1,float16,float16,0,22.363143920898438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,16,1,128,1,float16,float16,0,10.405489349365235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,16,16,128,1,float16,float16,0,11.932032012939453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,16,2,128,1,float16,float16,0,10.487075042724609
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,16,4,128,1,float16,float16,0,11.163420867919921
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,16,1,128,1,float16,float16,0,4.93818244934082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,16,16,128,1,float16,float16,0,5.870123291015625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,16,1,128,1,float16,float16,0,21.368177795410155
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,16,2,128,1,float16,float16,0,21.483981323242187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,16,4,128,1,float16,float16,0,22.769656372070312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,16,8,128,1,float16,float16,0,11.402841949462891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,16,8,128,1,float16,float16,0,23.302275085449217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,16,1,128,1,float16,float16,0,2.5505231857299804
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,16,16,128,1,float16,float16,0,3.170742416381836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,16,4,128,1,float16,float16,0,5.359673690795899
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,16,2,128,1,float16,float16,0,5.570683288574219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,16,8,128,1,float16,float16,0,5.4701793670654295
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,16,1,128,1,float16,float16,0,1.4186160087585449
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,16,16,128,1,float16,float16,0,1.5535615921020507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,16,2,128,1,float16,float16,0,1.3179120063781737
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,16,2,128,1,float16,float16,0,2.704628753662109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,16,4,128,1,float16,float16,0,1.3976384162902833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,16,4,128,1,float16,float16,0,2.642780876159668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,16,8,128,1,float16,float16,0,1.3607808113098145
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,16,8,128,1,float16,float16,0,2.79333438873291
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,16,1,128,1,float16,float16,0,6.08740463256836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,16,2,128,1,float16,float16,0,6.21749267578125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,16,16,128,1,float16,float16,0,7.341675567626953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,16,4,128,1,float16,float16,0,6.746009826660156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,16,16,128,1,float16,float16,0,3.3433151245117188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,16,1,128,1,float16,float16,0,2.851108741760254
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,16,2,128,1,float16,float16,0,3.077467155456543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,16,1,128,1,float16,float16,0,12.858224487304687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,16,2,128,1,float16,float16,0,12.901153564453125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,16,4,128,1,float16,float16,0,13.200515747070312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,16,16,128,1,float16,float16,0,1.7825887680053711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,16,8,128,1,float16,float16,0,6.65545425415039
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,16,4,128,1,float16,float16,0,2.8936256408691405
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,16,1,128,1,float16,float16,0,1.4827808380126952
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,16,8,128,1,float16,float16,0,3.154284858703613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,16,8,128,1,float16,float16,0,14.335847473144531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,16,2,128,1,float16,float16,0,1.4679295539855957
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,16,8,128,1,float16,float16,0,1.5616607666015625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,16,4,128,1,float16,float16,0,1.4844320297241211
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,16,16,128,1,float16,float16,0,0.9579327583312989
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,16,1,128,1,float16,float16,0,0.8367504119873047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,16,4,128,1,float16,float16,0,0.7925583839416503
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,16,8,128,1,float16,float16,0,0.8469072341918945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,16,2,128,1,float16,float16,0,0.7669600009918213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,16,1,128,1,float16,float16,0,5.507364654541016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,16,2,128,1,float16,float16,0,5.667571258544922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,16,16,128,1,float16,float16,0,6.8210700988769535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,16,4,128,1,float16,float16,0,5.563806533813477
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,16,1,128,1,float16,float16,0,2.4605968475341795
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,16,16,128,1,float16,float16,0,3.423863983154297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,16,1,128,1,float16,float16,0,11.253529357910157
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,16,2,128,1,float16,float16,0,11.120369720458985
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,16,4,128,1,float16,float16,0,11.955209350585937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,16,8,128,1,float16,float16,0,6.192006301879883
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,16,2,128,1,float16,float16,0,2.769286346435547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,16,8,128,1,float16,float16,0,12.647621154785156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,16,16,128,1,float16,float16,0,1.6862943649291993
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,16,1,128,1,float16,float16,0,1.304537582397461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,16,4,128,1,float16,float16,0,2.7677536010742188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,16,1,128,1,float16,float16,0,0.6979504108428956
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,16,8,128,1,float16,float16,0,2.9435327529907225
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,16,2,128,1,float16,float16,0,1.2699711799621582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,16,2,128,1,float16,float16,0,0.6992959976196289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,16,16,128,1,float16,float16,0,0.8708224296569824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,16,4,128,1,float16,float16,0,0.788267183303833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,16,4,128,1,float16,float16,0,1.349454402923584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,16,2,128,1,float16,float16,0,0.37915360927581787
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,16,8,128,1,float16,float16,0,1.5293055534362794
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,16,8,128,1,float16,float16,0,0.7478352069854737
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,16,16,128,1,float16,float16,0,0.49480319023132324
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,16,4,128,1,float16,float16,0,0.37644000053405763
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,16,8,128,1,float16,float16,0,0.4239488124847412
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,16,1,128,1,float16,float16,0,2.9005407333374023
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,16,1,128,1,float16,float16,0,0.3882783889770508
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,16,16,128,1,float16,float16,0,4.160134506225586
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,16,1,128,1,float16,float16,0,6.737214660644531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,16,2,128,1,float16,float16,0,6.861705780029297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,16,4,128,1,float16,float16,0,3.1820255279541017
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,16,2,128,1,float16,float16,0,3.1512832641601562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,16,4,128,1,float16,float16,0,6.975888061523437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,16,8,128,1,float16,float16,0,7.689785766601562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,16,8,128,1,float16,float16,0,3.602089691162109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,16,1,128,1,float16,float16,0,1.527524757385254
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,16,4,128,1,float16,float16,0,1.5893391609191894
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,16,2,128,1,float16,float16,0,1.6204288482666016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,16,16,128,1,float16,float16,0,2.0826656341552736
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,16,16,128,1,float16,float16,0,1.0728256225585937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,16,8,128,1,float16,float16,0,1.7781248092651367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,16,2,128,1,float16,float16,0,0.8636063575744629
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,16,1,128,1,float16,float16,0,0.8113264083862305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,16,4,128,1,float16,float16,0,0.8268655776977539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,16,1,128,1,float16,float16,0,0.429315185546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,16,16,128,1,float16,float16,0,0.549886417388916
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,16,8,128,1,float16,float16,0,0.8755439758300781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,16,2,128,1,float16,float16,0,0.4098832130432129
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,16,16,128,1,float16,float16,0,0.26933119297027586
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,16,2,128,1,float16,float16,0,0.2588479995727539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,16,8,128,1,float16,float16,0,0.40968642234802244
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,16,4,128,1,float16,float16,0,0.2436511993408203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,16,1,128,1,float16,float16,0,0.24395520687103273
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,16,8,128,1,float16,float16,0,0.24931199550628663
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,16,2,128,1,float16,float16,0,2.596552085876465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,16,1,128,1,float16,float16,0,2.6629247665405273
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,16,4,128,1,float16,float16,0,0.44078240394592283
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,16,16,128,1,float16,float16,0,4.1758544921875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,16,1,128,1,float16,float16,0,6.086281585693359
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,16,16,128,1,float16,float16,0,2.101795196533203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,16,2,128,1,float16,float16,0,6.383721542358399
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,16,1,128,1,float16,float16,0,1.2552783966064454
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,16,4,128,1,float16,float16,0,6.6313232421875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,16,4,128,1,float16,float16,0,3.307428741455078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,16,8,128,1,float16,float16,0,3.549662399291992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,16,8,128,1,float16,float16,0,7.109616088867187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,16,2,128,1,float16,float16,0,1.524062442779541
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,16,16,128,1,float16,float16,0,1.0734640121459962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,16,4,128,1,float16,float16,0,1.3927264213562012
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,16,2,128,1,float16,float16,0,0.7394544124603272
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,16,1,128,1,float16,float16,0,0.754640007019043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,16,4,128,1,float16,float16,0,0.7495408058166504
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,16,8,128,1,float16,float16,0,1.9193504333496094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,16,16,128,1,float16,float16,0,0.5713071823120117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,16,2,128,1,float16,float16,0,0.34311680793762206
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,16,4,128,1,float16,float16,0,0.39142398834228515
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,16,8,128,1,float16,float16,0,0.4487055778503418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,16,8,128,1,float16,float16,0,0.8746720314025879
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,16,16,128,1,float16,float16,0,0.2353584051132202
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,16,4,128,1,float16,float16,0,0.24471840858459473
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,16,16,128,1,float16,float16,0,0.15562399625778198
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,16,8,128,1,float16,float16,0,0.23201119899749756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,16,2,128,1,float16,float16,0,0.22406721115112305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,16,2,128,1,float16,float16,0,0.13893920183181763
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,16,8,128,1,float16,float16,0,0.14340959787368773
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,16,1,128,1,float16,float16,0,0.12914079427719116
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,16,1,128,1,float16,float16,0,0.35796959400177003
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,16,16,128,1,float16,float16,0,2.732961654663086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,16,1,128,1,float16,float16,0,3.408214569091797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,16,2,128,1,float16,float16,0,3.3659183502197267
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,16,1,128,1,float16,float16,0,0.2261359930038452
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,16,4,128,1,float16,float16,0,3.5721057891845702
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,16,8,128,1,float16,float16,0,4.215332794189453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,16,1,128,1,float16,float16,0,1.4785519599914552
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,16,2,128,1,float16,float16,0,1.5637311935424805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,16,1,128,1,float16,float16,0,0.9396623611450196
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,16,16,128,1,float16,float16,0,1.4087408065795899
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,16,4,128,1,float16,float16,0,1.9057807922363281
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,16,4,128,1,float16,float16,0,0.9031696319580078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,16,2,128,1,float16,float16,0,0.852137565612793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,16,16,128,1,float16,float16,0,0.7023744106292724
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,16,1,128,1,float16,float16,0,0.4187935829162598
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,16,2,128,1,float16,float16,0,0.4602015972137451
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,16,8,128,1,float16,float16,0,1.094153594970703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,16,1,128,1,float16,float16,0,0.21452319622039795
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,16,16,128,1,float16,float16,0,0.3524127960205078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,16,8,128,1,float16,float16,0,0.5237855911254883
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,16,8,128,1,float16,float16,0,0.24588320255279542
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,16,2,128,1,float16,float16,0,0.24027519226074218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,16,4,128,1,float16,float16,0,0.24176480770111083
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,16,16,128,1,float16,float16,0,0.14809440374374389
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,16,2,128,1,float16,float16,0,0.128329598903656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,16,4,128,1,float16,float16,0,0.1373855948448181
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,16,4,128,1,float16,float16,0,0.14355520009994507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,16,8,128,1,float16,float16,0,0.15335359573364257
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,16,16,128,1,float16,float16,0,0.0981231987476349
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,16,1,128,1,float16,float16,0,0.0845359981060028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,16,2,128,1,float16,float16,0,0.08367519974708557
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,16,4,128,1,float16,float16,0,0.09310399889945983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,16,4,128,1,float16,float16,0,0.47336640357971194
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,16,8,128,1,float16,float16,0,2.2897455215454103
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,16,1,128,1,float16,float16,0,0.13602399826049805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,16,1,128,1,float16,float16,0,3.198711967468262
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,16,2,128,1,float16,float16,0,3.361318588256836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,16,8,128,1,float16,float16,0,0.09035040140151977
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,16,4,128,1,float16,float16,0,3.491948699951172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,16,1,128,1,float16,float16,0,1.4578240394592286
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,16,8,128,1,float16,float16,0,4.456105422973633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,16,2,128,1,float16,float16,0,1.612388801574707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,16,1,128,1,float16,float16,0,0.83712797164917
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,16,4,128,1,float16,float16,0,1.864156723022461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,16,16,128,1,float16,float16,0,1.670382308959961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,16,4,128,1,float16,float16,0,0.9195055961608887
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,16,16,128,1,float16,float16,0,3.2733665466308595
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,16,8,128,1,float16,float16,0,2.1941423416137695
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,16,1,128,1,float16,float16,0,0.40510239601135256
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,16,16,128,1,float16,float16,0,0.8453200340270997
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,16,2,128,1,float16,float16,0,0.4269392013549805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,16,8,128,1,float16,float16,0,1.1727487564086914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,16,4,128,1,float16,float16,0,0.4653600215911865
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,16,16,128,1,float16,float16,0,0.395249605178833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,16,1,128,1,float16,float16,0,0.2119215965270996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,16,4,128,1,float16,float16,0,0.23552000522613525
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,16,16,128,1,float16,float16,0,0.1534767985343933
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,16,1,128,1,float16,float16,0,0.1246783971786499
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,16,8,128,1,float16,float16,0,0.2818351984024048
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,16,2,128,1,float16,float16,0,0.11840159893035888
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,16,8,128,1,float16,float16,0,0.12572480440139772
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,16,16,128,1,float16,float16,0,0.08528800010681152
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,16,1,128,1,float16,float16,0,0.0684112012386322
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,16,2,128,1,float16,float16,0,0.07435680031776429
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,16,4,128,1,float16,float16,0,0.07901920080184936
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,16,8,128,1,float16,float16,0,0.08161759972572327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,16,16,128,1,float16,float16,0,0.04628799855709076
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,16,1,128,1,float16,float16,0,0.04073919951915741
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,16,2,128,1,float16,float16,0,0.04357439875602722
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,16,4,128,1,float16,float16,0,0.04361119866371155
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,16,8,128,1,float16,float16,0,0.0420879989862442
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,16,2,128,1,float16,float16,0,0.7900335788726807
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,16,2,128,1,float16,float16,0,0.2032383918762207
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,16,8,128,1,float16,float16,0,0.555779218673706
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,16,1,128,1,float16,float16,0,2.303696060180664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,16,4,128,1,float16,float16,0,0.12139040231704712
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,16,2,128,1,float16,float16,0,2.5289087295532227
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,16,4,128,1,float16,float16,0,3.1140928268432617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,16,1,128,1,float16,float16,0,1.1122703552246094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,16,2,128,1,float16,float16,0,1.2821887969970702
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,16,4,128,1,float16,float16,0,1.5384559631347656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,16,1,128,1,float16,float16,0,0.5731967926025391
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,16,2,128,1,float16,float16,0,0.6552656173706055
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,16,16,128,1,float16,float16,0,1.8314847946166992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,16,16,128,1,float16,float16,0,3.3924606323242186
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,16,8,128,1,float16,float16,0,4.338825607299805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,16,8,128,1,float16,float16,0,2.316419219970703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,16,4,128,1,float16,float16,0,0.7603519916534424
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,16,1,128,1,float16,float16,0,0.3113231897354126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,16,8,128,1,float16,float16,0,1.1711584091186524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,16,16,128,1,float16,float16,0,0.4627391815185547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,16,2,128,1,float16,float16,0,0.32583041191101075
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,16,1,128,1,float16,float16,0,0.15298080444335938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,16,4,128,1,float16,float16,0,0.389684796333313
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,16,2,128,1,float16,float16,0,0.1556815981864929
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,16,8,128,1,float16,float16,0,0.6037199974060059
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,16,4,128,1,float16,float16,0,0.18904160261154174
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,16,1,128,1,float16,float16,0,0.08482080101966857
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,16,2,128,1,float16,float16,0,0.08444640040397644
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,16,16,128,1,float16,float16,0,0.16128159761428834
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,16,4,128,1,float16,float16,0,0.0869871973991394
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,16,1,128,1,float16,float16,0,0.05053439736366272
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,16,8,128,1,float16,float16,0,0.09342399835586548
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,16,4,128,1,float16,float16,0,0.05156159996986389
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,16,2,128,1,float16,float16,0,0.05086560249328613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,16,16,128,1,float16,float16,0,0.03838880062103271
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,16,8,128,1,float16,float16,0,0.05449119806289673
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,16,1,128,1,float16,float16,0,0.03475840091705322
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,16,4,128,1,float16,float16,0,0.035262399911880495
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,16,8,128,1,float16,float16,0,0.0356799989938736
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,16,16,128,1,float16,float16,0,0.03316160142421722
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,16,2,128,1,float16,float16,0,0.031350401043891904
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,16,4,128,1,float16,float16,0,0.03137120008468628
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,16,8,128,1,float16,float16,0,0.0323199987411499
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,16,1,128,1,float16,float16,0,0.9840831756591797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,16,2,128,1,float16,float16,0,1.1708944320678711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,16,16,128,1,float16,float16,0,0.8967967987060547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,16,4,128,1,float16,float16,0,1.4719167709350587
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,16,8,128,1,float16,float16,0,0.24926400184631348
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,16,1,128,1,float16,float16,0,0.47883038520812987
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,16,2,128,1,float16,float16,0,0.5531023979187012
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,16,16,128,1,float16,float16,0,0.062028801441192626
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,16,2,128,1,float16,float16,0,0.03474400043487549
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,16,1,128,1,float16,float16,0,0.22528159618377686
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,16,16,128,1,float16,float16,0,1.63885440826416
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,16,1,128,1,float16,float16,0,0.031409600377082826
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,16,8,128,1,float16,float16,0,2.075142478942871
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,16,4,128,1,float16,float16,0,0.7231904029846191
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,16,2,128,1,float16,float16,0,0.25599520206451415
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,16,16,128,1,float16,float16,0,0.8272831916809082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,16,4,128,1,float16,float16,0,0.3288336038589478
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,16,8,128,1,float16,float16,0,1.0506832122802734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,16,4,128,1,float16,float16,0,0.10509120225906372
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,16,2,128,1,float16,float16,0,0.0836736023426056
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,16,16,128,1,float16,float16,0,0.3566015958786011
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,16,16,128,1,float16,float16,0,0.08238080143928528
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,16,1,128,1,float16,float16,0,0.08136320114135742
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,16,2,128,1,float16,float16,0,0.04726240038871765
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,16,8,128,1,float16,float16,0,0.17144479751586914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,16,4,128,1,float16,float16,0,0.048695999383926394
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,16,8,128,1,float16,float16,0,0.051943999528884885
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,16,16,128,1,float16,float16,0,0.034852799773216245
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,16,1,128,1,float16,float16,0,0.028571200370788575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,16,8,128,1,float16,float16,0,0.5122079849243164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,16,1,128,1,float16,float16,0,0.04617120027542114
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,16,8,128,1,float16,float16,0,0.03152799904346466
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,16,2,128,1,float16,float16,0,0.02900480031967163
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,16,16,128,1,float16,float16,0,0.020475199818611144
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,16,1,128,1,float16,float16,0,0.017308799922466277
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,16,4,128,1,float16,float16,0,0.01794400066137314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,16,8,128,1,float16,float16,0,0.018779200315475465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,16,2,128,1,float16,float16,0,0.017374399304389953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,16,16,128,1,float16,float16,0,0.017828799784183502
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,16,2,128,1,float16,float16,0,0.016220800578594208
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,16,1,128,1,float16,float16,0,0.0160863995552063
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,16,4,128,1,float16,float16,0,0.0164015993475914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,16,16,128,1,float16,float16,0,0.016672000288963318
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,16,4,128,1,float16,float16,0,0.015988799929618835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,16,2,128,1,float16,float16,0,0.015889599919319153
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,16,8,128,1,float16,float16,0,0.016147199273109435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,16,1,128,1,float16,float16,0,0.48109917640686034
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,16,2,128,1,float16,float16,0,0.5490655899047852
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,16,4,128,1,float16,float16,0,0.7423520088195801
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,16,1,128,1,float16,float16,0,0.22425758838653564
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,16,8,128,1,float16,float16,0,1.0352383613586427
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,16,2,128,1,float16,float16,0,0.25603361129760743
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,16,16,128,1,float16,float16,0,0.8302767753601075
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,16,4,128,1,float16,float16,0,0.33079841136932375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,16,2,128,1,float16,float16,0,0.06622560024261474
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,16,16,128,1,float16,float16,0,0.3421008110046387
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,16,8,128,1,float16,float16,0,0.5077568054199219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,16,1,128,1,float16,float16,0,0.06466879844665527
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,16,4,128,1,float16,float16,0,0.08527039885520935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,16,8,128,1,float16,float16,0,0.16909279823303222
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,16,16,128,1,float16,float16,0,0.061607998609542844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,16,2,128,1,float16,float16,0,0.0381520003080368
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,16,4,128,1,float16,float16,0,0.039787200093269345
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,16,8,128,1,float16,float16,0,0.042987200617790225
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,16,16,128,1,float16,float16,0,0.02853280007839203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,16,1,128,1,float16,float16,0,0.02227199971675873
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,16,2,128,1,float16,float16,0,0.022752000391483305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,16,4,128,1,float16,float16,0,0.02345760017633438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,16,8,128,1,float16,float16,0,0.025060799717903138
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,16,16,128,1,float16,float16,0,0.016755199432373045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,16,1,128,1,float16,float16,0,0.013704000413417817
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,16,2,128,1,float16,float16,0,0.01377280056476593
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,16,4,128,1,float16,float16,0,0.0143312007188797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,16,8,128,1,float16,float16,0,0.015048000216484069
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,16,16,128,1,float16,float16,0,0.013793599605560303
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,16,1,128,1,float16,float16,0,0.012356799840927125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,16,2,128,1,float16,float16,0,0.012478400021791458
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,16,4,128,1,float16,float16,0,0.012567999958992004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,16,8,128,1,float16,float16,0,0.016891199350357055
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,16,8,128,1,float16,float16,0,0.013072000443935394
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,16,16,128,1,float16,float16,0,0.012800000607967377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,16,1,128,1,float16,float16,0,0.015756799280643462
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,16,1,128,1,float16,float16,0,0.012043199688196182
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,16,4,128,1,float16,float16,0,0.012129600346088409
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,16,16,128,1,float16,float16,0,0.012249600142240524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,16,8,128,1,float16,float16,0,0.01231039986014366
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,16,2,128,1,float16,float16,0,0.012212800234556198
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,16,4,128,1,float16,float16,0,0.012015999853610992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,16,2,128,1,float16,float16,0,0.011907199770212174
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,16,8,128,1,float16,float16,0,0.012068799883127212
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,16,1,128,1,float16,float16,0,0.2254784107208252
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,16,2,128,1,float16,float16,0,0.253383994102478
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,16,4,128,1,float16,float16,0,0.32875039577484133
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,16,1,128,1,float16,float16,0,0.06337760090827942
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,16,8,128,1,float16,float16,0,0.5032559871673584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,16,2,128,1,float16,float16,0,0.06439999938011169
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,16,4,128,1,float16,float16,0,0.07366080284118652
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,16,16,128,1,float16,float16,0,0.3509167909622192
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,16,16,128,1,float16,float16,0,0.09478719830513001
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,16,8,128,1,float16,float16,0,0.17210079431533815
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,16,1,128,1,float16,float16,0,0.0359360009431839
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,16,4,128,1,float16,float16,0,0.038254401087760924
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,16,8,128,1,float16,float16,0,0.0412416011095047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,16,16,128,1,float16,float16,0,0.03819519877433777
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,16,2,128,1,float16,float16,0,0.036795198917388916
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,16,1,128,1,float16,float16,0,0.021425600349903106
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,16,2,128,1,float16,float16,0,0.02168319970369339
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,16,4,128,1,float16,float16,0,0.022668799757957457
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,16,16,128,1,float16,float16,0,0.021804800629615782
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,16,1,128,1,float16,float16,0,0.01345279961824417
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,16,2,128,1,float16,float16,0,0.013471999764442444
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,16,1,128,1,float16,float16,0,0.03751679956912994
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,16,8,128,1,float16,float16,0,0.01478080004453659
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,16,16,128,1,float16,float16,0,0.013665600121021271
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,16,1,128,1,float16,float16,0,0.012200000137090683
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,16,2,128,1,float16,float16,0,0.012225600332021714
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,16,4,128,1,float16,float16,0,0.012385600060224534
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,16,8,128,1,float16,float16,0,0.012912000715732574
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,16,16,128,1,float16,float16,0,0.012303999811410903
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,16,1,128,1,float16,float16,0,0.011795199662446975
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,16,4,128,1,float16,float16,0,0.011912000179290772
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,16,2,128,1,float16,float16,0,0.011897599697113037
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,16,8,128,1,float16,float16,0,0.012107200175523757
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,16,16,128,1,float16,float16,0,0.011671999841928482
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,16,4,128,1,float16,float16,0,0.029838401079177856
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,16,2,128,1,float16,float16,0,0.011774399876594543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,16,4,128,1,float16,float16,0,0.011992000043392181
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,16,8,128,1,float16,float16,0,0.011881600320339202
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,16,1,128,1,float16,float16,0,0.011692799627780914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,16,16,128,1,float16,float16,0,0.011686400324106217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,16,2,128,1,float16,float16,0,0.01175839975476265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,16,4,128,1,float16,float16,0,0.011847999691963196
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,16,1,128,1,float16,float16,0,0.011897599697113037
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,16,1,128,1,float16,float16,0,0.06401439905166625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,16,2,128,1,float16,float16,0,0.06523680090904235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,16,4,128,1,float16,float16,0,0.07869920134544373
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,16,8,128,1,float16,float16,0,0.19132479429244995
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,16,16,128,1,float16,float16,0,0.12775520086288453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,16,1,128,1,float16,float16,0,0.035867199301719666
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,16,2,128,1,float16,float16,0,0.03694559931755066
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,16,4,128,1,float16,float16,0,0.038624000549316403
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,16,8,128,1,float16,float16,0,0.06253439784049988
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,16,16,128,1,float16,float16,0,0.06029599905014038
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,16,1,128,1,float16,float16,0,0.021550400555133818
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,16,2,128,1,float16,float16,0,0.02179040014743805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,16,4,128,1,float16,float16,0,0.022648000717163087
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,16,16,128,1,float16,float16,0,0.033036801218986514
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,16,8,128,1,float16,float16,0,0.0349151998758316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,16,1,128,1,float16,float16,0,0.013619199395179749
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,16,2,128,1,float16,float16,0,0.013675199449062347
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,16,8,128,1,float16,float16,0,0.020185600221157073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,16,4,128,1,float16,float16,0,0.014435200393199921
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,16,16,128,1,float16,float16,0,0.018932799994945525
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,16,1,128,1,float16,float16,0,0.012095999717712403
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,16,8,128,1,float16,float16,0,0.02427999973297119
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,16,2,128,1,float16,float16,0,0.01234399974346161
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,16,4,128,1,float16,float16,0,0.012275200337171555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,16,8,128,1,float16,float16,0,0.01268800050020218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,16,4,128,1,float16,float16,0,0.01398719996213913
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,16,16,128,1,float16,float16,0,0.012521600723266602
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,16,4,128,1,float16,float16,0,0.01186399981379509
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,16,8,128,1,float16,float16,0,0.011727999895811081
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,16,1,128,1,float16,float16,0,0.01165440008044243
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,16,16,128,1,float16,float16,0,0.011721599847078323
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,16,2,128,1,float16,float16,0,0.011686400324106217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,16,4,128,1,float16,float16,0,0.011707200109958649
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,16,16,128,1,float16,float16,0,0.011260800063610077
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,16,8,128,1,float16,float16,0,0.011505600064992905
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,16,1,128,1,float16,float16,0,0.011622399836778641
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,16,2,128,1,float16,float16,0,0.011630400270223617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,16,1,128,1,float16,float16,0,0.01170239970088005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,16,8,128,1,float16,float16,0,0.011236800253391266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,16,4,128,1,float16,float16,0,0.01188800036907196
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,16,16,128,1,float16,float16,0,0.011208000034093857
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,16,2,128,1,float16,float16,0,0.01165279969573021
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,16,1,128,1,float16,float16,0,0.0119439996778965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,16,8,128,1,float16,float16,0,0.01120320037007332
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,16,4,128,1,float16,float16,0,0.011843200027942657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,16,1,128,1,float16,float16,0,0.03673279881477356
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,16,4,128,1,float16,float16,0,0.060412800312042235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,16,8,128,1,float16,float16,0,0.011724799871444702
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,16,2,128,1,float16,float16,0,0.038001599907875064
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,16,8,128,1,float16,float16,0,0.10356160402297973
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,16,1,128,1,float16,float16,0,0.021827200055122377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,16,16,128,1,float16,float16,0,0.10155839920043945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,16,2,128,1,float16,float16,0,0.02229599952697754
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,16,4,128,1,float16,float16,0,0.03389120101928711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,16,1,128,1,float16,float16,0,0.013836799561977387
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,16,16,128,1,float16,float16,0,0.05660799741744995
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,16,8,128,1,float16,float16,0,0.031585600972175595
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,16,16,128,1,float16,float16,0,0.031001600623130798
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,16,1,128,1,float16,float16,0,0.0123648002743721
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,16,2,128,1,float16,float16,0,0.012404800206422806
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,16,4,128,1,float16,float16,0,0.01249919980764389
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,16,8,128,1,float16,float16,0,0.01852000057697296
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,16,16,128,1,float16,float16,0,0.018038399517536163
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,16,1,128,1,float16,float16,0,0.011737599968910217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,16,2,128,1,float16,float16,0,0.011691199988126755
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,16,4,128,1,float16,float16,0,0.011519999802112579
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,16,8,128,1,float16,float16,0,0.01165279969573021
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,16,16,128,1,float16,float16,0,0.011742399632930755
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,16,1,128,1,float16,float16,0,0.011585599929094314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,16,2,128,1,float16,float16,0,0.011615999788045884
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,16,4,128,1,float16,float16,0,0.0112527996301651
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,16,8,128,1,float16,float16,0,0.011259199678897857
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,16,16,128,1,float16,float16,0,0.011327999830245971
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,16,1,128,1,float16,float16,0,0.011550399661064147
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,16,2,128,1,float16,float16,0,0.011577600240707397
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,16,4,128,1,float16,float16,0,0.011196800321340562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,16,8,128,1,float16,float16,0,0.011070399731397628
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,16,16,128,1,float16,float16,0,0.011166399717330933
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,16,1,128,1,float16,float16,0,0.011585599929094314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,16,2,128,1,float16,float16,0,0.01159999966621399
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,16,4,128,1,float16,float16,0,0.011217600107192994
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,16,8,128,1,float16,float16,0,0.01096000000834465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,16,16,128,1,float16,float16,0,0.011161600053310395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,16,1,128,1,float16,float16,0,0.011513599753379821
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,16,2,128,1,float16,float16,0,0.011537600308656693
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,16,4,128,1,float16,float16,0,0.011153600364923476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,16,8,128,1,float16,float16,0,0.010929600149393082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,16,1,128,1,float16,float16,0,0.019539199769496918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,16,1,128,1,float16,float16,0,0.011956799775362015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,16,2,128,1,float16,float16,0,0.03136320114135742
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,16,4,128,1,float16,float16,0,0.05516800284385681
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,16,8,128,1,float16,float16,0,0.09685919880867004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,16,16,128,1,float16,float16,0,0.09542080163955688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,16,1,128,1,float16,float16,0,0.012750400602817536
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,16,4,128,1,float16,float16,0,0.030542400479316712
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,16,8,128,1,float16,float16,0,0.0543503999710083
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,16,16,128,1,float16,float16,0,0.05368800163269043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,16,1,128,1,float16,float16,0,0.011900799721479416
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,16,2,128,1,float16,float16,0,0.012166400253772736
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,16,2,128,1,float16,float16,0,0.014099200069904328
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,16,8,128,1,float16,float16,0,0.0577023983001709
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,16,4,128,1,float16,float16,0,0.01806560009717941
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,16,4,128,1,float16,float16,0,0.020054399967193604
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,16,8,128,1,float16,float16,0,0.030219200253486633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,16,2,128,1,float16,float16,0,0.011398400366306304
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,16,4,128,1,float16,float16,0,0.011903999745845795
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,16,16,128,1,float16,float16,0,0.01728159934282303
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,16,2,128,1,float16,float16,0,0.010918399691581726
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,16,1,128,1,float16,float16,0,0.01119839996099472
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,16,4,128,1,float16,float16,0,0.011028800159692764
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,16,16,128,1,float16,float16,0,0.011353600025177001
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,16,1,128,1,float16,float16,0,0.010831999778747558
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,16,2,128,1,float16,float16,0,0.010878399759531022
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,16,4,128,1,float16,float16,0,0.010873600095510482
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,16,8,128,1,float16,float16,0,0.01101280003786087
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,16,16,128,1,float16,float16,0,0.01120000034570694
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,16,1,128,1,float16,float16,0,0.01072160005569458
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,16,2,128,1,float16,float16,0,0.010755199939012527
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,16,4,128,1,float16,float16,0,0.010828799754381179
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,16,8,128,1,float16,float16,0,0.010876800119876861
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,16,16,128,1,float16,float16,0,0.011025600135326385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,16,1,128,1,float16,float16,0,0.010751999914646149
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,16,2,128,1,float16,float16,0,0.010785599797964096
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,16,4,128,1,float16,float16,0,0.010873600095510482
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,16,8,128,1,float16,float16,0,0.0108255997300148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,16,16,128,1,float16,float16,0,0.011006399989128113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,16,2,128,1,float16,float16,0,0.018486399948596955
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,16,1,128,1,float16,float16,0,0.010651200264692306
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,16,2,128,1,float16,float16,0,0.010844799876213073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,16,4,128,1,float16,float16,0,0.010731200128793717
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,16,8,128,1,float16,float16,0,0.010817600041627884
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,16,16,128,1,float16,float16,0,0.029553601145744325
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,16,8,128,1,float16,float16,0,0.01791200041770935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,16,8,128,1,float16,float16,0,0.011539199948310852
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,16,2,128,1,float16,float16,0,0.011699199676513672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,16,1,128,1,float16,float16,0,0.011486399918794632
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,12,1,128,1,float16,float16,0,15.196893310546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,12,2,128,1,float16,float16,0,15.778709411621094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,12,4,128,1,float16,float16,0,15.636030578613282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,12,1,128,1,float16,float16,0,7.599689483642578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,12,2,128,1,float16,float16,0,7.813855743408203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,12,4,128,1,float16,float16,0,7.723540496826172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,12,1,128,1,float16,float16,0,30.8511474609375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,12,1,128,1,float16,float16,0,3.8137359619140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,12,2,128,1,float16,float16,0,31.5850830078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,12,2,128,1,float16,float16,0,3.7022079467773437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,12,4,128,1,float16,float16,0,31.97327575683594
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,12,4,128,1,float16,float16,0,4.123751831054688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,12,1,128,1,float16,float16,0,9.378593444824219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,12,1,128,1,float16,float16,0,19.006422424316405
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,12,2,128,1,float16,float16,0,19.299458312988282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,12,4,128,1,float16,float16,0,19.392288208007812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,12,2,128,1,float16,float16,0,9.204003143310548
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,12,1,128,1,float16,float16,0,4.100851058959961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,12,2,128,1,float16,float16,0,4.354260635375977
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,12,4,128,1,float16,float16,0,9.397262573242188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,12,4,128,1,float16,float16,0,4.353710556030274
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,12,1,128,1,float16,float16,0,2.0631296157836916
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,12,2,128,1,float16,float16,0,2.114841651916504
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,12,4,128,1,float16,float16,0,2.1762208938598633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,12,1,128,1,float16,float16,0,12.197782135009765
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,12,2,128,1,float16,float16,0,13.120747375488282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,12,4,128,1,float16,float16,0,13.57138671875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,12,1,128,1,float16,float16,0,5.962390518188476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,12,2,128,1,float16,float16,0,6.074078369140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,12,4,128,1,float16,float16,0,6.241268920898437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,12,12,128,1,float16,float16,0,3.120854377746582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,12,12,128,1,float16,float16,0,3.941912078857422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,12,1,128,1,float16,float16,0,2.882923126220703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,12,2,128,1,float16,float16,0,3.026326370239258
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,12,4,128,1,float16,float16,0,3.06268310546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,12,12,128,1,float16,float16,0,2.2649696350097654
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,12,12,128,1,float16,float16,0,1.7935056686401367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,12,12,128,1,float16,float16,0,6.549654388427735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,12,12,128,1,float16,float16,0,9.458483123779297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,12,1,128,1,float16,float16,0,1.4910351753234863
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,12,12,128,1,float16,float16,0,8.024443054199219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,12,12,128,1,float16,float16,0,16.368067932128906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,12,2,128,1,float16,float16,0,1.4822463989257812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,12,12,128,1,float16,float16,0,4.714982223510742
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,12,4,128,1,float16,float16,0,1.4855199813842774
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,12,12,128,1,float16,float16,0,8.916734313964843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,12,2,128,1,float16,float16,0,7.635591888427735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,12,1,128,1,float16,float16,0,7.732569885253906
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,12,12,128,1,float16,float16,0,4.3297119140625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,12,4,128,1,float16,float16,0,8.420212554931641
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,12,1,128,1,float16,float16,0,3.715206527709961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,12,12,128,1,float16,float16,0,2.1920352935791017
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,12,1,128,1,float16,float16,0,16.209255981445313
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,12,2,128,1,float16,float16,0,16.209788513183593
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,12,2,128,1,float16,float16,0,3.8775089263916014
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,12,1,128,1,float16,float16,0,1.9602224349975585
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,12,4,128,1,float16,float16,0,4.060168075561523
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,12,2,128,1,float16,float16,0,1.953291130065918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,12,12,128,1,float16,float16,0,1.2631263732910156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,12,1,128,1,float16,float16,0,1.085208034515381
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,12,4,128,1,float16,float16,0,16.98089599609375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,12,2,128,1,float16,float16,0,1.0129152297973634
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,12,4,128,1,float16,float16,0,1.959596824645996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,12,4,128,1,float16,float16,0,1.02041597366333
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,12,1,128,1,float16,float16,0,4.359980773925781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,12,2,128,1,float16,float16,0,4.589899063110352
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,12,4,128,1,float16,float16,0,4.812631988525391
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,12,12,128,1,float16,float16,0,5.148438262939453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,12,1,128,1,float16,float16,0,2.186479949951172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,12,12,128,1,float16,float16,0,2.5458927154541016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,12,2,128,1,float16,float16,0,2.3168703079223634
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,12,1,128,1,float16,float16,0,9.13280029296875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,12,4,128,1,float16,float16,0,2.3478672027587892
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,12,12,128,1,float16,float16,0,1.3269536018371582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,12,1,128,1,float16,float16,0,1.0561152458190919
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,12,2,128,1,float16,float16,0,0.9742815971374512
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,12,2,128,1,float16,float16,0,9.68896484375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,12,4,128,1,float16,float16,0,9.685079956054688
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,12,4,128,1,float16,float16,0,1.217027187347412
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,12,1,128,1,float16,float16,0,0.6508431911468506
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,12,12,128,1,float16,float16,0,0.7473983764648438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,12,4,128,1,float16,float16,0,0.6029871940612793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,12,2,128,1,float16,float16,0,0.5970960140228272
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,12,12,128,1,float16,float16,0,2.4160911560058596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,12,1,128,1,float16,float16,0,3.935852813720703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,12,2,128,1,float16,float16,0,4.285623931884766
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,12,4,128,1,float16,float16,0,4.396527862548828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,12,1,128,1,float16,float16,0,1.8838272094726562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,12,12,128,1,float16,float16,0,4.925323104858398
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,12,12,128,1,float16,float16,0,1.353865623474121
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,12,2,128,1,float16,float16,0,2.0456960678100584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,12,1,128,1,float16,float16,0,1.0371647834777833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,12,2,128,1,float16,float16,0,0.967801570892334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,12,4,128,1,float16,float16,0,2.171183967590332
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,12,1,128,1,float16,float16,0,0.5342271804809571
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,12,2,128,1,float16,float16,0,0.528331184387207
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,12,12,128,1,float16,float16,0,0.7086959838867187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,12,12,128,1,float16,float16,0,0.3269216060638428
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,12,4,128,1,float16,float16,0,1.0347552299499512
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,12,1,128,1,float16,float16,0,8.099456024169921
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,12,1,128,1,float16,float16,0,0.30362720489501954
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,12,4,128,1,float16,float16,0,0.6297967910766602
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,12,2,128,1,float16,float16,0,0.3246623992919922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,12,4,128,1,float16,float16,0,0.3096719980239868
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,12,2,128,1,float16,float16,0,9.510848236083984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,12,1,128,1,float16,float16,0,1.9541664123535156
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,12,2,128,1,float16,float16,0,2.2022192001342775
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,12,12,128,1,float16,float16,0,2.993707275390625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,12,12,128,1,float16,float16,0,1.6253360748291015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,12,1,128,1,float16,float16,0,1.234540843963623
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,12,4,128,1,float16,float16,0,2.857731246948242
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,12,1,128,1,float16,float16,0,4.948432159423828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,12,2,128,1,float16,float16,0,1.0850864410400392
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,12,4,128,1,float16,float16,0,9.169102478027344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,12,2,128,1,float16,float16,0,4.852947235107422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,12,4,128,1,float16,float16,0,5.153646469116211
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,12,12,128,1,float16,float16,0,0.8443663597106934
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,12,1,128,1,float16,float16,0,0.6384640216827393
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,12,4,128,1,float16,float16,0,1.2088735580444336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,12,4,128,1,float16,float16,0,0.6560383796691894
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,12,1,128,1,float16,float16,0,0.3904639959335327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,12,4,128,1,float16,float16,0,0.3778480052947998
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,12,12,128,1,float16,float16,0,0.39120800495147706
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,12,2,128,1,float16,float16,0,0.6608975887298584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,12,12,128,1,float16,float16,0,0.22699038982391356
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,12,2,128,1,float16,float16,0,0.21231679916381835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,12,2,128,1,float16,float16,0,0.34649760723114015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,12,1,128,1,float16,float16,0,0.20194718837738038
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,12,4,128,1,float16,float16,0,0.22021119594573973
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,12,12,128,1,float16,float16,0,1.5829839706420898
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,12,1,128,1,float16,float16,0,1.95467529296875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,12,2,128,1,float16,float16,0,2.0765472412109376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,12,4,128,1,float16,float16,0,2.251163291931152
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,12,12,128,1,float16,float16,0,3.145707130432129
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,12,1,128,1,float16,float16,0,0.9502832412719726
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,12,2,128,1,float16,float16,0,1.161678409576416
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,12,12,128,1,float16,float16,0,0.8763744354248046
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,12,4,128,1,float16,float16,0,1.2499024391174316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,12,2,128,1,float16,float16,0,0.5282688140869141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,12,1,128,1,float16,float16,0,0.5350656032562255
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,12,1,128,1,float16,float16,0,4.461687850952148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,12,1,128,1,float16,float16,0,0.3108479976654053
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,12,12,128,1,float16,float16,0,0.4225791931152344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,12,4,128,1,float16,float16,0,0.6191999912261963
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,12,2,128,1,float16,float16,0,4.587740707397461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,12,2,128,1,float16,float16,0,0.2988672018051147
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,12,1,128,1,float16,float16,0,0.17017120122909546
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,12,4,128,1,float16,float16,0,0.28540959358215334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,12,2,128,1,float16,float16,0,0.17484480142593384
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,12,4,128,1,float16,float16,0,0.18576799631118773
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,12,1,128,1,float16,float16,0,0.12079839706420899
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,12,2,128,1,float16,float16,0,0.11635199785232545
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,12,12,128,1,float16,float16,0,0.21383678913116455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,12,4,128,1,float16,float16,0,5.010348892211914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,12,12,128,1,float16,float16,0,0.15260000228881837
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,12,1,128,1,float16,float16,0,1.1573023796081543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,12,2,128,1,float16,float16,0,1.232646369934082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,12,12,128,1,float16,float16,0,2.0854303359985353
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,12,1,128,1,float16,float16,0,2.3614511489868164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,12,2,128,1,float16,float16,0,2.5980255126953127
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,12,4,128,1,float16,float16,0,2.80928955078125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,12,12,128,1,float16,float16,0,1.0509471893310547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,12,4,128,1,float16,float16,0,1.3893119812011718
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,12,2,128,1,float16,float16,0,0.7473552227020264
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,12,1,128,1,float16,float16,0,0.590283203125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,12,4,128,1,float16,float16,0,0.11553599834442138
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,12,12,128,1,float16,float16,0,0.2549407958984375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,12,1,128,1,float16,float16,0,0.3118016004562378
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,12,12,128,1,float16,float16,0,0.5400224208831788
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,12,1,128,1,float16,float16,0,0.17002880573272705
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,12,2,128,1,float16,float16,0,0.3593584060668945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,12,4,128,1,float16,float16,0,0.40207839012145996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,12,2,128,1,float16,float16,0,0.1767583966255188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,12,4,128,1,float16,float16,0,0.18224639892578126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,12,1,128,1,float16,float16,0,0.10572960376739501
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,12,2,128,1,float16,float16,0,0.11895840167999268
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,12,12,128,1,float16,float16,0,0.13364800214767455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,12,12,128,1,float16,float16,0,0.06748480200767518
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,12,1,128,1,float16,float16,0,0.06641600131988526
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,12,4,128,1,float16,float16,0,0.12321280241012574
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,12,4,128,1,float16,float16,0,0.07169759869575501
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,12,1,128,1,float16,float16,0,1.2870512008666992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,12,4,128,1,float16,float16,0,0.7588143825531006
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,12,1,128,1,float16,float16,0,2.639801597595215
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,12,12,128,1,float16,float16,0,2.6463296890258787
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,12,2,128,1,float16,float16,0,2.908451271057129
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,12,2,128,1,float16,float16,0,0.06286399960517883
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,12,4,128,1,float16,float16,0,3.3774032592773438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,12,2,128,1,float16,float16,0,1.4252976417541503
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,12,1,128,1,float16,float16,0,0.652185583114624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,12,1,128,1,float16,float16,0,0.35909280776977537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,12,2,128,1,float16,float16,0,0.7383935928344727
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,12,4,128,1,float16,float16,0,0.8258624076843262
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,12,4,128,1,float16,float16,0,1.6368511199951172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,12,12,128,1,float16,float16,0,0.719371223449707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,12,12,128,1,float16,float16,0,1.4167920112609864
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,12,12,128,1,float16,float16,0,0.3547231912612915
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,12,2,128,1,float16,float16,0,0.36793439388275145
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,12,1,128,1,float16,float16,0,0.10562560558319092
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,12,12,128,1,float16,float16,0,0.13804320096969605
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,12,2,128,1,float16,float16,0,0.17417759895324708
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,12,4,128,1,float16,float16,0,0.4143360137939453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,12,4,128,1,float16,float16,0,0.20141758918762206
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,12,2,128,1,float16,float16,0,0.1027616024017334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,12,12,128,1,float16,float16,0,0.08827679753303527
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,12,4,128,1,float16,float16,0,0.10253119468688965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,12,12,128,1,float16,float16,0,0.060524797439575194
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,12,4,128,1,float16,float16,0,0.05780159831047058
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,12,2,128,1,float16,float16,0,0.056884801387786864
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,12,1,128,1,float16,float16,0,0.04150719940662384
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,12,2,128,1,float16,float16,0,0.04866400063037872
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,12,4,128,1,float16,float16,0,0.05132160186767578
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,12,1,128,1,float16,float16,0,0.884819221496582
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,12,2,128,1,float16,float16,0,1.046356773376465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,12,1,128,1,float16,float16,0,0.1799183964729309
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,12,1,128,1,float16,float16,0,1.795102310180664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,12,2,128,1,float16,float16,0,2.0622352600097655
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,12,1,128,1,float16,float16,0,0.05816320180892944
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,12,12,128,1,float16,float16,0,2.6304975509643556
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,12,4,128,1,float16,float16,0,2.7932607650756838
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,12,1,128,1,float16,float16,0,0.4507440090179443
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,12,12,128,1,float16,float16,0,1.383351993560791
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,12,4,128,1,float16,float16,0,1.4151247978210448
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,12,2,128,1,float16,float16,0,0.526639986038208
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,12,12,128,1,float16,float16,0,0.6753568172454834
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,12,4,128,1,float16,float16,0,0.7181935787200928
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,12,1,128,1,float16,float16,0,0.24801759719848632
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,12,4,128,1,float16,float16,0,0.33147521018981935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,12,2,128,1,float16,float16,0,0.26572959423065184
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,12,12,128,1,float16,float16,0,0.09118880033493042
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,12,2,128,1,float16,float16,0,0.12241439819335938
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,12,4,128,1,float16,float16,0,0.12945120334625243
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,12,1,128,1,float16,float16,0,0.11464639902114868
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,12,12,128,1,float16,float16,0,0.29428958892822266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,12,1,128,1,float16,float16,0,0.0662608027458191
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,12,2,128,1,float16,float16,0,0.06703199744224549
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,12,12,128,1,float16,float16,0,0.05222560167312622
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,12,4,128,1,float16,float16,0,0.0749776005744934
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,12,1,128,1,float16,float16,0,0.044417598843574525
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,12,2,128,1,float16,float16,0,0.04302560091018677
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,12,12,128,1,float16,float16,0,0.0457616001367569
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,12,4,128,1,float16,float16,0,0.04460479915142059
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,12,1,128,1,float16,float16,0,0.03887520134449005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,12,2,128,1,float16,float16,0,0.03936479985713959
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,12,2,128,1,float16,float16,0,0.027582401037216188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,12,12,128,1,float16,float16,0,0.029177600145339967
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,12,4,128,1,float16,float16,0,0.04356319904327392
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,12,1,128,1,float16,float16,0,0.027775999903678895
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,12,4,128,1,float16,float16,0,0.027635198831558228
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,12,1,128,1,float16,float16,0,0.40359840393066404
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,12,1,128,1,float16,float16,0,0.8487504005432129
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,12,12,128,1,float16,float16,0,0.6228847980499268
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,12,2,128,1,float16,float16,0,0.9311967849731445
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,12,4,128,1,float16,float16,0,0.6771183967590332
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,12,12,128,1,float16,float16,0,1.243064022064209
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,12,4,128,1,float16,float16,0,1.2556351661682128
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,12,4,128,1,float16,float16,0,0.2630944013595581
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,12,12,128,1,float16,float16,0,0.20733280181884767
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,12,4,128,1,float16,float16,0,0.06997119784355163
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,12,12,128,1,float16,float16,0,0.04613440036773682
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,12,1,128,1,float16,float16,0,0.06378880143165588
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,12,2,128,1,float16,float16,0,0.038395199179649356
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,12,1,128,1,float16,float16,0,0.03680480122566223
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,12,2,128,1,float16,float16,0,0.06624000072479248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,12,4,128,1,float16,float16,0,0.039801600575447085
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,12,12,128,1,float16,float16,0,0.029315200448036195
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,12,1,128,1,float16,float16,0,0.024694399535655977
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,12,2,128,1,float16,float16,0,0.025696000456809996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,12,4,128,1,float16,float16,0,0.02640640139579773
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,12,12,128,1,float16,float16,0,0.019100800156593323
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,12,1,128,1,float16,float16,0,0.01677599996328354
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,12,2,128,1,float16,float16,0,0.01700640022754669
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,12,4,128,1,float16,float16,0,0.017444799840450286
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,12,12,128,1,float16,float16,0,0.017323200404644013
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,12,1,128,1,float16,float16,0,0.016169600188732147
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,12,2,128,1,float16,float16,0,0.016279999911785126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,12,1,128,1,float16,float16,0,0.015775999426841734
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,12,12,128,1,float16,float16,0,0.01650719940662384
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,12,2,128,1,float16,float16,0,0.015958400070667268
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,12,4,128,1,float16,float16,0,0.015996800363063814
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,12,1,128,1,float16,float16,0,0.37292799949645994
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,12,2,128,1,float16,float16,0,0.4488815784454346
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,12,4,128,1,float16,float16,0,0.6258927822113037
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,12,1,128,1,float16,float16,0,0.16015839576721191
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,12,2,128,1,float16,float16,0,0.19062720537185668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,12,12,128,1,float16,float16,0,0.6132959842681884
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,12,4,128,1,float16,float16,0,0.2567519903182983
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,12,12,128,1,float16,float16,0,0.20559680461883545
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,12,2,128,1,float16,float16,0,0.053908801078796385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,12,12,128,1,float16,float16,0,0.03999040126800537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,12,4,128,1,float16,float16,0,0.05682399868965149
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,12,1,128,1,float16,float16,0,0.17700959444046022
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,12,1,128,1,float16,float16,0,0.030527999997138976
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,12,2,128,1,float16,float16,0,0.03146719932556152
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,12,2,128,1,float16,float16,0,0.20733439922332764
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,12,1,128,1,float16,float16,0,0.0211776003241539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,12,12,128,1,float16,float16,0,0.025887998938560485
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,12,4,128,1,float16,float16,0,0.022433599829673766
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,12,2,128,1,float16,float16,0,0.021849599480628968
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,12,12,128,1,float16,float16,0,0.015476800501346588
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,12,1,128,1,float16,float16,0,0.013174399733543396
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,12,2,128,1,float16,float16,0,0.013422399759292603
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,12,4,128,1,float16,float16,0,0.013851200044155122
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,12,12,128,1,float16,float16,0,0.013326400518417358
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,12,1,128,1,float16,float16,0,0.012399999797344208
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,12,4,128,1,float16,float16,0,0.012444800138473511
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,12,12,128,1,float16,float16,0,0.01263359934091568
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,12,1,128,1,float16,float16,0,0.011907199770212174
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,12,2,128,1,float16,float16,0,0.01204800009727478
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,12,4,128,1,float16,float16,0,0.012123200297355651
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,12,12,128,1,float16,float16,0,0.01234399974346161
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,12,4,128,1,float16,float16,0,0.016334399580955505
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,12,2,128,1,float16,float16,0,0.01196800023317337
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,12,4,128,1,float16,float16,0,0.011908800154924393
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,12,1,128,1,float16,float16,0,0.16745760440826415
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,12,2,128,1,float16,float16,0,0.1916592001914978
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,12,4,128,1,float16,float16,0,0.26197760105133056
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,12,12,128,1,float16,float16,0,0.21820640563964844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,12,1,128,1,float16,float16,0,0.050537598133087155
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,12,2,128,1,float16,float16,0,0.05194079875946045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,12,4,128,1,float16,float16,0,0.0667967975139618
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,12,12,128,1,float16,float16,0,0.05400480031967163
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,12,1,128,1,float16,float16,0,0.029267200827598573
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,12,2,128,1,float16,float16,0,0.03011679947376251
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,12,4,128,1,float16,float16,0,0.03722079992294312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,12,2,128,1,float16,float16,0,0.48552799224853516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,12,12,128,1,float16,float16,0,0.03033280074596405
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,12,1,128,1,float16,float16,0,0.020635199546813966
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,12,1,128,1,float16,float16,0,0.05184800028800964
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,12,4,128,1,float16,float16,0,0.021671999990940095
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,12,12,128,1,float16,float16,0,0.020751999318599702
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,12,1,128,1,float16,float16,0,0.012907199561595917
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,12,2,128,1,float16,float16,0,0.021270400285720824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,12,2,128,1,float16,float16,0,0.013081599771976472
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,12,4,128,1,float16,float16,0,0.013553600013256072
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,12,12,128,1,float16,float16,0,0.012956799566745758
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,12,4,128,1,float16,float16,0,0.03301759958267212
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,12,1,128,1,float16,float16,0,0.012150400131940842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,12,2,128,1,float16,float16,0,0.012123200297355651
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,12,4,128,1,float16,float16,0,0.012331199645996094
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,12,2,128,1,float16,float16,0,0.011846400052309036
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,12,12,128,1,float16,float16,0,0.012062399834394454
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,12,12,128,1,float16,float16,0,0.011723200231790543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,12,4,128,1,float16,float16,0,0.012019199877977371
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,12,1,128,1,float16,float16,0,0.012172800302505494
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,12,1,128,1,float16,float16,0,0.011721599847078323
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,12,2,128,1,float16,float16,0,0.011748799681663513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,12,12,128,1,float16,float16,0,0.01170559972524643
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,12,2,128,1,float16,float16,0,0.011774399876594543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,12,4,128,1,float16,float16,0,0.011700800061225891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,12,1,128,1,float16,float16,0,0.011987199634313583
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,12,2,128,1,float16,float16,0,0.012355200201272964
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,12,1,128,1,float16,float16,0,0.0509984016418457
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,12,2,128,1,float16,float16,0,0.0630079984664917
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,12,4,128,1,float16,float16,0,0.06624959707260132
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,12,1,128,1,float16,float16,0,0.02933279871940613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,12,2,128,1,float16,float16,0,0.035545599460601804
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,12,12,128,1,float16,float16,0,0.0835312008857727
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,12,12,128,1,float16,float16,0,0.04734399914741516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,12,1,128,1,float16,float16,0,0.020732800662517547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,12,4,128,1,float16,float16,0,0.021673600375652313
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,12,12,128,1,float16,float16,0,0.02648319900035858
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,12,2,128,1,float16,float16,0,0.02146400064229965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,12,1,128,1,float16,float16,0,0.013177600502967835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,12,2,128,1,float16,float16,0,0.013279999792575835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,12,4,128,1,float16,float16,0,0.013804799318313599
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,12,12,128,1,float16,float16,0,0.018963199853897095
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,12,2,128,1,float16,float16,0,0.012035199999809265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,12,4,128,1,float16,float16,0,0.012110400199890136
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,12,1,128,1,float16,float16,0,0.011715199798345566
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,12,12,128,1,float16,float16,0,0.01218239963054657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,12,2,128,1,float16,float16,0,0.01183520033955574
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,12,4,128,1,float16,float16,0,0.011671999841928482
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,12,1,128,1,float16,float16,0,0.011692799627780914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,12,2,128,1,float16,float16,0,0.011711999773979187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,12,12,128,1,float16,float16,0,0.011879999935626984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,12,4,128,1,float16,float16,0,0.011593600362539291
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,12,12,128,1,float16,float16,0,0.011403200030326844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,12,2,128,1,float16,float16,0,0.011718399822711945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,12,1,128,1,float16,float16,0,0.011879999935626984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,12,12,128,1,float16,float16,0,0.011432000249624253
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,12,1,128,1,float16,float16,0,0.01186240017414093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,12,4,128,1,float16,float16,0,0.011681599915027619
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,12,1,128,1,float16,float16,0,0.035515201091766355
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,12,2,128,1,float16,float16,0,0.036483201384544375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,12,4,128,1,float16,float16,0,0.05958880186080932
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,12,12,128,1,float16,float16,0,0.07710719704627991
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,12,4,128,1,float16,float16,0,0.011785600334405899
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,12,1,128,1,float16,float16,0,0.021252800524234772
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,12,2,128,1,float16,float16,0,0.021411199867725373
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,12,12,128,1,float16,float16,0,0.04395999908447266
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,12,4,128,1,float16,float16,0,0.0332175999879837
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,12,1,128,1,float16,float16,0,0.013374400138854981
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,12,4,128,1,float16,float16,0,0.03675200045108795
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,12,2,128,1,float16,float16,0,0.013620799779891968
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,12,4,128,1,float16,float16,0,0.019259199500083923
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,12,1,128,1,float16,float16,0,0.012318400293588638
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,12,12,128,1,float16,float16,0,0.024823999404907225
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,12,4,128,1,float16,float16,0,0.012244799733161926
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,12,1,128,1,float16,float16,0,0.011691199988126755
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,12,2,128,1,float16,float16,0,0.011407999694347382
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,12,1,128,1,float16,float16,0,0.011939200013875962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,12,4,128,1,float16,float16,0,0.011307200044393539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,12,1,128,1,float16,float16,0,0.011579199880361556
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,12,12,128,1,float16,float16,0,0.011979199945926666
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,12,2,128,1,float16,float16,0,0.011340799927711486
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,12,12,128,1,float16,float16,0,0.011407999694347382
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,12,4,128,1,float16,float16,0,0.011420799791812897
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,12,1,128,1,float16,float16,0,0.011552000045776367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,12,4,128,1,float16,float16,0,0.010974399745464325
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,12,2,128,1,float16,float16,0,0.011908800154924393
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,12,4,128,1,float16,float16,0,0.011432000249624253
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,12,12,128,1,float16,float16,0,0.011310400068759918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,12,1,128,1,float16,float16,0,0.011959999799728394
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,12,2,128,1,float16,float16,0,0.011708799749612808
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,12,12,128,1,float16,float16,0,0.011612799763679505
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,12,2,128,1,float16,float16,0,0.011692799627780914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,12,4,128,1,float16,float16,0,0.011327999830245971
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,12,1,128,1,float16,float16,0,0.019811199605464937
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,12,2,128,1,float16,float16,0,0.0318448007106781
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,12,4,128,1,float16,float16,0,0.05555840134620667
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,12,12,128,1,float16,float16,0,0.07258399724960327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,12,1,128,1,float16,float16,0,0.012848000228404998
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,12,1,128,1,float16,float16,0,0.011825600266456604
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,12,2,128,1,float16,float16,0,0.01881600022315979
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,12,2,128,1,float16,float16,0,0.012111999839544297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,12,4,128,1,float16,float16,0,0.03057120144367218
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,12,12,128,1,float16,float16,0,0.017875200510025023
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,12,1,128,1,float16,float16,0,0.012012799829244613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,12,12,128,1,float16,float16,0,0.024001599848270418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,12,1,128,1,float16,float16,0,0.01165440008044243
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,12,4,128,1,float16,float16,0,0.011638399958610535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,12,2,128,1,float16,float16,0,0.011633600294589996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,12,12,128,1,float16,float16,0,0.01738079935312271
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,12,1,128,1,float16,float16,0,0.011566399782896041
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,12,2,128,1,float16,float16,0,0.010950399935245514
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,12,12,128,1,float16,float16,0,0.011499200016260147
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,12,4,128,1,float16,float16,0,0.011710400134325028
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,12,4,128,1,float16,float16,0,0.010992000252008438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,12,1,128,1,float16,float16,0,0.010859200358390808
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,12,4,128,1,float16,float16,0,0.010902400314807891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,12,12,128,1,float16,float16,0,0.011374399811029435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,12,1,128,1,float16,float16,0,0.011519999802112579
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,12,1,128,1,float16,float16,0,0.010812799632549285
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,12,2,128,1,float16,float16,0,0.010868799686431885
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,12,4,128,1,float16,float16,0,0.010964799672365189
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,12,12,128,1,float16,float16,0,0.011129599809646607
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,12,1,128,1,float16,float16,0,0.01080000028014183
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,12,2,128,1,float16,float16,0,0.01085119992494583
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,12,4,128,1,float16,float16,0,0.01080320030450821
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,12,12,128,1,float16,float16,0,0.011168000102043153
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,12,1,128,1,float16,float16,0,0.010860799998044967
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,12,2,128,1,float16,float16,0,0.010678400099277497
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,12,4,128,1,float16,float16,0,0.010841599851846694
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,12,12,128,1,float16,float16,0,0.041991999745368956
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,12,2,128,1,float16,float16,0,0.012196800112724305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,12,4,128,1,float16,float16,0,0.0181551992893219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,12,2,128,1,float16,float16,0,0.011392000317573547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,12,2,128,1,float16,float16,0,0.011470399796962738
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,8,1,128,1,float16,float16,0,10.144966125488281
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,8,8,128,1,float16,float16,0,10.924398040771484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,8,2,128,1,float16,float16,0,9.889798736572265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,8,8,128,1,float16,float16,0,4.98780632019043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,8,4,128,1,float16,float16,0,10.72775650024414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,8,8,128,1,float16,float16,0,2.7605215072631837
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,8,1,128,1,float16,float16,0,2.4494415283203126
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,8,1,128,1,float16,float16,0,5.124856185913086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,8,2,128,1,float16,float16,0,4.992153549194336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,8,4,128,1,float16,float16,0,5.190945434570312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,8,2,128,1,float16,float16,0,2.4174640655517576
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,8,4,128,1,float16,float16,0,2.5381519317626955
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,8,1,128,1,float16,float16,0,20.18543243408203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,8,2,128,1,float16,float16,0,20.76007537841797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,8,4,128,1,float16,float16,0,21.051272583007812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,8,1,128,1,float16,float16,0,5.6347393035888675
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,8,2,128,1,float16,float16,0,5.840016174316406
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,8,8,128,1,float16,float16,0,6.287936019897461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,8,8,128,1,float16,float16,0,3.0466144561767576
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,8,4,128,1,float16,float16,0,6.1075889587402346
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,8,8,128,1,float16,float16,0,1.6195455551147462
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,8,1,128,1,float16,float16,0,2.7796335220336914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,8,1,128,1,float16,float16,0,1.4796095848083497
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,8,2,128,1,float16,float16,0,2.8078447341918946
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,8,4,128,1,float16,float16,0,2.960840034484863
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,8,1,128,1,float16,float16,0,11.823251342773437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,8,2,128,1,float16,float16,0,1.4606143951416015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,8,2,128,1,float16,float16,0,11.813768005371093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,8,4,128,1,float16,float16,0,1.5365455627441407
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,8,4,128,1,float16,float16,0,12.15591049194336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,8,8,128,1,float16,float16,0,2.1265392303466797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,8,1,128,1,float16,float16,0,3.923828887939453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,8,8,128,1,float16,float16,0,4.389702224731446
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,8,2,128,1,float16,float16,0,4.015331268310547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,8,4,128,1,float16,float16,0,4.1685951232910154
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,8,1,128,1,float16,float16,0,1.8854320526123047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,8,2,128,1,float16,float16,0,2.0455888748168944
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,8,8,128,1,float16,float16,0,1.3147919654846192
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,8,1,128,1,float16,float16,0,1.0499152183532714
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,8,2,128,1,float16,float16,0,1.0856783866882325
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,8,4,128,1,float16,float16,0,2.074198341369629
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,8,1,128,1,float16,float16,0,8.620673370361327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,8,4,128,1,float16,float16,0,1.0808176040649413
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,8,2,128,1,float16,float16,0,8.615521240234376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,8,4,128,1,float16,float16,0,8.61534423828125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,8,8,128,1,float16,float16,0,2.9114448547363283
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,8,2,128,1,float16,float16,0,5.561515045166016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,8,1,128,1,float16,float16,0,5.654926300048828
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,8,8,128,1,float16,float16,0,6.362702560424805
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,8,4,128,1,float16,float16,0,5.78319206237793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,8,1,128,1,float16,float16,0,2.6282480239868162
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,8,8,128,1,float16,float16,0,1.4758000373840332
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,8,1,128,1,float16,float16,0,1.3321711540222168
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,8,2,128,1,float16,float16,0,2.5606687545776365
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,8,2,128,1,float16,float16,0,1.3536080360412597
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,8,4,128,1,float16,float16,0,2.7878095626831056
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,8,8,128,1,float16,float16,0,0.7443984031677247
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,8,1,128,1,float16,float16,0,0.7289584159851075
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,8,2,128,1,float16,float16,0,0.72358078956604
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,8,1,128,1,float16,float16,0,11.026165008544922
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,8,2,128,1,float16,float16,0,11.075788879394532
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,8,4,128,1,float16,float16,0,0.816260814666748
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,8,4,128,1,float16,float16,0,11.64637451171875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,8,2,128,1,float16,float16,0,2.878593635559082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,8,1,128,1,float16,float16,0,2.830816078186035
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,8,4,128,1,float16,float16,0,1.328007984161377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,8,8,128,1,float16,float16,0,3.3569713592529298
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,8,4,128,1,float16,float16,0,3.1745296478271485
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,8,1,128,1,float16,float16,0,1.491865634918213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,8,2,128,1,float16,float16,0,1.3915776252746581
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,8,8,128,1,float16,float16,0,1.7180992126464845
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,8,1,128,1,float16,float16,0,5.926510238647461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,8,4,128,1,float16,float16,0,1.5650511741638184
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,8,2,128,1,float16,float16,0,6.041019058227539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,8,2,128,1,float16,float16,0,0.7062431812286377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,8,8,128,1,float16,float16,0,0.9533056259155274
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,8,1,128,1,float16,float16,0,0.7502208232879639
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,8,1,128,1,float16,float16,0,0.4762671947479248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,8,4,128,1,float16,float16,0,6.8261566162109375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,8,4,128,1,float16,float16,0,0.8691408157348632
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,8,4,128,1,float16,float16,0,0.4302544116973877
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,8,2,128,1,float16,float16,0,0.509991979598999
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,8,1,128,1,float16,float16,0,2.5618528366088866
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,8,2,128,1,float16,float16,0,2.6587968826293946
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,8,8,128,1,float16,float16,0,0.45506877899169923
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,8,4,128,1,float16,float16,0,2.8217615127563476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,8,8,128,1,float16,float16,0,3.213550567626953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,8,1,128,1,float16,float16,0,1.3368687629699707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,8,2,128,1,float16,float16,0,1.2561887741088866
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,8,8,128,1,float16,float16,0,1.6599056243896484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,8,4,128,1,float16,float16,0,1.433521556854248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,8,1,128,1,float16,float16,0,5.48046875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,8,8,128,1,float16,float16,0,0.8609888076782226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,8,1,128,1,float16,float16,0,0.6711103916168213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,8,2,128,1,float16,float16,0,5.760116958618164
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,8,2,128,1,float16,float16,0,0.6619584083557128
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,8,8,128,1,float16,float16,0,0.4143072128295898
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,8,2,128,1,float16,float16,0,0.3798655986785889
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,8,4,128,1,float16,float16,0,0.7474480152130127
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,8,4,128,1,float16,float16,0,0.39584319591522216
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,8,8,128,1,float16,float16,0,0.2532831907272339
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,8,1,128,1,float16,float16,0,0.41397600173950194
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,8,4,128,1,float16,float16,0,6.16146240234375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,8,2,128,1,float16,float16,0,0.24497439861297607
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,8,1,128,1,float16,float16,0,0.2394239902496338
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,8,4,128,1,float16,float16,0,0.24593920707702638
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,8,1,128,1,float16,float16,0,1.3630720138549806
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,8,2,128,1,float16,float16,0,1.4721440315246581
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,8,4,128,1,float16,float16,0,1.6792272567749023
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,8,8,128,1,float16,float16,0,2.008478355407715
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,8,1,128,1,float16,float16,0,0.7415056228637695
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,8,8,128,1,float16,float16,0,1.0070544242858888
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,8,1,128,1,float16,float16,0,3.0029743194580076
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,8,2,128,1,float16,float16,0,0.8399439811706543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,8,2,128,1,float16,float16,0,3.1209743499755858
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,8,4,128,1,float16,float16,0,3.4527889251708985
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,8,8,128,1,float16,float16,0,0.5379776000976563
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,8,1,128,1,float16,float16,0,0.41049761772155763
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,8,2,128,1,float16,float16,0,0.4972991943359375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,8,8,128,1,float16,float16,0,0.2946399927139282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,8,2,128,1,float16,float16,0,0.2440095901489258
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,8,4,128,1,float16,float16,0,0.2473439931869507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,8,8,128,1,float16,float16,0,0.1832784056663513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,8,1,128,1,float16,float16,0,0.2905424118041992
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,8,1,128,1,float16,float16,0,0.18543360233306885
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,8,2,128,1,float16,float16,0,0.16491520404815674
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,8,8,128,1,float16,float16,0,2.0720367431640625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,8,1,128,1,float16,float16,0,2.8175119400024413
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,8,4,128,1,float16,float16,0,0.868779182434082
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,8,4,128,1,float16,float16,0,0.4385488033294678
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,8,2,128,1,float16,float16,0,3.211638259887695
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,8,4,128,1,float16,float16,0,3.3293998718261717
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,8,4,128,1,float16,float16,0,0.1624959945678711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,8,1,128,1,float16,float16,0,1.3023119926452638
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,8,8,128,1,float16,float16,0,1.0683024406433106
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,8,1,128,1,float16,float16,0,0.7321807861328125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,8,4,128,1,float16,float16,0,0.8549296379089355
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,8,8,128,1,float16,float16,0,0.5343567848205566
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,8,2,128,1,float16,float16,0,1.5843184471130372
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,8,4,128,1,float16,float16,0,1.6955392837524415
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,8,2,128,1,float16,float16,0,0.844974422454834
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,8,2,128,1,float16,float16,0,0.3652368068695068
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,8,8,128,1,float16,float16,0,0.23412320613861085
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,8,1,128,1,float16,float16,0,0.21200320720672608
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,8,4,128,1,float16,float16,0,0.443939208984375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,8,8,128,1,float16,float16,0,0.15584319829940796
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,8,1,128,1,float16,float16,0,0.12252320051193237
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,8,2,128,1,float16,float16,0,0.251310396194458
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,8,4,128,1,float16,float16,0,0.22778398990631105
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,8,4,128,1,float16,float16,0,0.13750239610671997
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,8,1,128,1,float16,float16,0,0.08613600134849549
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,8,2,128,1,float16,float16,0,0.12862720489501953
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,8,2,128,1,float16,float16,0,0.07190399765968322
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,8,4,128,1,float16,float16,0,0.08455520272254943
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,8,1,128,1,float16,float16,0,0.872332763671875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,8,2,128,1,float16,float16,0,1.0384575843811035
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,8,8,128,1,float16,float16,0,1.5100272178649903
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,8,1,128,1,float16,float16,0,1.7629552841186524
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,8,2,128,1,float16,float16,0,1.9000320434570312
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,8,4,128,1,float16,float16,0,2.3306816101074217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,8,1,128,1,float16,float16,0,0.3444655895233154
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,8,8,128,1,float16,float16,0,0.7892911911010743
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,8,1,128,1,float16,float16,0,0.4570608139038086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,8,4,128,1,float16,float16,0,1.15806884765625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,8,2,128,1,float16,float16,0,0.5134704113006592
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,8,8,128,1,float16,float16,0,0.08822240233421326
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,8,1,128,1,float16,float16,0,0.22911520004272462
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,8,4,128,1,float16,float16,0,0.6000368118286132
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,8,8,128,1,float16,float16,0,0.4122176170349121
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,8,2,128,1,float16,float16,0,0.13070720434188843
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,8,4,128,1,float16,float16,0,0.13733279705047607
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,8,8,128,1,float16,float16,0,0.14864799976348878
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,8,1,128,1,float16,float16,0,0.13954399824142455
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,8,2,128,1,float16,float16,0,0.23850080966949463
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,8,8,128,1,float16,float16,0,0.09026399850845337
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,8,2,128,1,float16,float16,0,0.08287519812583924
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,8,1,128,1,float16,float16,0,0.09047039747238159
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,8,4,128,1,float16,float16,0,0.08506720066070557
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,8,8,128,1,float16,float16,0,0.06662719845771789
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,8,2,128,1,float16,float16,0,0.058804798126220706
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,8,1,128,1,float16,float16,0,0.051881599426269534
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,8,4,128,1,float16,float16,0,0.06515359878540039
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,8,1,128,1,float16,float16,0,0.8991888046264649
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,8,2,128,1,float16,float16,0,1.0111295700073242
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,8,4,128,1,float16,float16,0,1.2364255905151367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,8,1,128,1,float16,float16,0,1.7715087890625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,8,8,128,1,float16,float16,0,1.7206575393676757
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,8,2,128,1,float16,float16,0,2.0145343780517577
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,8,1,128,1,float16,float16,0,0.4597792148590088
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,8,4,128,1,float16,float16,0,2.457700729370117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,8,8,128,1,float16,float16,0,0.8931728363037109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,8,4,128,1,float16,float16,0,0.27830240726470945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,8,2,128,1,float16,float16,0,0.5188240051269531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,8,8,128,1,float16,float16,0,0.47403841018676757
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,8,4,128,1,float16,float16,0,0.6537392139434814
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,8,1,128,1,float16,float16,0,0.24250080585479736
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,8,2,128,1,float16,float16,0,0.2609040021896362
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,8,1,128,1,float16,float16,0,0.12793920040130616
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,8,2,128,1,float16,float16,0,0.13158559799194336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,8,4,128,1,float16,float16,0,0.13899999856948853
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,8,2,128,1,float16,float16,0,0.08567839860916138
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,8,4,128,1,float16,float16,0,0.33218719959259035
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,8,1,128,1,float16,float16,0,0.08449919819831848
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,8,8,128,1,float16,float16,0,0.09678400158882142
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,8,8,128,1,float16,float16,0,0.05729280114173889
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,8,4,128,1,float16,float16,0,0.08814560174942017
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,8,1,128,1,float16,float16,0,0.05294240117073059
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,8,1,128,1,float16,float16,0,0.03901920020580292
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,8,4,128,1,float16,float16,0,0.05510879755020141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,8,8,128,1,float16,float16,0,0.04209119975566864
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,8,4,128,1,float16,float16,0,0.040067198872566226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,8,2,128,1,float16,float16,0,0.055276799201965335
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,8,1,128,1,float16,float16,0,0.6297728061676026
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,8,1,128,1,float16,float16,0,1.2619983673095703
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,8,2,128,1,float16,float16,0,1.638711929321289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,8,8,128,1,float16,float16,0,1.7066848754882813
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,8,2,128,1,float16,float16,0,0.8235360145568847
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,8,4,128,1,float16,float16,0,2.255088043212891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,8,8,128,1,float16,float16,0,0.19535839557647705
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,8,1,128,1,float16,float16,0,0.32784960269927976
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,8,2,128,1,float16,float16,0,0.0392879992723465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,8,8,128,1,float16,float16,0,0.9345567703247071
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,8,4,128,1,float16,float16,0,1.176473617553711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,8,8,128,1,float16,float16,0,0.4335184097290039
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,8,2,128,1,float16,float16,0,0.3909663915634155
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,8,4,128,1,float16,float16,0,0.6066383838653564
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,8,2,128,1,float16,float16,0,0.18849120140075684
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,8,1,128,1,float16,float16,0,0.15775200128555297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,8,2,128,1,float16,float16,0,0.09084640145301819
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,8,8,128,1,float16,float16,0,0.16002240180969238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,8,1,128,1,float16,float16,0,0.08882880210876465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,8,4,128,1,float16,float16,0,0.2532592058181763
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,8,4,128,1,float16,float16,0,0.09602400064468383
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,8,8,128,1,float16,float16,0,0.06300640106201172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,8,1,128,1,float16,float16,0,0.052051198482513425
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,8,8,128,1,float16,float16,0,0.037982401251792905
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,8,2,128,1,float16,float16,0,0.05310239791870117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,8,4,128,1,float16,float16,0,0.05662879943847656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,8,2,128,1,float16,float16,0,0.03511680066585541
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,8,8,128,1,float16,float16,0,0.0328575998544693
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,8,1,128,1,float16,float16,0,0.034667199850082396
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,8,4,128,1,float16,float16,0,0.03629119992256165
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,8,2,128,1,float16,float16,0,0.03154079914093018
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,8,4,128,1,float16,float16,0,0.032120001316070554
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,8,1,128,1,float16,float16,0,0.024622400104999543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,8,8,128,1,float16,float16,0,0.026492801308631898
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,8,2,128,1,float16,float16,0,0.024795199930667877
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,8,1,128,1,float16,float16,0,0.25540480613708494
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,8,1,128,1,float16,float16,0,0.5513519763946533
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,8,2,128,1,float16,float16,0,0.32675039768218994
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,8,2,128,1,float16,float16,0,0.7326032161712647
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,8,8,128,1,float16,float16,0,0.8272432327270508
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,8,4,128,1,float16,float16,0,1.0485520362854004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,8,4,128,1,float16,float16,0,0.5076015949249267
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,8,1,128,1,float16,float16,0,0.08283039927482605
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,8,2,128,1,float16,float16,0,0.10262080430984497
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,8,1,128,1,float16,float16,0,0.04740639925003052
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,8,4,128,1,float16,float16,0,0.17388800382614136
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,8,2,128,1,float16,float16,0,0.049614399671554565
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,8,4,128,1,float16,float16,0,0.052129602432250975
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,8,1,128,1,float16,float16,0,0.02894560098648071
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,8,4,128,1,float16,float16,0,0.031590399146080014
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,8,8,128,1,float16,float16,0,0.02027200013399124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,8,1,128,1,float16,float16,0,0.017502400279045104
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,8,2,128,1,float16,float16,0,0.017903999984264375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,8,8,128,1,float16,float16,0,0.01772959977388382
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,8,4,128,1,float16,float16,0,0.01886560022830963
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,8,1,128,1,float16,float16,0,0.01637279987335205
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,8,2,128,1,float16,float16,0,0.016523200273513793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,8,4,128,1,float16,float16,0,0.016953599452972413
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,8,1,128,1,float16,float16,0,0.03145920038223267
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,8,8,128,1,float16,float16,0,0.016627199947834015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,8,1,128,1,float16,float16,0,0.01597920060157776
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,8,4,128,1,float16,float16,0,0.016161599755287172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,8,8,128,1,float16,float16,0,0.016075199842453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,8,1,128,1,float16,float16,0,0.01584320068359375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,8,2,128,1,float16,float16,0,0.01578560024499893
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,8,4,128,1,float16,float16,0,0.025832000374794006
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,8,4,128,1,float16,float16,0,0.015966400504112244
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,8,1,128,1,float16,float16,0,0.25500319004058836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,8,2,128,1,float16,float16,0,0.3296447992324829
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,8,4,128,1,float16,float16,0,0.5056240081787109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,8,8,128,1,float16,float16,0,0.3382575988769531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,8,1,128,1,float16,float16,0,0.06723039746284484
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,8,2,128,1,float16,float16,0,0.07638239860534668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,8,8,128,1,float16,float16,0,0.06059200167655945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,8,1,128,1,float16,float16,0,0.03872480094432831
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,8,4,128,1,float16,float16,0,0.1713536024093628
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,8,2,128,1,float16,float16,0,0.040591999888420105
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,8,8,128,1,float16,float16,0,0.028697600960731505
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,8,8,128,1,float16,float16,0,0.35002400875091555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,8,2,128,1,float16,float16,0,0.023975999653339387
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,8,1,128,1,float16,float16,0,0.02354400008916855
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,8,8,128,1,float16,float16,0,0.016678400337696075
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,8,4,128,1,float16,float16,0,0.02528960108757019
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,8,8,128,1,float16,float16,0,0.03456639945507049
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,8,2,128,1,float16,float16,0,0.014316800236701965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,8,4,128,1,float16,float16,0,0.015167999267578124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,8,2,128,1,float16,float16,0,0.029731199145317078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,8,1,128,1,float16,float16,0,0.012537600100040435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,8,2,128,1,float16,float16,0,0.012729600071907043
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,8,4,128,1,float16,float16,0,0.013167999684810638
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,8,8,128,1,float16,float16,0,0.01279360055923462
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,8,1,128,1,float16,float16,0,0.012080000340938568
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,8,4,128,1,float16,float16,0,0.012240000069141388
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,8,8,128,1,float16,float16,0,0.01231039986014366
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,8,1,128,1,float16,float16,0,0.011934400349855424
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,8,2,128,1,float16,float16,0,0.011958400160074234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,8,4,128,1,float16,float16,0,0.012039999663829803
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,8,8,128,1,float16,float16,0,0.011939200013875962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,8,2,128,1,float16,float16,0,0.01600960046052933
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,8,1,128,1,float16,float16,0,0.011736000329256058
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,8,2,128,1,float16,float16,0,0.011796800047159195
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,8,4,128,1,float16,float16,0,0.011838400363922119
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,8,1,128,1,float16,float16,0,0.06545119881629943
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,8,2,128,1,float16,float16,0,0.07479680180549622
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,8,8,128,1,float16,float16,0,0.08395360112190246
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,8,4,128,1,float16,float16,0,0.1713919997215271
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,8,1,128,1,float16,float16,0,0.03748640120029449
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,8,2,128,1,float16,float16,0,0.03900800049304962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,8,4,128,1,float16,float16,0,0.04220960140228271
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,8,8,128,1,float16,float16,0,0.038468798995018004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,8,1,128,1,float16,float16,0,0.02221599966287613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,8,2,128,1,float16,float16,0,0.02301599979400635
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,8,8,128,1,float16,float16,0,0.022129599750041962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,8,1,128,1,float16,float16,0,0.013948799669742584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,8,4,128,1,float16,float16,0,0.04360480010509491
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,8,2,128,1,float16,float16,0,0.014425599575042724
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,8,8,128,1,float16,float16,0,0.06469119787216186
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,8,8,128,1,float16,float16,0,0.013670399785041809
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,8,4,128,1,float16,float16,0,0.015214399993419647
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,8,1,128,1,float16,float16,0,0.012342400103807449
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,8,2,128,1,float16,float16,0,0.012398400157690049
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,8,1,128,1,float16,float16,0,0.01406559944152832
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,8,8,128,1,float16,float16,0,0.012316799908876418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,8,1,128,1,float16,float16,0,0.011924800276756287
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,8,8,128,1,float16,float16,0,0.013892799615859985
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,8,8,128,1,float16,float16,0,0.011593600362539291
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,8,4,128,1,float16,float16,0,0.012100800126791
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,8,1,128,1,float16,float16,0,0.011814399808645248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,8,2,128,1,float16,float16,0,0.011779200285673141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,8,4,128,1,float16,float16,0,0.011963199824094772
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,8,8,128,1,float16,float16,0,0.011326400190591812
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,8,1,128,1,float16,float16,0,0.011633600294589996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,8,2,128,1,float16,float16,0,0.011711999773979187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,8,4,128,1,float16,float16,0,0.01165120005607605
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,8,2,128,1,float16,float16,0,0.01212640032172203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,8,8,128,1,float16,float16,0,0.011231999844312668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,8,1,128,1,float16,float16,0,0.01170559972524643
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,8,2,128,1,float16,float16,0,0.011608000099658965
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,8,4,128,1,float16,float16,0,0.012041600048542022
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,8,1,128,1,float16,float16,0,0.037987199425697324
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,8,2,128,1,float16,float16,0,0.03935680091381073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,8,4,128,1,float16,float16,0,0.06355199813842774
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,8,1,128,1,float16,float16,0,0.022305600345134735
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,8,2,128,1,float16,float16,0,0.023289600014686586
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,8,4,128,1,float16,float16,0,0.035201600193977355
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,8,8,128,1,float16,float16,0,0.061536002159118655
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,8,8,128,1,float16,float16,0,0.0333983987569809
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,8,1,128,1,float16,float16,0,0.01422239989042282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,8,2,128,1,float16,float16,0,0.014588800072669984
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,8,8,128,1,float16,float16,0,0.019475199282169342
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,8,1,128,1,float16,float16,0,0.012620800733566284
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,8,4,128,1,float16,float16,0,0.013124799728393555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,8,8,128,1,float16,float16,0,0.012305600196123123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,8,1,128,1,float16,float16,0,0.011852800101041793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,8,2,128,1,float16,float16,0,0.011899200081825257
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,8,4,128,1,float16,float16,0,0.011787199974060058
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,8,1,128,1,float16,float16,0,0.011718399822711945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,8,2,128,1,float16,float16,0,0.011681599915027619
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,8,4,128,1,float16,float16,0,0.01136159971356392
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,8,8,128,1,float16,float16,0,0.011159999668598175
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,8,1,128,1,float16,float16,0,0.011617600172758102
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,8,2,128,1,float16,float16,0,0.011615999788045884
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,8,4,128,1,float16,float16,0,0.011235199868679047
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,8,4,128,1,float16,float16,0,0.02452480047941208
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,8,8,128,1,float16,float16,0,0.011020799726247787
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,8,1,128,1,float16,float16,0,0.011628799885511399
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,8,2,128,1,float16,float16,0,0.011598400026559829
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,8,4,128,1,float16,float16,0,0.011700800061225891
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,8,8,128,1,float16,float16,0,0.010969600081443787
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,8,1,128,1,float16,float16,0,0.01156959980726242
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,8,2,128,1,float16,float16,0,0.011553599685430526
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,8,4,128,1,float16,float16,0,0.011163199692964554
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,8,1,128,1,float16,float16,0,0.02322559952735901
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,8,4,128,1,float16,float16,0,0.012910400331020356
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,8,2,128,1,float16,float16,0,0.03478719890117645
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,8,4,128,1,float16,float16,0,0.05870559811592102
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,8,1,128,1,float16,float16,0,0.014767999947071075
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,8,8,128,1,float16,float16,0,0.05759519934654236
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,8,4,128,1,float16,float16,0,0.03244799971580505
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,8,1,128,1,float16,float16,0,0.012932799756526947
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,8,8,128,1,float16,float16,0,0.03177280128002167
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,8,2,128,1,float16,float16,0,0.012960000336170197
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,8,8,128,1,float16,float16,0,0.01844799965620041
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,8,4,128,1,float16,float16,0,0.019299200177192687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,8,1,128,1,float16,float16,0,0.012465599924325943
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,8,2,128,1,float16,float16,0,0.012195199728012085
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,8,4,128,1,float16,float16,0,0.012361600250005721
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,8,4,128,1,float16,float16,0,0.021028800308704375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,8,8,128,1,float16,float16,0,0.012038400024175644
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,8,1,128,1,float16,float16,0,0.011956799775362015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,8,2,128,1,float16,float16,0,0.011593600362539291
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,8,8,128,1,float16,float16,0,0.011444800347089768
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,8,4,128,1,float16,float16,0,0.011582399904727935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,8,8,128,1,float16,float16,0,0.011558400094509124
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,8,1,128,1,float16,float16,0,0.011566399782896041
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,8,8,128,1,float16,float16,0,0.011019200086593628
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,8,4,128,1,float16,float16,0,0.01173119992017746
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,8,1,128,1,float16,float16,0,0.011580800265073776
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,8,4,128,1,float16,float16,0,0.011043199896812439
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,8,2,128,1,float16,float16,0,0.011479999870061874
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,8,8,128,1,float16,float16,0,0.010939200222492219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,8,2,128,1,float16,float16,0,0.011208000034093857
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,8,1,128,1,float16,float16,0,0.011929599940776825
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,8,8,128,1,float16,float16,0,0.010947199910879135
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,8,1,128,1,float16,float16,0,0.011844799667596818
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,8,4,128,1,float16,float16,0,0.011112000048160552
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,8,1,128,1,float16,float16,0,0.019739200174808503
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,8,2,128,1,float16,float16,0,0.031585600972175595
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,8,4,128,1,float16,float16,0,0.05539839863777161
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,8,2,128,1,float16,float16,0,0.012276799976825714
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,8,8,128,1,float16,float16,0,0.054211199283599854
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,8,1,128,1,float16,float16,0,0.012928000092506409
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,8,2,128,1,float16,float16,0,0.018638400733470915
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,8,4,128,1,float16,float16,0,0.030675199627876282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,8,8,128,1,float16,float16,0,0.029873600602149962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,8,2,128,1,float16,float16,0,0.01218239963054657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,8,1,128,1,float16,float16,0,0.01218239963054657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,8,4,128,1,float16,float16,0,0.018060800433158875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,8,8,128,1,float16,float16,0,0.017899200320243835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,8,1,128,1,float16,float16,0,0.011451199650764465
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,8,2,128,1,float16,float16,0,0.011744000017642975
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,8,4,128,1,float16,float16,0,0.011644800007343293
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,8,1,128,1,float16,float16,0,0.010921599715948105
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,8,8,128,1,float16,float16,0,0.01154559999704361
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,8,2,128,1,float16,float16,0,0.012670400738716125
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,8,2,128,1,float16,float16,0,0.010911999642848969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,8,1,128,1,float16,float16,0,0.010831999778747558
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,8,4,128,1,float16,float16,0,0.011348800361156463
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,8,2,128,1,float16,float16,0,0.01085439994931221
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,8,8,128,1,float16,float16,0,0.010915199667215348
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,8,4,128,1,float16,float16,0,0.011268799751996994
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,8,1,128,1,float16,float16,0,0.010870400071144103
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,8,2,128,1,float16,float16,0,0.011180800199508668
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,8,2,128,1,float16,float16,0,0.011457599699497223
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,8,8,128,1,float16,float16,0,0.011360000073909759
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,8,1,128,1,float16,float16,0,0.011006399989128113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,8,2,128,1,float16,float16,0,0.011212799698114395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,8,4,128,1,float16,float16,0,0.01114400029182434
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,8,4,128,1,float16,float16,0,0.010967999696731567
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,8,8,128,1,float16,float16,0,0.011182399839162827
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,8,2,128,1,float16,float16,0,0.010753600299358368
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,8,1,128,1,float16,float16,0,0.010911999642848969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,8,2,128,1,float16,float16,0,0.011099199950695037
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,8,4,128,1,float16,float16,0,0.011321599781513213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,8,2,128,1,float16,float16,0,0.02046400010585785
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,8,8,128,1,float16,float16,0,0.011337599903345107
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,8,4,128,1,float16,float16,0,0.010865599662065507
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,4,1,128,1,float16,float16,0,4.625601577758789
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,4,1,128,1,float16,float16,0,2.329140853881836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,4,4,128,1,float16,float16,0,5.325497436523437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,4,2,128,1,float16,float16,0,2.4059728622436523
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,4,2,128,1,float16,float16,0,5.223278427124024
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,4,4,128,1,float16,float16,0,1.5550784111022948
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,4,1,128,1,float16,float16,0,1.3739279747009276
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,4,2,128,1,float16,float16,0,1.566500759124756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,4,4,128,1,float16,float16,0,2.6245872497558596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,4,1,128,1,float16,float16,0,2.7256048202514647
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,4,1,128,1,float16,float16,0,9.99773941040039
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,4,4,128,1,float16,float16,0,3.036215972900391
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,4,2,128,1,float16,float16,0,2.84674072265625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,4,2,128,1,float16,float16,0,10.38349609375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,4,4,128,1,float16,float16,0,1.535854434967041
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,4,4,128,1,float16,float16,0,0.9196864128112793
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,4,1,128,1,float16,float16,0,1.5236288070678712
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,4,1,128,1,float16,float16,0,5.7293041229248045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,4,1,128,1,float16,float16,0,0.9061424255371093
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,4,2,128,1,float16,float16,0,0.9495632171630859
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,4,2,128,1,float16,float16,0,1.6256288528442382
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,4,2,128,1,float16,float16,0,5.917870330810547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,4,1,128,1,float16,float16,0,1.0510128021240235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,4,4,128,1,float16,float16,0,1.096615982055664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,4,1,128,1,float16,float16,0,1.9051216125488282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,4,2,128,1,float16,float16,0,1.9373327255249024
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,4,4,128,1,float16,float16,0,0.6468463897705078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,4,4,128,1,float16,float16,0,2.1143167495727537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,4,1,128,1,float16,float16,0,0.5964352130889893
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,4,2,128,1,float16,float16,0,1.1308223724365234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,4,2,128,1,float16,float16,0,0.6763472080230712
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,4,1,128,1,float16,float16,0,3.8480831146240235
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,4,2,128,1,float16,float16,0,4.055462265014649
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,4,4,128,1,float16,float16,0,1.3908464431762695
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,4,1,128,1,float16,float16,0,2.410323143005371
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,4,2,128,1,float16,float16,0,2.5981536865234376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,4,1,128,1,float16,float16,0,1.3747424125671386
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,4,4,128,1,float16,float16,0,2.742710494995117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,4,4,128,1,float16,float16,0,0.7456511974334716
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,4,1,128,1,float16,float16,0,0.7859583854675293
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,4,2,128,1,float16,float16,0,0.7529456138610839
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,4,1,128,1,float16,float16,0,0.4279471874237061
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,4,4,128,1,float16,float16,0,0.5487984180450439
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,4,2,128,1,float16,float16,0,0.4783487796783447
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,4,1,128,1,float16,float16,0,5.120775985717773
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,4,2,128,1,float16,float16,0,5.4017585754394535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,4,1,128,1,float16,float16,0,1.3546799659729003
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,4,4,128,1,float16,float16,0,1.6343360900878907
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,4,1,128,1,float16,float16,0,0.7727456092834473
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,4,4,128,1,float16,float16,0,1.0357215881347657
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,4,1,128,1,float16,float16,0,2.8951999664306642
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,4,2,128,1,float16,float16,0,1.435649585723877
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,4,2,128,1,float16,float16,0,0.8136351585388184
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,4,2,128,1,float16,float16,0,2.9928400039672853
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,4,4,128,1,float16,float16,0,0.46041121482849123
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,4,4,128,1,float16,float16,0,0.27723519802093505
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,4,1,128,1,float16,float16,0,0.2934479951858521
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,4,1,128,1,float16,float16,0,0.5370304107666015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,4,2,128,1,float16,float16,0,0.48749279975891113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,4,2,128,1,float16,float16,0,0.3541408061981201
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,4,4,128,1,float16,float16,0,0.8356032371520996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,4,1,128,1,float16,float16,0,1.2565903663635254
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,4,2,128,1,float16,float16,0,1.3586912155151367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,4,4,128,1,float16,float16,0,1.594985580444336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,4,1,128,1,float16,float16,0,0.6252575874328613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,4,2,128,1,float16,float16,0,0.7765520095825196
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,4,1,128,1,float16,float16,0,2.520801544189453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,4,1,128,1,float16,float16,0,0.3839087963104248
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,4,4,128,1,float16,float16,0,0.26814401149749756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,4,2,128,1,float16,float16,0,2.8121456146240233
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,4,2,128,1,float16,float16,0,1.5741583824157714
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,4,4,128,1,float16,float16,0,0.16738239526748658
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,4,2,128,1,float16,float16,0,0.1442352056503296
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,4,1,128,1,float16,float16,0,0.17684799432754517
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,4,4,128,1,float16,float16,0,1.0863583564758301
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,4,1,128,1,float16,float16,0,1.5816783905029297
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,4,2,128,1,float16,float16,0,1.7489328384399414
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,4,1,128,1,float16,float16,0,0.8076815605163574
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,4,4,128,1,float16,float16,0,0.4473775863647461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,4,2,128,1,float16,float16,0,0.3666320085525513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,4,4,128,1,float16,float16,0,0.6005887985229492
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,4,4,128,1,float16,float16,0,0.24444000720977782
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,4,2,128,1,float16,float16,0,0.8979920387268067
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,4,1,128,1,float16,float16,0,0.23489599227905272
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,4,1,128,1,float16,float16,0,0.2947999954223633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,4,2,128,1,float16,float16,0,0.45582242012023927
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,4,1,128,1,float16,float16,0,0.4509888172149658
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,4,2,128,1,float16,float16,0,0.2457119941711426
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,4,4,128,1,float16,float16,0,0.14360159635543823
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,4,1,128,1,float16,float16,0,0.1492079973220825
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,4,2,128,1,float16,float16,0,0.09925439953804016
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,4,1,128,1,float16,float16,0,0.09008640050888062
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,4,1,128,1,float16,float16,0,0.7995728015899658
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,4,2,128,1,float16,float16,0,0.924937629699707
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,4,4,128,1,float16,float16,0,1.147494411468506
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,4,1,128,1,float16,float16,0,1.5796079635620117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,4,2,128,1,float16,float16,0,1.8139072418212892
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,4,1,128,1,float16,float16,0,0.41817121505737304
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,4,4,128,1,float16,float16,0,0.5999728202819824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,4,2,128,1,float16,float16,0,0.4651792049407959
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,4,4,128,1,float16,float16,0,0.29362239837646487
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,4,2,128,1,float16,float16,0,0.2546224117279053
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,4,4,128,1,float16,float16,0,0.1474511981010437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,4,2,128,1,float16,float16,0,0.23090240955352784
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,4,1,128,1,float16,float16,0,0.23131999969482422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,4,1,128,1,float16,float16,0,0.13798400163650512
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,4,2,128,1,float16,float16,0,0.15150560140609742
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,4,4,128,1,float16,float16,0,0.09275040030479431
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,4,1,128,1,float16,float16,0,0.08855519890785217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,4,2,128,1,float16,float16,0,0.09196000099182129
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,4,1,128,1,float16,float16,0,0.05842080116271973
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,4,2,128,1,float16,float16,0,0.061559998989105226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,4,4,128,1,float16,float16,0,0.06198080182075501
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,4,4,128,1,float16,float16,0,0.1076159954071045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,4,2,128,1,float16,float16,0,0.1398095965385437
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,4,1,128,1,float16,float16,0,0.24791359901428223
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,4,2,128,1,float16,float16,0,0.28430399894714353
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,4,4,128,1,float16,float16,0,0.7830687999725342
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,4,1,128,1,float16,float16,0,0.5046895980834961
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,4,2,128,1,float16,float16,0,0.588862419128418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,4,1,128,1,float16,float16,0,0.9702447891235352
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,4,2,128,1,float16,float16,0,1.1509424209594727
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,4,4,128,1,float16,float16,0,0.1578879952430725
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,4,4,128,1,float16,float16,0,0.09818879961967468
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,4,1,128,1,float16,float16,0,0.14467040300369263
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,4,2,128,1,float16,float16,0,0.10273120403289795
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,4,1,128,1,float16,float16,0,0.09038079977035522
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,4,1,128,1,float16,float16,0,0.05871840119361878
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,4,4,128,1,float16,float16,0,0.04108000099658966
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,4,1,128,1,float16,float16,0,0.04132480025291443
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,4,2,128,1,float16,float16,0,0.04026240110397339
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,4,1,128,1,float16,float16,0,0.5208816051483154
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,4,4,128,1,float16,float16,0,0.87674560546875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,4,1,128,1,float16,float16,0,1.0165056228637694
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,4,2,128,1,float16,float16,0,1.2480688095092773
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,4,4,128,1,float16,float16,0,0.3958336114883423
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,4,2,128,1,float16,float16,0,0.6481584072113037
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,4,4,128,1,float16,float16,0,0.46643681526184083
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,4,1,128,1,float16,float16,0,0.26622719764709474
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,4,2,128,1,float16,float16,0,0.3294271945953369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,4,4,128,1,float16,float16,0,0.20563840866088867
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,4,1,128,1,float16,float16,0,0.13648639917373656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,4,2,128,1,float16,float16,0,0.1459391951560974
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,4,2,128,1,float16,float16,0,0.06336799860000611
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,4,4,128,1,float16,float16,0,0.09964479804039002
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,4,1,128,1,float16,float16,0,0.09417120218276978
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,4,4,128,1,float16,float16,0,0.07115839719772339
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,4,2,128,1,float16,float16,0,0.09588000178337097
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,4,1,128,1,float16,float16,0,0.054769599437713624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,4,4,128,1,float16,float16,0,0.04064480066299438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,4,2,128,1,float16,float16,0,0.05478559732437134
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,4,1,128,1,float16,float16,0,0.03953920006752014
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,4,2,128,1,float16,float16,0,0.039843198657035825
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,4,4,128,1,float16,float16,0,0.030505600571632385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,4,1,128,1,float16,float16,0,0.029481598734855653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,4,2,128,1,float16,float16,0,0.02993920147418976
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,4,1,128,1,float16,float16,0,0.38611838817596433
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,4,2,128,1,float16,float16,0,0.5557504177093506
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,4,1,128,1,float16,float16,0,0.814731216430664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,4,4,128,1,float16,float16,0,0.8786335945129394
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,4,2,128,1,float16,float16,0,1.145470428466797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,4,4,128,1,float16,float16,0,0.46312799453735354
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,4,1,128,1,float16,float16,0,0.19071359634399415
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,4,4,128,1,float16,float16,0,0.16359360218048097
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,4,1,128,1,float16,float16,0,0.09096320271492005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,4,4,128,1,float16,float16,0,0.06284160017967225
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,4,1,128,1,float16,float16,0,0.05633760094642639
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,4,2,128,1,float16,float16,0,0.09812960028648376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,4,2,128,1,float16,float16,0,0.14828319549560548
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,4,4,128,1,float16,float16,0,0.03768639862537384
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,4,1,128,1,float16,float16,0,0.03518719971179962
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,4,2,128,1,float16,float16,0,0.05750719904899597
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,4,2,128,1,float16,float16,0,0.03576799929141998
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,4,4,128,1,float16,float16,0,0.056796801090240476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,4,4,128,1,float16,float16,0,0.03324959874153137
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,4,2,128,1,float16,float16,0,0.0323168009519577
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,4,1,128,1,float16,float16,0,0.024900799989700316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,4,4,128,1,float16,float16,0,0.020531199872493744
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,4,2,128,1,float16,float16,0,0.025515198707580566
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,4,1,128,1,float16,float16,0,0.019947199523448943
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,4,2,128,1,float16,float16,0,0.02014559954404831
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,4,1,128,1,float16,float16,0,0.09910719990730285
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,4,1,128,1,float16,float16,0,0.32698080539703367
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,4,4,128,1,float16,float16,0,0.34858078956604005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,4,2,128,1,float16,float16,0,0.1771775960922241
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,4,4,128,1,float16,float16,0,0.09046080112457275
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,4,1,128,1,float16,float16,0,0.04954879879951477
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,4,2,128,1,float16,float16,0,0.05271520018577576
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,4,4,128,1,float16,float16,0,0.03532159924507141
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,4,1,128,1,float16,float16,0,0.030342400074005127
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,4,2,128,1,float16,float16,0,0.031867200136184694
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,4,4,128,1,float16,float16,0,0.020185600221157073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,4,1,128,1,float16,float16,0,0.01796640008687973
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,4,2,128,1,float16,float16,0,0.018833599984645844
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,4,4,128,1,float16,float16,0,0.017708800733089447
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,4,1,128,1,float16,float16,0,0.016571199893951415
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,4,2,128,1,float16,float16,0,0.01690720021724701
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,4,4,128,1,float16,float16,0,0.016654400527477263
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,4,2,128,1,float16,float16,0,0.016251200437545778
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,4,4,128,1,float16,float16,0,0.01613599956035614
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,4,1,128,1,float16,float16,0,0.01587039977312088
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,4,2,128,1,float16,float16,0,0.015931199491024017
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,4,4,128,1,float16,float16,0,0.01592320054769516
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,4,1,128,1,float16,float16,0,0.015732799470424653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,4,2,128,1,float16,float16,0,0.01576800048351288
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,4,1,128,1,float16,float16,0,0.0854319989681244
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,4,2,128,1,float16,float16,0,0.2553567886352539
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,4,2,128,1,float16,float16,0,0.1717311978340149
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,4,4,128,1,float16,float16,0,0.05525919795036316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,4,2,128,1,float16,float16,0,0.04384320080280304
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,4,4,128,1,float16,float16,0,0.02913280129432678
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,4,1,128,1,float16,float16,0,0.041068801283836366
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,4,1,128,1,float16,float16,0,0.024409599602222443
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,4,2,128,1,float16,float16,0,0.02584959864616394
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,4,4,128,1,float16,float16,0,0.017080000042915343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,4,1,128,1,float16,float16,0,0.014628799259662628
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,4,2,128,1,float16,float16,0,0.015460799634456634
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,4,4,128,1,float16,float16,0,0.013892799615859985
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,4,1,128,1,float16,float16,0,0.012703999876976013
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,4,1,128,1,float16,float16,0,0.03142080008983612
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,4,2,128,1,float16,float16,0,0.013062399625778199
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,4,4,128,1,float16,float16,0,0.012758399546146392
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,4,4,128,1,float16,float16,0,0.01218400001525879
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,4,2,128,1,float16,float16,0,0.012326399981975555
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,4,2,128,1,float16,float16,0,0.012065599858760833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,4,4,128,1,float16,float16,0,0.026041600108146667
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,4,1,128,1,float16,float16,0,0.011902400106191636
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,4,2,128,1,float16,float16,0,0.012025599926710128
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,4,4,128,1,float16,float16,0,0.011844799667596818
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,4,1,128,1,float16,float16,0,0.011844799667596818
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,4,2,128,1,float16,float16,0,0.011791999638080596
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,4,4,128,1,float16,float16,0,0.03890720009803772
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,4,1,128,1,float16,float16,0,0.03997440040111542
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,4,2,128,1,float16,float16,0,0.025135999917984007
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,4,1,128,1,float16,float16,0,0.02354079931974411
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,4,2,128,1,float16,float16,0,0.5080463886260986
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,4,4,128,1,float16,float16,0,0.02248159945011139
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,4,1,128,1,float16,float16,0,0.014620800316333771
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,4,4,128,1,float16,float16,0,0.01393119990825653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,4,2,128,1,float16,float16,0,0.015464000403881073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,4,1,128,1,float16,float16,0,0.012812800705432892
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,4,2,128,1,float16,float16,0,0.013232000172138214
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,4,4,128,1,float16,float16,0,0.012316799908876418
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,4,2,128,1,float16,float16,0,0.012068799883127212
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,4,1,128,1,float16,float16,0,0.016092799603939056
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,4,1,128,1,float16,float16,0,0.011883199959993363
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,4,2,128,1,float16,float16,0,0.011883199959993363
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,4,4,128,1,float16,float16,0,0.011417599767446518
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,4,1,128,1,float16,float16,0,0.011727999895811081
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,4,2,128,1,float16,float16,0,0.011727999895811081
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,4,4,128,1,float16,float16,0,0.01128000020980835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,4,1,128,1,float16,float16,0,0.01165120005607605
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,4,2,128,1,float16,float16,0,0.01170239970088005
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,4,4,128,1,float16,float16,0,0.011238399893045425
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,4,1,128,1,float16,float16,0,0.011577600240707397
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,4,2,128,1,float16,float16,0,0.011627200245857238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,4,1,128,1,float16,float16,0,0.024249599874019624
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,4,2,128,1,float16,float16,0,0.036432000994682315
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,4,1,128,1,float16,float16,0,0.012214399874210358
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,4,4,128,1,float16,float16,0,0.03416320085525513
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,4,1,128,1,float16,float16,0,0.015136000514030457
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,4,1,128,1,float16,float16,0,0.01204800009727478
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,4,4,128,1,float16,float16,0,0.011956799775362015
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,4,2,128,1,float16,float16,0,0.021323199570178985
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,4,4,128,1,float16,float16,0,0.019687999784946442
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,4,1,128,1,float16,float16,0,0.013441599905490875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,4,4,128,1,float16,float16,0,0.012622399628162384
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,4,2,128,1,float16,float16,0,0.01377120018005371
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,4,4,128,1,float16,float16,0,0.01148959994316101
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,4,2,128,1,float16,float16,0,0.012371200323104858
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,4,2,128,1,float16,float16,0,0.011420799791812897
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,4,2,128,1,float16,float16,0,0.0431551992893219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,4,1,128,1,float16,float16,0,0.01165440008044243
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,4,4,128,1,float16,float16,0,0.011531200259923935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,4,4,128,1,float16,float16,0,0.011006399989128113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,4,2,128,1,float16,float16,0,0.011640000343322753
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,4,1,128,1,float16,float16,0,0.011953599750995636
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,4,4,128,1,float16,float16,0,0.011310400068759918
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,4,2,128,1,float16,float16,0,0.011457599699497223
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,4,1,128,1,float16,float16,0,0.012036799639463424
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,4,4,128,1,float16,float16,0,0.011247999966144562
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,4,1,128,1,float16,float16,0,0.011508800089359283
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,4,4,128,1,float16,float16,0,0.011596799641847611
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,4,2,128,1,float16,float16,0,0.011488000303506852
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,4,1,128,1,float16,float16,0,0.02126239985227585
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,4,2,128,1,float16,float16,0,0.033390399813652036
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,4,1,128,1,float16,float16,0,0.013582399487495423
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,4,4,128,1,float16,float16,0,0.032355201244354245
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,4,2,128,1,float16,float16,0,0.01951040029525757
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,4,4,128,1,float16,float16,0,0.0187376007437706
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,4,2,128,1,float16,float16,0,0.012358400225639343
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,4,4,128,1,float16,float16,0,0.011931200325489045
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,4,1,128,1,float16,float16,0,0.012624000012874604
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,4,1,128,1,float16,float16,0,0.011630400270223617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,4,2,128,1,float16,float16,0,0.011627200245857238
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,4,1,128,1,float16,float16,0,0.01120320037007332
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,4,4,128,1,float16,float16,0,0.011540800333023071
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,4,4,128,1,float16,float16,0,0.011025600135326385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,4,2,128,1,float16,float16,0,0.010966400057077408
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,4,1,128,1,float16,float16,0,0.01143999993801117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,4,4,128,1,float16,float16,0,0.010943999886512757
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,4,2,128,1,float16,float16,0,0.01093600019812584
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,4,4,128,1,float16,float16,0,0.010876800119876861
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,4,1,128,1,float16,float16,0,0.011025600135326385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,4,2,128,1,float16,float16,0,0.010921599715948105
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,4,4,128,1,float16,float16,0,0.010868799686431885
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,4,1,128,1,float16,float16,0,0.010982400178909302
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,4,2,128,1,float16,float16,0,0.010871999710798264
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,4,1,128,1,float16,float16,0,0.019529600441455842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,4,1,128,1,float16,float16,0,0.01234079971909523
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,4,2,128,1,float16,float16,0,0.03127520084381104
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,4,1,128,1,float16,float16,0,0.01250240057706833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,4,4,128,1,float16,float16,0,0.030638399720191955
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,4,1,128,1,float16,float16,0,0.01215519979596138
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,4,4,128,1,float16,float16,0,0.01814880073070526
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,4,1,128,1,float16,float16,0,0.012070400267839431
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,4,2,128,1,float16,float16,0,0.01223199963569641
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,4,2,128,1,float16,float16,0,0.011215999722480774
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,4,4,128,1,float16,float16,0,0.011894399672746659
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,4,2,128,1,float16,float16,0,0.01141439974308014
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,4,1,128,1,float16,float16,0,0.011534400284290314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,4,1,128,1,float16,float16,0,0.01188960000872612
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,4,4,128,1,float16,float16,0,0.011372800171375274
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,4,2,128,1,float16,float16,0,0.011208000034093857
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,4,4,128,1,float16,float16,0,0.01122720018029213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,4,1,128,1,float16,float16,0,0.011187200248241425
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,4,2,128,1,float16,float16,0,0.011228799819946289
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,4,4,128,1,float16,float16,0,0.011151999980211259
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,4,1,128,1,float16,float16,0,0.011145599931478501
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,4,2,128,1,float16,float16,0,0.011153600364923476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,4,4,128,1,float16,float16,0,0.01117279976606369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,4,1,128,1,float16,float16,0,0.011156799644231797
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,4,2,128,1,float16,float16,0,0.011137600243091583
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,4,4,128,1,float16,float16,0,0.011150400340557098
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,4,1,128,1,float16,float16,0,0.011238399893045425
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,4,2,128,1,float16,float16,0,0.011163199692964554
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,4,2,128,1,float16,float16,0,0.011017599701881408
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,4,1,128,1,float16,float16,0,0.01141119971871376
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,2,2,128,1,float16,float16,0,1.3194416046142579
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,4,2,128,1,float16,float16,0,0.018406400084495546
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,2,1,128,1,float16,float16,0,1.2571248054504394
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,2,2,128,1,float16,float16,0,2.5693695068359377
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,4,1,128,1,float16,float16,0,0.010862399637699128
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,2,2,128,1,float16,float16,0,0.8474896430969239
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,2,1,128,1,float16,float16,0,2.397622489929199
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,2,2,128,1,float16,float16,0,0.8205583572387696
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,2,1,128,1,float16,float16,0,0.893404769897461
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,2,2,128,1,float16,float16,0,1.5507167816162108
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,2,1,128,1,float16,float16,0,1.3547167778015137
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,2,2,128,1,float16,float16,0,0.5232431888580322
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,2,1,128,1,float16,float16,0,0.5628992080688476
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16384,2,1,128,1,float16,float16,0,4.811819076538086
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,12288,2,1,128,1,float16,float16,0,2.6034944534301756
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,2,2,128,1,float16,float16,0,0.6462319850921631
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,2,1,128,1,float16,float16,0,0.9779647827148438
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,2,1,128,1,float16,float16,0,0.7394991874694824
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,2,2,128,1,float16,float16,0,1.0925984382629395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,2,2,128,1,float16,float16,0,0.4683055877685547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,2,1,128,1,float16,float16,0,0.4125967979431152
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,10240,2,1,128,1,float16,float16,0,1.9378623962402344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,2,1,128,1,float16,float16,0,0.7747312068939209
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,2,2,128,1,float16,float16,0,0.8314160346984864
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,2,1,128,1,float16,float16,0,0.6928224086761474
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,2,1,128,1,float16,float16,0,0.42713279724121095
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,2,2,128,1,float16,float16,0,0.49094400405883787
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,2,2,128,1,float16,float16,0,1.3906304359436035
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,2,1,128,1,float16,float16,0,1.2700575828552245
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,2,1,128,1,float16,float16,0,0.276307201385498
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,2,2,128,1,float16,float16,0,0.33574719429016114
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,2,2,128,1,float16,float16,0,0.26792159080505373
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,2,1,128,1,float16,float16,0,0.42777280807495116
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,8192,2,1,128,1,float16,float16,0,2.5077871322631835
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,2,1,128,1,float16,float16,0,0.814908790588379
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,2,1,128,1,float16,float16,0,0.25397119522094724
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,2,2,128,1,float16,float16,0,0.15574400424957274
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,2,1,128,1,float16,float16,0,0.16971999406814575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,6144,2,1,128,1,float16,float16,0,1.5234975814819336
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,2,2,128,1,float16,float16,0,0.24240639209747314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,2,2,128,1,float16,float16,0,0.5451519966125489
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,2,1,128,1,float16,float16,0,0.7638783931732178
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,2,2,128,1,float16,float16,0,0.8821167945861816
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,2,1,128,1,float16,float16,0,0.42281599044799806
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,2,1,128,1,float16,float16,0,0.23822081089019775
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,2,1,128,1,float16,float16,0,0.16383039951324463
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,2,1,128,1,float16,float16,0,0.10035680532455445
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,4096,2,1,128,1,float16,float16,0,1.5280464172363282
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,2,1,128,1,float16,float16,0,0.46064958572387693
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,2,2,128,1,float16,float16,0,0.5680624008178711
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,2,2,128,1,float16,float16,0,0.27654080390930175
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,3072,2,1,128,1,float16,float16,0,0.9077136039733886
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,2,2,128,1,float16,float16,0,0.4846047878265381
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,2,1,128,1,float16,float16,0,0.2552896022796631
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,2,1,128,1,float16,float16,0,0.15759040117263795
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,2,2,128,1,float16,float16,0,0.9019776344299316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,2,1,128,1,float16,float16,0,0.062187200784683226
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,2,2,128,1,float16,float16,0,0.05912479758262634
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,2,2,128,1,float16,float16,0,0.10906080007553101
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,2,1,128,1,float16,float16,0,0.09382240176200866
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,2,1,128,1,float16,float16,0,0.2366640090942383
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,2,2,128,1,float16,float16,0,0.28438239097595214
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,2,1,128,1,float16,float16,0,0.4802688121795654
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,2,2,128,1,float16,float16,0,0.5969039916992187
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,2,2,128,1,float16,float16,0,0.1567456007003784
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,2,1,128,1,float16,float16,0,0.15160319805145264
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,2,2,128,1,float16,float16,0,0.09251999855041504
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,2048,2,1,128,1,float16,float16,0,0.9179167747497559
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,2,1,128,1,float16,float16,0,0.059243202209472656
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,2,1,128,1,float16,float16,0,0.0917472004890442
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,2,2,128,1,float16,float16,0,0.1830639958381653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,2,2,128,1,float16,float16,0,0.03794879913330078
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,2,2,128,1,float16,float16,0,0.09519680142402649
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,2,1,128,1,float16,float16,0,0.03823040127754211
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,2,1,128,1,float16,float16,0,0.15251519680023193
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,2,2,128,1,float16,float16,0,0.1642367959022522
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,2,1,128,1,float16,float16,0,0.28860321044921877
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,2,2,128,1,float16,float16,0,0.39763200283050537
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,2,2,128,1,float16,float16,0,0.10556319952011109
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,2,1,128,1,float16,float16,0,0.09911360144615174
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,2,2,128,1,float16,float16,0,0.06576480269432068
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1536,2,1,128,1,float16,float16,0,0.589518404006958
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,2,2,128,1,float16,float16,0,0.040601599216461184
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,2,1,128,1,float16,float16,0,0.06355839967727661
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,2,2,128,1,float16,float16,0,0.030048000812530517
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,2,2,128,1,float16,float16,0,0.15410399436950684
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,2,2,128,1,float16,float16,0,0.1986016035079956
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,2,1,128,1,float16,float16,0,0.14685920476913453
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,2,1,128,1,float16,float16,0,0.32734880447387693
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,2,2,128,1,float16,float16,0,0.4486847877502441
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,2,2,128,1,float16,float16,0,0.10223679542541504
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1024,2,1,128,1,float16,float16,0,0.6534431934356689
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,2,1,128,1,float16,float16,0,0.09679679870605469
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,2,2,128,1,float16,float16,0,0.05704799890518188
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,2,1,128,1,float16,float16,0,0.05538560152053833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,2,2,128,1,float16,float16,0,0.04101119935512543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,2,1,128,1,float16,float16,0,0.02994079887866974
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,2,1,128,1,float16,float16,0,0.0398032009601593
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,2,2,128,1,float16,float16,0,0.02446240037679672
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,2,1,128,1,float16,float16,0,0.024164800345897675
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,2,2,128,1,float16,float16,0,0.059971201419830325
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,2,2,128,1,float16,float16,0,0.15778399705886842
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,2,1,128,1,float16,float16,0,0.24825119972229004
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,2,2,128,1,float16,float16,0,0.42582240104675295
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,512,2,1,128,1,float16,float16,0,0.5564544200897217
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,2,2,128,1,float16,float16,0,0.06590240001678467
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,2,1,128,1,float16,float16,0,0.09930400252342224
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,2,1,128,1,float16,float16,0,0.058492797613143924
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,2,2,128,1,float16,float16,0,0.03821440041065216
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,2,1,128,1,float16,float16,0,0.03585439920425415
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,2,2,128,1,float16,float16,0,0.033406400680541994
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,2,1,128,1,float16,float16,0,0.029665601253509522
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,2,1,128,1,float16,float16,0,0.0255295991897583
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,2,2,128,1,float16,float16,0,0.026044800877571106
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,2,1,128,1,float16,float16,0,0.020364800095558168
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,2,2,128,1,float16,float16,0,0.020771199464797975
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,2,2,128,1,float16,float16,0,0.018246400356292724
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,2,1,128,1,float16,float16,0,0.018177600204944612
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,2,1,128,1,float16,float16,0,0.05310720205307007
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,2,2,128,1,float16,float16,0,0.03542880117893219
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,2,2,128,1,float16,float16,0,0.07729759812355042
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,256,2,1,128,1,float16,float16,0,0.17668160200119018
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,2,1,128,1,float16,float16,0,0.032102400064468385
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,2,2,128,1,float16,float16,0,0.02056639939546585
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,2,2,128,1,float16,float16,0,0.017710399627685548
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,2,1,128,1,float16,float16,0,0.016918399930000307
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,2,2,128,1,float16,float16,0,0.01663679927587509
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,2,2,128,1,float16,float16,0,0.016092799603939056
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,2,1,128,1,float16,float16,0,0.01629440039396286
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,2,2,128,1,float16,float16,0,0.015939199924468996
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,2,1,128,1,float16,float16,0,0.015857599675655365
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,2,2,128,1,float16,float16,0,0.015795199573040007
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,2,1,128,1,float16,float16,0,0.015772800147533416
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,2,2,128,1,float16,float16,0,0.0304639995098114
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,2,2,128,1,float16,float16,0,0.029684799909591674
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,128,2,1,128,1,float16,float16,0,0.044865599274635314
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,2,1,128,1,float16,float16,0,0.026743999123573302
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,2,2,128,1,float16,float16,0,0.017470400035381316
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,2,1,128,1,float16,float16,0,0.015812799334526062
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,2,2,128,1,float16,float16,0,0.014382399618625641
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,2,1,128,1,float16,float16,0,0.013352000713348388
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,2,2,128,1,float16,float16,0,0.012783999741077422
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,2,2,128,1,float16,float16,0,0.012167999893426895
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,2,1,128,1,float16,float16,0,0.012188799679279327
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,2,2,128,1,float16,float16,0,0.011966399848461151
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,2,2,128,1,float16,float16,0,0.011919999867677689
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,2,1,128,1,float16,float16,0,0.011923199892044068
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,2,2,128,1,float16,float16,0,0.011785600334405899
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,2,1,128,1,float16,float16,0,0.011774399876594543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,64,2,1,128,1,float16,float16,0,0.026291200518608095
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,2,2,128,1,float16,float16,0,0.023123200237751006
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,2,1,128,1,float16,float16,0,0.016161599755287172
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,2,2,128,1,float16,float16,0,0.014351999759674073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,2,1,128,1,float16,float16,0,0.03198719918727875
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,2,1,128,1,float16,float16,0,0.04009599983692169
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,2,1,128,1,float16,float16,0,0.013571199774742127
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,2,2,128,1,float16,float16,0,0.012689599394798278
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,2,2,128,1,float16,float16,0,0.011582399904727935
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,2,1,128,1,float16,float16,0,0.01199679970741272
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,2,2,128,1,float16,float16,0,0.011393599957227708
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,2,1,128,1,float16,float16,0,0.011822400242090225
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,2,2,128,1,float16,float16,0,0.011212799698114395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,2,1,128,1,float16,float16,0,0.011723200231790543
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,2,2,128,1,float16,float16,0,0.011184000223875046
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,2,1,128,1,float16,float16,0,0.011671999841928482
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,2,2,128,1,float16,float16,0,0.01117279976606369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,2,1,128,1,float16,float16,0,0.011631999909877778
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,32,2,1,128,1,float16,float16,0,0.02223680019378662
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,2,2,128,1,float16,float16,0,0.020491200685501098
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,2,1,128,1,float16,float16,0,0.013979199528694152
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,2,2,128,1,float16,float16,0,0.013076800107955932
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,2,1,128,1,float16,float16,0,0.019332799315452575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,2,1,128,1,float16,float16,0,0.012577599287033081
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,2,2,128,1,float16,float16,0,0.011868800222873687
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,2,2,128,1,float16,float16,0,0.011161600053310395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,2,1,128,1,float16,float16,0,0.016046400368213653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,2,1,128,1,float16,float16,0,0.011376000195741653
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,2,1,128,1,float16,float16,0,0.012012799829244613
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,2,2,128,1,float16,float16,0,0.011052799969911575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,2,1,128,1,float16,float16,0,0.011195199936628342
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,2,2,128,1,float16,float16,0,0.011046399921178817
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,2,2,128,1,float16,float16,0,0.010950399935245514
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,2,1,128,1,float16,float16,0,0.011321599781513213
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,2,1,128,1,float16,float16,0,0.011079999804496764
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,2,2,128,1,float16,float16,0,0.011044800281524658
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,2,1,128,1,float16,float16,0,0.010976000130176545
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,2,1,128,1,float16,float16,0,0.013147200644016265
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,2,2,128,1,float16,float16,0,0.019393600523471832
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,2,2,128,1,float16,float16,0,0.012361600250005721
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,2,1,128,1,float16,float16,0,0.012139199674129486
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,2,1,128,1,float16,float16,0,0.01143999993801117
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,2,2,128,1,float16,float16,0,0.011023999750614166
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,2,1,128,1,float16,float16,0,0.01101600006222725
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,2,2,128,1,float16,float16,0,0.010937599837779999
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,2,1,128,1,float16,float16,0,0.010915199667215348
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,2,2,128,1,float16,float16,0,0.010878399759531022
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,2,1,128,1,float16,float16,0,0.010911999642848969
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,2,1,128,1,float16,float16,0,0.012608000636100769
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,2,1,128,1,float16,float16,0,0.01080320030450821
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,2,1,128,1,float16,float16,0,0.012019199877977371
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,2,2,128,1,float16,float16,0,0.0108255997300148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,2,1,128,1,float16,float16,0,0.010830400139093399
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,1,2,1,128,1,float16,float16,0,0.01945919990539551
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,2,2,128,1,float16,float16,0,0.018452799320220946
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,2,1,128,1,float16,float16,0,0.012486399710178375
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,2,2,128,1,float16,float16,0,0.011958400160074234
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,2,1,128,1,float16,float16,0,0.011718399822711945
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,2,2,128,1,float16,float16,0,0.011355199664831162
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,2,1,128,1,float16,float16,0,0.011222399771213531
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,2,2,128,1,float16,float16,0,0.010878399759531022
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,2,1,128,1,float16,float16,0,0.010846400260925293
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,2,2,128,1,float16,float16,0,0.010868799686431885
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,2,2,128,1,float16,float16,0,0.010838399827480315
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,2,1,128,1,float16,float16,0,0.010856000334024429
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,2,1,128,1,float16,float16,0,0.010907199978828431
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,2,2,128,1,float16,float16,0,0.010756800323724747
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,2,1,128,1,float16,float16,0,0.010843200236558914
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,2,2,128,1,float16,float16,0,0.010844799876213073
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,2,1,128,1,float16,float16,0,0.010704000294208527
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,2,1,128,1,float16,float16,0,0.0127920001745224
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16384,1,1,128,1,float16,float16,0,0.5135983943939209
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16384,1,1,128,1,float16,float16,0,0.8005887985229492
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,12288,1,1,128,1,float16,float16,0,0.46544637680053713
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,12288,1,1,128,1,float16,float16,0,0.24988799095153807
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16384,1,1,128,1,float16,float16,0,1.2860608100891113
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,12288,1,1,128,1,float16,float16,0,0.8016592025756836
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,10240,1,1,128,1,float16,float16,0,0.21726880073547364
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,10240,1,1,128,1,float16,float16,0,0.3868383884429932
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,256,16,2,1,128,1,float16,float16,0,0.020732800662517547
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,2,2,128,1,float16,float16,0,0.01159520000219345
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,10240,1,1,128,1,float16,float16,0,0.623472023010254
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,8192,1,1,128,1,float16,float16,0,0.16651040315628052
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,8192,1,1,128,1,float16,float16,0,0.3069776058197021
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,2,2,128,1,float16,float16,0,0.010899200290441512
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,8192,1,1,128,1,float16,float16,0,0.4454160213470459
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,6144,1,1,128,1,float16,float16,0,0.26630558967590334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,8192,1,1,128,1,float16,float16,0,0.7660287857055664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,6144,1,1,128,1,float16,float16,0,0.09694880247116089
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,6144,1,1,128,1,float16,float16,0,0.15358400344848633
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,6144,1,1,128,1,float16,float16,0,0.45556321144104006
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,4096,1,1,128,1,float16,float16,0,0.26040000915527345
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,4096,1,1,128,1,float16,float16,0,0.16533600091934203
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,4096,1,1,128,1,float16,float16,0,0.09520959854125977
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,4096,1,1,128,1,float16,float16,0,0.45148482322692873
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,3072,1,1,128,1,float16,float16,0,0.10533599853515625
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,3072,1,1,128,1,float16,float16,0,0.0589680016040802
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,3072,1,1,128,1,float16,float16,0,0.042019200325012204
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,3072,1,1,128,1,float16,float16,0,0.2676352024078369
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,3072,1,1,128,1,float16,float16,0,0.1703312039375305
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,2048,1,1,128,1,float16,float16,0,0.0921616017818451
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,2048,1,1,128,1,float16,float16,0,0.16024320125579833
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,2048,1,1,128,1,float16,float16,0,0.03858720064163208
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,2048,1,1,128,1,float16,float16,0,0.029344001412391664
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,2048,1,1,128,1,float16,float16,0,0.28715360164642334
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1536,1,1,128,1,float16,float16,0,0.10837759971618652
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1536,1,1,128,1,float16,float16,0,0.16761280298233033
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1536,1,1,128,1,float16,float16,0,0.06656479835510254
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1536,1,1,128,1,float16,float16,0,0.04061599969863892
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1536,1,1,128,1,float16,float16,0,0.030350399017333985
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1536,1,1,128,1,float16,float16,0,0.02473440021276474
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1024,1,1,128,1,float16,float16,0,0.057126402854919434
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1024,1,1,128,1,float16,float16,0,0.040987199544906615
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1024,1,1,128,1,float16,float16,0,0.030513599514961243
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1024,1,1,128,1,float16,float16,0,0.20344960689544678
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1024,1,1,128,1,float16,float16,0,0.023073600232601167
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1024,1,1,128,1,float16,float16,0,0.024796800315380098
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,512,1,1,128,1,float16,float16,0,0.03832640051841736
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,512,1,1,128,1,float16,float16,0,0.15827360153198242
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,512,1,1,128,1,float16,float16,0,0.032950401306152344
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,512,1,1,128,1,float16,float16,0,0.06629120111465454
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,512,1,1,128,1,float16,float16,0,0.026177600026130676
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,512,1,1,128,1,float16,float16,0,0.01842560023069382
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,512,1,1,128,1,float16,float16,0,0.02067680060863495
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,512,1,1,128,1,float16,float16,0,0.01775359958410263
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,256,1,1,128,1,float16,float16,0,0.035750401020050046
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,256,1,1,128,1,float16,float16,0,0.017990399897098542
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,256,1,1,128,1,float16,float16,0,0.021080000698566435
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,256,1,1,128,1,float16,float16,0,0.016206400096416475
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,256,1,1,128,1,float16,float16,0,0.015964800119400026
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,4096,1,1,128,1,float16,float16,0,0.05712959766387939
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,256,1,1,128,1,float16,float16,0,0.01568319946527481
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,256,1,1,128,1,float16,float16,0,0.015987199544906617
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,128,1,1,128,1,float16,float16,0,0.014572800695896148
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,128,1,1,128,1,float16,float16,0,0.017953599989414214
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,128,1,1,128,1,float16,float16,0,0.013052800297737121
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,128,1,1,128,1,float16,float16,0,0.012379200011491776
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,128,1,1,128,1,float16,float16,0,0.011950399726629257
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,128,1,1,128,1,float16,float16,0,0.011800000071525573
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,64,1,1,128,1,float16,float16,0,0.014983999729156493
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,64,1,1,128,1,float16,float16,0,0.012931199371814727
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,64,1,1,128,1,float16,float16,0,0.012055999785661697
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,64,1,1,128,1,float16,float16,0,0.011428800225257874
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,64,1,1,128,1,float16,float16,0,0.011393599957227708
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,64,1,1,128,1,float16,float16,0,0.011212799698114395
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,64,1,1,128,1,float16,float16,0,0.011190400272607804
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,64,1,1,128,1,float16,float16,0,0.011068800091743469
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,2048,1,1,128,1,float16,float16,0,0.059592002630233766
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,32,1,1,128,1,float16,float16,0,0.012308800220489502
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,32,1,1,128,1,float16,float16,0,0.013571199774742127
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,32,1,1,128,1,float16,float16,0,0.011155200004577637
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,32,1,1,128,1,float16,float16,0,0.011827199906110763
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,32,1,1,128,1,float16,float16,0,0.011001600325107575
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,32,1,1,128,1,float16,float16,0,0.010873600095510482
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,32,1,1,128,1,float16,float16,0,0.010883200168609618
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,32,1,1,128,1,float16,float16,0,0.011388800293207168
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,16,1,1,128,1,float16,float16,0,0.013123199343681335
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,16,1,1,128,1,float16,float16,0,0.012001600116491318
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,16,1,1,128,1,float16,float16,0,0.010948800295591355
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,16,1,1,128,1,float16,float16,0,0.011640000343322753
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,16,1,1,128,1,float16,float16,0,0.01093439981341362
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,16,1,1,128,1,float16,float16,0,0.010868799686431885
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,16,1,1,128,1,float16,float16,0,0.010795199871063232
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,16,1,1,128,1,float16,float16,0,0.01106560006737709
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,128,1,1,1,128,1,float16,float16,0,0.012505599856376648
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,64,1,1,1,128,1,float16,float16,0,0.011745599657297134
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,1,1,1,128,1,float16,float16,0,0.010841599851846694
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1,1,1,128,1,float16,float16,0,0.01141280010342598
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,1,1,1,128,1,float16,float16,0,0.010920000076293946
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,4,1,1,1,128,1,float16,float16,0,0.010718400031328202
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,1,1,1,128,1,float16,float16,0,0.010840000212192535
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,32,1024,1,1,128,1,float16,float16,0,0.10315200090408325
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,16,256,1,1,128,1,float16,float16,0,0.016540800034999848
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,8,128,1,1,128,1,float16,float16,0,0.012031999975442886
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,1,128,1,1,128,1,float16,float16,0,0.011787199974060058
SGLang,0.5.9,NVIDIA L40S,context_attention,flash_attention,2,1,1,1,128,1,float16,float16,0,0.010955200344324113
