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,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,96,1,128,1,float16,float16,0,117.67823486328125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,96,96,128,1,float16,float16,0,129.63681640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,96,2,128,1,float16,float16,0,117.0471923828125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,96,4,128,1,float16,float16,0,118.5892822265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,96,1,128,1,float16,float16,0,58.08583984375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,96,1,128,1,float16,float16,0,233.6639404296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,96,96,128,1,float16,float16,0,64.85524291992188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,96,2,128,1,float16,float16,0,57.67659301757813
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,96,2,128,1,float16,float16,0,235.118603515625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,96,4,128,1,float16,float16,0,239.642626953125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,96,8,128,1,float16,float16,0,243.1498291015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,96,1,128,1,float16,float16,0,28.257281494140624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,96,96,128,1,float16,float16,0,31.990579223632814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,96,8,128,1,float16,float16,0,121.19029541015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,96,4,128,1,float16,float16,0,59.1383544921875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,96,8,128,1,float16,float16,0,59.243109130859374
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,96,2,128,1,float16,float16,0,28.393472290039064
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,96,4,128,1,float16,float16,0,28.981707763671874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,96,8,128,1,float16,float16,0,29.647308349609375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,96,2,128,1,float16,float16,0,66.340966796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,96,1,128,1,float16,float16,0,66.41679077148437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,96,96,128,1,float16,float16,0,76.72330322265626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,96,4,128,1,float16,float16,0,68.93870239257812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,96,1,128,1,float16,float16,0,133.87227783203124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,96,2,128,1,float16,float16,0,136.04822998046876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,96,4,128,1,float16,float16,0,138.0801025390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,96,96,128,1,float16,float16,0,37.385369873046876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,96,1,128,1,float16,float16,0,32.79600219726562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,96,2,128,1,float16,float16,0,33.01340026855469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,96,8,128,1,float16,float16,0,136.2491943359375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,96,1,128,1,float16,float16,0,16.15447082519531
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,96,96,128,1,float16,float16,0,18.46307830810547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,96,4,128,1,float16,float16,0,33.06501159667969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,96,2,128,1,float16,float16,0,15.867594909667968
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,96,8,128,1,float16,float16,0,68.36295776367187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,96,4,128,1,float16,float16,0,16.263320922851562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,96,8,128,1,float16,float16,0,33.408102416992186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,96,8,128,1,float16,float16,0,16.546815490722658
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,96,1,128,1,float16,float16,0,46.476287841796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,96,2,128,1,float16,float16,0,46.528564453125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,96,4,128,1,float16,float16,0,46.64289245605469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,96,96,128,1,float16,float16,0,54.3414794921875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,96,2,128,1,float16,float16,0,22.77217254638672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,96,1,128,1,float16,float16,0,93.51470336914062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,96,96,128,1,float16,float16,0,26.618008422851563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,96,1,128,1,float16,float16,0,22.58252868652344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,96,2,128,1,float16,float16,0,94.3299072265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,96,4,128,1,float16,float16,0,96.73897094726563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,96,8,128,1,float16,float16,0,96.36874389648438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,96,8,128,1,float16,float16,0,48.01674194335938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,96,1,128,1,float16,float16,0,11.022335815429688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,96,96,128,1,float16,float16,0,13.220863342285156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,96,2,128,1,float16,float16,0,11.445862579345704
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,96,4,128,1,float16,float16,0,11.779634857177735
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,96,8,128,1,float16,float16,0,11.470489501953125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,96,4,128,1,float16,float16,0,22.968882751464843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,96,8,128,1,float16,float16,0,23.511135864257813
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,96,1,128,1,float16,float16,0,60.265576171875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,96,2,128,1,float16,float16,0,61.126043701171874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,96,4,128,1,float16,float16,0,62.008673095703124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,96,96,128,1,float16,float16,0,74.5248779296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,96,1,128,1,float16,float16,0,29.685147094726563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,96,96,128,1,float16,float16,0,35.83375244140625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,96,1,128,1,float16,float16,0,124.46248779296874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,96,2,128,1,float16,float16,0,30.349465942382814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,96,2,128,1,float16,float16,0,123.138916015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,96,4,128,1,float16,float16,0,125.00428466796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,96,8,128,1,float16,float16,0,127.63990478515625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,96,8,128,1,float16,float16,0,62.85311279296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,96,1,128,1,float16,float16,0,14.701210021972656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,96,96,128,1,float16,float16,0,17.68826904296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,96,2,128,1,float16,float16,0,14.602854919433593
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,96,4,128,1,float16,float16,0,30.45447692871094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,96,4,128,1,float16,float16,0,15.237017822265624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,96,8,128,1,float16,float16,0,15.461631774902344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,96,8,128,1,float16,float16,0,30.68968811035156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,96,96,128,1,float16,float16,0,8.661759948730468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,96,1,128,1,float16,float16,0,7.312640380859375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,96,2,128,1,float16,float16,0,7.193548583984375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,96,4,128,1,float16,float16,0,7.116697692871094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,96,8,128,1,float16,float16,0,7.4302978515625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,96,1,128,1,float16,float16,0,34.691122436523436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,96,2,128,1,float16,float16,0,34.036581420898436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,96,4,128,1,float16,float16,0,35.177728271484376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,96,96,128,1,float16,float16,0,43.31356201171875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,96,1,128,1,float16,float16,0,69.79855346679688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,96,2,128,1,float16,float16,0,71.1605224609375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,96,4,128,1,float16,float16,0,72.0847900390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,96,8,128,1,float16,float16,0,72.95554809570312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,96,96,128,1,float16,float16,0,21.164288330078126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,96,1,128,1,float16,float16,0,16.842291259765624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,96,2,128,1,float16,float16,0,17.396018981933594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,96,8,128,1,float16,float16,0,35.37100830078125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,96,96,128,1,float16,float16,0,10.428518676757813
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,96,1,128,1,float16,float16,0,8.423423767089844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,96,4,128,1,float16,float16,0,17.342976379394532
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,96,2,128,1,float16,float16,0,8.438374328613282
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,96,4,128,1,float16,float16,0,8.315033721923829
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,96,8,128,1,float16,float16,0,17.674290466308594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,96,8,128,1,float16,float16,0,8.673996734619141
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,96,96,128,1,float16,float16,0,5.0860542297363285
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,96,1,128,1,float16,float16,0,3.765398406982422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,96,4,128,1,float16,float16,0,4.108902359008789
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,96,2,128,1,float16,float16,0,4.064972686767578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,96,8,128,1,float16,float16,0,4.0840190887451175
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,96,1,128,1,float16,float16,0,31.29595031738281
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,96,2,128,1,float16,float16,0,31.65767822265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,96,4,128,1,float16,float16,0,32.89323425292969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,96,96,128,1,float16,float16,0,42.578329467773436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,96,1,128,1,float16,float16,0,65.04427490234374
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,96,2,128,1,float16,float16,0,65.7680908203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,96,4,128,1,float16,float16,0,67.09074096679687
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,96,1,128,1,float16,float16,0,15.955763244628907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,96,96,128,1,float16,float16,0,20.53729248046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,96,8,128,1,float16,float16,0,67.87359008789062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,96,2,128,1,float16,float16,0,15.801446533203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,96,8,128,1,float16,float16,0,32.8848388671875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,96,4,128,1,float16,float16,0,16.179405212402344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,96,96,128,1,float16,float16,0,10.346288299560547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,96,1,128,1,float16,float16,0,7.39210205078125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,96,2,128,1,float16,float16,0,7.731249237060547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,96,8,128,1,float16,float16,0,16.6129150390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,96,4,128,1,float16,float16,0,8.003890991210938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,96,8,128,1,float16,float16,0,7.847321319580078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,96,96,128,1,float16,float16,0,5.039155197143555
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,96,1,128,1,float16,float16,0,3.691775894165039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,96,2,128,1,float16,float16,0,3.7534721374511717
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,96,4,128,1,float16,float16,0,3.708313751220703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,96,96,128,1,float16,float16,0,2.536137580871582
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,96,8,128,1,float16,float16,0,3.7132240295410157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,96,1,128,1,float16,float16,0,1.744327926635742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,96,2,128,1,float16,float16,0,1.7582592010498046
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,96,4,128,1,float16,float16,0,1.8290176391601562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,96,8,128,1,float16,float16,0,1.827225685119629
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,96,1,128,1,float16,float16,0,17.965618896484376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,96,2,128,1,float16,float16,0,18.329548645019532
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,96,4,128,1,float16,float16,0,18.72578582763672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,96,96,128,1,float16,float16,0,25.708953857421875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,96,1,128,1,float16,float16,0,36.97059936523438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,96,2,128,1,float16,float16,0,36.95692749023438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,96,4,128,1,float16,float16,0,38.6671630859375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,96,8,128,1,float16,float16,0,38.43722229003906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,96,1,128,1,float16,float16,0,9.064345550537109
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,96,2,128,1,float16,float16,0,8.870502471923828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,96,96,128,1,float16,float16,0,12.689049530029298
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,96,8,128,1,float16,float16,0,18.898074340820312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,96,4,128,1,float16,float16,0,9.291110229492187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,96,1,128,1,float16,float16,0,4.106649780273438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,96,8,128,1,float16,float16,0,9.598770904541016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,96,96,128,1,float16,float16,0,6.357657623291016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,96,2,128,1,float16,float16,0,4.197017669677734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,96,4,128,1,float16,float16,0,4.122828674316406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,96,8,128,1,float16,float16,0,4.215091323852539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,96,96,128,1,float16,float16,0,3.1337472915649416
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,96,1,128,1,float16,float16,0,2.036787223815918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,96,2,128,1,float16,float16,0,2.0155391693115234
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,96,4,128,1,float16,float16,0,2.1230592727661133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,96,8,128,1,float16,float16,0,2.1876224517822265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,96,96,128,1,float16,float16,0,1.5759360313415527
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,96,1,128,1,float16,float16,0,1.0280943870544434
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,96,2,128,1,float16,float16,0,1.0446847915649413
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,96,4,128,1,float16,float16,0,1.0851327896118164
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,96,8,128,1,float16,float16,0,1.0891263961791993
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,96,2,128,1,float16,float16,0,16.968499755859376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,96,1,128,1,float16,float16,0,16.894309997558594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,96,4,128,1,float16,float16,0,17.806028747558592
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,96,1,128,1,float16,float16,0,34.35914001464844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,96,96,128,1,float16,float16,0,26.701412963867188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,96,2,128,1,float16,float16,0,35.231640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,96,4,128,1,float16,float16,0,36.544561767578124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,96,8,128,1,float16,float16,0,37.043661499023436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,96,1,128,1,float16,float16,0,8.380313873291016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,96,96,128,1,float16,float16,0,13.375640869140625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,96,2,128,1,float16,float16,0,8.667903900146484
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,96,4,128,1,float16,float16,0,8.790835571289062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,96,8,128,1,float16,float16,0,18.11584014892578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,96,1,128,1,float16,float16,0,4.046694564819336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,96,8,128,1,float16,float16,0,8.965478515625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,96,96,128,1,float16,float16,0,6.637773132324218
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,96,2,128,1,float16,float16,0,3.868467330932617
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,96,4,128,1,float16,float16,0,4.211404800415039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,96,8,128,1,float16,float16,0,4.210227203369141
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,96,96,128,1,float16,float16,0,3.2524799346923827
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,96,2,128,1,float16,float16,0,1.893120002746582
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,96,1,128,1,float16,float16,0,1.9063295364379882
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,96,4,128,1,float16,float16,0,1.9593727111816406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,96,8,128,1,float16,float16,0,2.0806655883789062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,96,96,128,1,float16,float16,0,1.6540672302246093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,96,1,128,1,float16,float16,0,0.9760255813598633
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,96,2,128,1,float16,float16,0,0.9788415908813477
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,96,8,128,1,float16,float16,0,1.0077183723449707
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,96,4,128,1,float16,float16,0,0.9923520088195801
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,96,1,128,1,float16,float16,0,0.5196800231933594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,96,96,128,1,float16,float16,0,0.830617618560791
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,96,2,128,1,float16,float16,0,0.5188096046447754
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,96,4,128,1,float16,float16,0,0.5328896045684814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,96,8,128,1,float16,float16,0,0.5430784225463867
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,96,2,128,1,float16,float16,0,9.766860961914062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,96,1,128,1,float16,float16,0,10.030899047851562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,96,4,128,1,float16,float16,0,10.476850891113282
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,96,1,128,1,float16,float16,0,20.182170104980468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,96,96,128,1,float16,float16,0,17.090611267089844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,96,2,128,1,float16,float16,0,20.33387451171875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,96,4,128,1,float16,float16,0,21.102796936035155
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,96,8,128,1,float16,float16,0,22.062643432617186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,96,1,128,1,float16,float16,0,4.561868667602539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,96,96,128,1,float16,float16,0,8.570111846923828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,96,2,128,1,float16,float16,0,4.552753448486328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,96,4,128,1,float16,float16,0,5.0864097595214846
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,96,8,128,1,float16,float16,0,10.927257537841797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,96,96,128,1,float16,float16,0,4.353638458251953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,96,8,128,1,float16,float16,0,5.313587188720703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,96,1,128,1,float16,float16,0,2.257304000854492
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,96,2,128,1,float16,float16,0,2.259404754638672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,96,4,128,1,float16,float16,0,2.2866943359375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,96,1,128,1,float16,float16,0,1.1587583541870117
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,96,8,128,1,float16,float16,0,2.463180732727051
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,96,96,128,1,float16,float16,0,2.156492805480957
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,96,2,128,1,float16,float16,0,1.166438388824463
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,96,4,128,1,float16,float16,0,1.1815936088562011
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,96,8,128,1,float16,float16,0,1.2050432205200194
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,96,96,128,1,float16,float16,0,1.0896368026733398
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,96,1,128,1,float16,float16,0,0.592844820022583
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,96,2,128,1,float16,float16,0,0.5892608165740967
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,96,4,128,1,float16,float16,0,0.6082560062408447
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,96,96,128,1,float16,float16,0,0.5578752040863038
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,96,1,128,1,float16,float16,0,0.3122688055038452
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,96,8,128,1,float16,float16,0,0.6295040130615235
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,96,2,128,1,float16,float16,0,0.30458879470825195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,96,4,128,1,float16,float16,0,0.31191039085388184
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,96,8,128,1,float16,float16,0,0.3147264003753662
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,96,2,128,1,float16,float16,0,9.794713592529297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,96,1,128,1,float16,float16,0,9.752985382080078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,96,1,128,1,float16,float16,0,19.901081848144532
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,96,4,128,1,float16,float16,0,10.15572509765625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,96,2,128,1,float16,float16,0,20.132762145996093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,96,4,128,1,float16,float16,0,20.878746032714844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,96,8,128,1,float16,float16,0,21.859890747070313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,96,96,128,1,float16,float16,0,19.724641418457033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,96,1,128,1,float16,float16,0,4.808958435058594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,96,4,128,1,float16,float16,0,4.7572479248046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,96,1,128,1,float16,float16,0,2.200217628479004
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,96,8,128,1,float16,float16,0,5.0443775177001955
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,96,96,128,1,float16,float16,0,5.076326370239258
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,96,2,128,1,float16,float16,0,4.721561431884766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,96,96,128,1,float16,float16,0,10.022911834716798
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,96,8,128,1,float16,float16,0,10.625484466552734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,96,4,128,1,float16,float16,0,2.317414474487305
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,96,2,128,1,float16,float16,0,2.2645248413085937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,96,1,128,1,float16,float16,0,1.0979328155517578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,96,8,128,1,float16,float16,0,2.422425651550293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,96,96,128,1,float16,float16,0,2.610380744934082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,96,4,128,1,float16,float16,0,1.1719679832458496
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,96,2,128,1,float16,float16,0,1.1094528198242188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,96,8,128,1,float16,float16,0,1.2024831771850586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,96,1,128,1,float16,float16,0,0.5745632171630859
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,96,2,128,1,float16,float16,0,0.5861375808715821
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,96,4,128,1,float16,float16,0,0.6033919811248779
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,96,96,128,1,float16,float16,0,1.2613632202148437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,96,8,128,1,float16,float16,0,0.621721601486206
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,96,1,128,1,float16,float16,0,0.3015680074691772
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,96,2,128,1,float16,float16,0,0.3010560035705566
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,96,96,128,1,float16,float16,0,0.6273536205291748
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,96,4,128,1,float16,float16,0,0.3166656017303467
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,96,96,128,1,float16,float16,0,0.32911360263824463
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,96,8,128,1,float16,float16,0,0.33080320358276366
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,96,1,128,1,float16,float16,0,0.1690111994743347
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,96,2,128,1,float16,float16,0,0.1655295968055725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,96,4,128,1,float16,float16,0,0.16547839641571044
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,96,8,128,1,float16,float16,0,0.17182719707489014
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,96,1,128,1,float16,float16,0,12.207564544677734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,96,1,128,1,float16,float16,0,6.025823974609375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,96,2,128,1,float16,float16,0,6.195916748046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,96,4,128,1,float16,float16,0,6.43056640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,96,2,128,1,float16,float16,0,12.856416320800781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,96,4,128,1,float16,float16,0,13.166181945800782
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,96,8,128,1,float16,float16,0,13.664512634277344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,96,96,128,1,float16,float16,0,18.880921936035158
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,96,2,128,1,float16,float16,0,2.934681510925293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,96,1,128,1,float16,float16,0,2.8327423095703126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,96,8,128,1,float16,float16,0,6.5110527038574215
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,96,4,128,1,float16,float16,0,2.9667791366577148
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,96,8,128,1,float16,float16,0,3.209369659423828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,96,96,128,1,float16,float16,0,9.536409759521485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,96,96,128,1,float16,float16,0,4.989798355102539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,96,1,128,1,float16,float16,0,1.3838335990905761
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,96,2,128,1,float16,float16,0,1.4353407859802245
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,96,4,128,1,float16,float16,0,1.4716927528381347
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,96,1,128,1,float16,float16,0,0.7185408115386963
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,96,8,128,1,float16,float16,0,1.5692288398742675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,96,96,128,1,float16,float16,0,2.4222719192504885
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,96,2,128,1,float16,float16,0,0.7434239864349366
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,96,4,128,1,float16,float16,0,0.7506432056427002
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,96,8,128,1,float16,float16,0,0.7808000087738037
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,96,1,128,1,float16,float16,0,0.3808255910873413
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,96,96,128,1,float16,float16,0,1.213644790649414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,96,2,128,1,float16,float16,0,0.39152638912200927
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,96,96,128,1,float16,float16,0,0.6015999794006348
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,96,1,128,1,float16,float16,0,0.20003840923309327
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,96,4,128,1,float16,float16,0,0.39485440254211424
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,96,8,128,1,float16,float16,0,0.4132863998413086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,96,2,128,1,float16,float16,0,0.20234239101409912
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,96,4,128,1,float16,float16,0,0.2141184091567993
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,96,8,128,1,float16,float16,0,0.22799360752105713
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,96,96,128,1,float16,float16,0,0.2647552013397217
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,96,1,128,1,float16,float16,0,0.10480639934539795
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,96,2,128,1,float16,float16,0,0.10511360168457032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,96,4,128,1,float16,float16,0,0.10675040483474732
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,96,96,128,1,float16,float16,0,0.09635519981384277
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,96,8,128,1,float16,float16,0,0.10910719633102417
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,96,1,128,1,float16,float16,0,0.06266400218009949
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,96,2,128,1,float16,float16,0,0.06343680024147033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,96,4,128,1,float16,float16,0,0.06369280219078063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,96,8,128,1,float16,float16,0,0.0686079978942871
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,96,1,128,1,float16,float16,0,5.052979278564453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,96,2,128,1,float16,float16,0,5.640806579589844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,96,4,128,1,float16,float16,0,5.798400115966797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,96,1,128,1,float16,float16,0,2.4934911727905273
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,96,2,128,1,float16,float16,0,2.8935680389404297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,96,4,128,1,float16,float16,0,3.0601728439331053
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,96,8,128,1,float16,float16,0,6.434355163574219
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,96,96,128,1,float16,float16,0,9.413222503662109
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,96,8,128,1,float16,float16,0,3.322982406616211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,96,1,128,1,float16,float16,0,1.3649408340454101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,96,2,128,1,float16,float16,0,1.4505935668945313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,96,4,128,1,float16,float16,0,1.4651391983032227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,96,8,128,1,float16,float16,0,1.6276479721069337
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,96,96,128,1,float16,float16,0,4.8330238342285154
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,96,96,128,1,float16,float16,0,2.4946144104003904
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,96,1,128,1,float16,float16,0,0.6804480075836181
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,96,2,128,1,float16,float16,0,0.7110144138336182
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,96,8,128,1,float16,float16,0,0.7929855823516846
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,96,96,128,1,float16,float16,0,1.2095999717712402
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,96,4,128,1,float16,float16,0,0.7110144138336182
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,96,2,128,1,float16,float16,0,0.35189759731292725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,96,1,128,1,float16,float16,0,0.34344959259033203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,96,4,128,1,float16,float16,0,0.36817920207977295
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,96,8,128,1,float16,float16,0,0.374726390838623
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,96,96,128,1,float16,float16,0,0.5945856094360351
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,96,1,128,1,float16,float16,0,0.14510079622268676
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,96,2,128,1,float16,float16,0,0.15877120494842528
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,96,96,128,1,float16,float16,0,0.24883038997650148
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,96,8,128,1,float16,float16,0,0.17756160497665405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,96,4,128,1,float16,float16,0,0.1657855987548828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,96,2,128,1,float16,float16,0,0.07767040133476258
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,96,1,128,1,float16,float16,0,0.07475200295448303
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,96,4,128,1,float16,float16,0,0.07623680233955384
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,96,8,128,1,float16,float16,0,0.07920640110969543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,96,96,128,1,float16,float16,0,0.07848960161209106
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,96,1,128,1,float16,float16,0,0.045465600490570066
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,96,2,128,1,float16,float16,0,0.048076799511909483
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,96,8,128,1,float16,float16,0,0.04782080054283142
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,96,4,128,1,float16,float16,0,0.04761599898338318
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,96,96,128,1,float16,float16,0,0.04782080054283142
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,96,2,128,1,float16,float16,0,0.033228799700737
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,96,1,128,1,float16,float16,0,0.03292160034179688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,96,4,128,1,float16,float16,0,0.03328000009059906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,96,8,128,1,float16,float16,0,0.03348479866981506
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,96,1,128,1,float16,float16,0,2.6465280532836912
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,96,4,128,1,float16,float16,0,2.954035186767578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,96,2,128,1,float16,float16,0,1.4049792289733887
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,96,1,128,1,float16,float16,0,1.3934080123901367
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,96,2,128,1,float16,float16,0,2.7215871810913086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,96,8,128,1,float16,float16,0,3.294464111328125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,96,96,128,1,float16,float16,0,4.705279922485351
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,96,4,128,1,float16,float16,0,1.4223360061645507
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,96,8,128,1,float16,float16,0,1.590937614440918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,96,1,128,1,float16,float16,0,0.6962687969207764
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,96,2,128,1,float16,float16,0,0.7022079944610595
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,96,96,128,1,float16,float16,0,2.4194047927856444
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,96,8,128,1,float16,float16,0,0.8564736366271972
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,96,4,128,1,float16,float16,0,0.7652863979339599
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,96,96,128,1,float16,float16,0,1.3105152130126954
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,96,1,128,1,float16,float16,0,0.34652159214019773
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,96,2,128,1,float16,float16,0,0.3527679920196533
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,96,4,128,1,float16,float16,0,0.34012160301208494
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,96,8,128,1,float16,float16,0,0.3736576080322266
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,96,96,128,1,float16,float16,0,0.6504960060119629
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,96,4,128,1,float16,float16,0,0.13987840414047242
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,96,1,128,1,float16,float16,0,0.11412479877471923
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,96,8,128,1,float16,float16,0,0.16624640226364135
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,96,2,128,1,float16,float16,0,0.12021759748458863
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,96,1,128,1,float16,float16,0,0.06108160018920898
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,96,96,128,1,float16,float16,0,0.2461184024810791
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,96,2,128,1,float16,float16,0,0.06062080264091492
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,96,8,128,1,float16,float16,0,0.06574079990386963
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,96,4,128,1,float16,float16,0,0.06297600269317627
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,96,1,128,1,float16,float16,0,0.038852798938751223
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,96,96,128,1,float16,float16,0,0.07055360078811646
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,96,2,128,1,float16,float16,0,0.03865599930286408
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,96,4,128,1,float16,float16,0,0.039526399970054624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,96,8,128,1,float16,float16,0,0.040345600247383116
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,96,1,128,1,float16,float16,0,0.029183998703956604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,96,96,128,1,float16,float16,0,0.04433920085430145
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,96,4,128,1,float16,float16,0,0.029440000653266907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,96,2,128,1,float16,float16,0,0.029388800263404846
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,96,8,128,1,float16,float16,0,0.029388800263404846
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,96,1,128,1,float16,float16,0,0.020428800582885744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,96,96,128,1,float16,float16,0,0.027286401391029357
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,96,2,128,1,float16,float16,0,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,96,4,128,1,float16,float16,0,0.020428800582885744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,96,8,128,1,float16,float16,0,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,96,1,128,1,float16,float16,0,1.2803071975708007
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,96,2,128,1,float16,float16,0,1.319219207763672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,96,4,128,1,float16,float16,0,1.3859840393066407
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,96,8,128,1,float16,float16,0,1.5459327697753906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,96,2,128,1,float16,float16,0,0.6548992156982422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,96,1,128,1,float16,float16,0,0.6762495994567871
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,96,4,128,1,float16,float16,0,0.6915071964263916
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,96,96,128,1,float16,float16,0,2.3412736892700194
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,96,8,128,1,float16,float16,0,0.7930367946624756
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,96,96,128,1,float16,float16,0,1.1905967712402343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,96,1,128,1,float16,float16,0,0.3466752052307129
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,96,2,128,1,float16,float16,0,0.3490303993225098
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,96,8,128,1,float16,float16,0,0.3732480049133301
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,96,4,128,1,float16,float16,0,0.34892799854278567
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,96,96,128,1,float16,float16,0,0.5849599838256836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,96,2,128,1,float16,float16,0,0.12108800411224366
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,96,1,128,1,float16,float16,0,0.10675200223922729
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,96,4,128,1,float16,float16,0,0.14673919677734376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,96,96,128,1,float16,float16,0,0.24780800342559814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,96,8,128,1,float16,float16,0,0.17018879652023317
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,96,2,128,1,float16,float16,0,0.059699201583862306
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,96,1,128,1,float16,float16,0,0.05872640013694763
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,96,4,128,1,float16,float16,0,0.060262399911880496
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,96,8,128,1,float16,float16,0,0.06359040141105651
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,96,1,128,1,float16,float16,0,0.037529599666595456
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,96,96,128,1,float16,float16,0,0.08698559999465942
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,96,2,128,1,float16,float16,0,0.03809280097484589
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,96,8,128,1,float16,float16,0,0.040191999077796935
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,96,4,128,1,float16,float16,0,0.037887999415397645
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,96,96,128,1,float16,float16,0,0.04935680031776428
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,96,1,128,1,float16,float16,0,0.02815999984741211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,96,2,128,1,float16,float16,0,0.028518399596214293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,96,4,128,1,float16,float16,0,0.028672000765800475
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,96,96,128,1,float16,float16,0,0.03333120048046112
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,96,8,128,1,float16,float16,0,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,96,1,128,1,float16,float16,0,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,96,2,128,1,float16,float16,0,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,96,4,128,1,float16,float16,0,0.026675200462341307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,96,8,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,96,1,128,1,float16,float16,0,0.01858399957418442
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,96,96,128,1,float16,float16,0,0.021503999829292297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,96,2,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,96,8,128,1,float16,float16,0,0.019097599387168884
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,96,4,128,1,float16,float16,0,0.01863040030002594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,96,1,128,1,float16,float16,0,0.6437376022338868
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,96,2,128,1,float16,float16,0,0.6548992156982422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,96,4,128,1,float16,float16,0,0.6830063819885254
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,96,8,128,1,float16,float16,0,0.7460815906524658
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,96,96,128,1,float16,float16,0,1.1819007873535157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,96,1,128,1,float16,float16,0,0.3267584085464478
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,96,2,128,1,float16,float16,0,0.3294719934463501
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,96,4,128,1,float16,float16,0,0.34145278930664064
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,96,8,128,1,float16,float16,0,0.3695616006851196
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,96,1,128,1,float16,float16,0,0.12098560333251954
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,96,96,128,1,float16,float16,0,0.5833216190338135
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,96,4,128,1,float16,float16,0,0.16358400583267213
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,96,2,128,1,float16,float16,0,0.12139519453048705
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,96,8,128,1,float16,float16,0,0.1761791944503784
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,96,96,128,1,float16,float16,0,0.2995136022567749
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,96,1,128,1,float16,float16,0,0.059033602476119995
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,96,2,128,1,float16,float16,0,0.058316802978515624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,96,4,128,1,float16,float16,0,0.06000639796257019
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,96,1,128,1,float16,float16,0,0.037273600697517395
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,96,8,128,1,float16,float16,0,0.06236159801483154
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,96,2,128,1,float16,float16,0,0.037785598635673524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,96,96,128,1,float16,float16,0,0.12308160066604615
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,96,4,128,1,float16,float16,0,0.037990400195121767
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,96,8,128,1,float16,float16,0,0.03865599930286408
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,96,1,128,1,float16,float16,0,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,96,4,128,1,float16,float16,0,0.028415998816490172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,96,8,128,1,float16,float16,0,0.028620800375938414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,96,2,128,1,float16,float16,0,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,96,96,128,1,float16,float16,0,0.03993600010871887
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,96,96,128,1,float16,float16,0,0.06880319714546204
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,96,1,128,1,float16,float16,0,0.019865599274635316
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,96,4,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,96,2,128,1,float16,float16,0,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,96,8,128,1,float16,float16,0,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,96,96,128,1,float16,float16,0,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,96,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,96,2,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,96,8,128,1,float16,float16,0,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,96,4,128,1,float16,float16,0,0.01857919991016388
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,96,96,128,1,float16,float16,0,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,96,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,96,8,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,96,2,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,96,4,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,96,1,128,1,float16,float16,0,0.32741599082946776
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,96,2,128,1,float16,float16,0,0.32875039577484133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,96,4,128,1,float16,float16,0,0.34037759304046633
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,96,8,128,1,float16,float16,0,0.3632128000259399
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,96,1,128,1,float16,float16,0,0.11156480312347412
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,96,2,128,1,float16,float16,0,0.11422719955444335
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,96,4,128,1,float16,float16,0,0.14028799533843994
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,96,96,128,1,float16,float16,0,0.8345088005065918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,96,8,128,1,float16,float16,0,0.18037760257720947
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,96,96,128,1,float16,float16,0,0.43438081741333007
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,96,1,128,1,float16,float16,0,0.05867519974708557
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,96,2,128,1,float16,float16,0,0.05980160236358643
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,96,4,128,1,float16,float16,0,0.061440002918243405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,96,8,128,1,float16,float16,0,0.07551519870758057
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,96,4,128,1,float16,float16,0,0.03829759955406189
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,96,96,128,1,float16,float16,0,0.19281920194625854
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,96,2,128,1,float16,float16,0,0.037836799025535585
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,96,1,128,1,float16,float16,0,0.03804160058498383
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,96,1,128,1,float16,float16,0,0.028262400627136232
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,96,8,128,1,float16,float16,0,0.04469760060310364
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,96,96,128,1,float16,float16,0,0.10275839567184449
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,96,2,128,1,float16,float16,0,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,96,8,128,1,float16,float16,0,0.028364801406860353
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,96,96,128,1,float16,float16,0,0.060108798742294314
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,96,4,128,1,float16,float16,0,0.02815999984741211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,96,1,128,1,float16,float16,0,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,96,2,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,96,4,128,1,float16,float16,0,0.01966080069541931
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,96,96,128,1,float16,float16,0,0.034764799475669864
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,96,8,128,1,float16,float16,0,0.020019200444221497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,96,2,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,96,1,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,96,8,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,96,4,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,96,96,128,1,float16,float16,0,0.026111999154090883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,96,1,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,96,2,128,1,float16,float16,0,0.018223999440670012
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,96,8,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,96,96,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,96,4,128,1,float16,float16,0,0.018379199504852294
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,96,1,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,96,2,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,96,4,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,96,8,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,64,1,128,1,float16,float16,0,79.86739501953124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,64,2,128,1,float16,float16,0,79.473046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,64,4,128,1,float16,float16,0,78.953466796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,64,64,128,1,float16,float16,0,87.73124389648437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,64,1,128,1,float16,float16,0,39.072564697265626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,64,64,128,1,float16,float16,0,42.689639282226565
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,64,2,128,1,float16,float16,0,38.90867309570312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,64,1,128,1,float16,float16,0,160.5297607421875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,64,2,128,1,float16,float16,0,161.18768310546875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,64,4,128,1,float16,float16,0,162.286083984375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,64,8,128,1,float16,float16,0,162.25673828125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,64,64,128,1,float16,float16,0,21.302937316894532
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,64,8,128,1,float16,float16,0,81.02753295898438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,64,4,128,1,float16,float16,0,39.71558532714844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,64,8,128,1,float16,float16,0,39.2943115234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,64,1,128,1,float16,float16,0,19.323443603515624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,64,2,128,1,float16,float16,0,19.476019287109374
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,64,4,128,1,float16,float16,0,19.578623962402343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,64,8,128,1,float16,float16,0,19.3818115234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,64,1,128,1,float16,float16,0,44.948019409179686
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,64,2,128,1,float16,float16,0,43.89222412109375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,64,64,128,1,float16,float16,0,51.11178283691406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,64,4,128,1,float16,float16,0,44.58316650390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,64,1,128,1,float16,float16,0,90.53245239257812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,64,2,128,1,float16,float16,0,89.85149536132812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,64,1,128,1,float16,float16,0,21.946775817871092
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,64,64,128,1,float16,float16,0,25.24385223388672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,64,2,128,1,float16,float16,0,21.697637939453124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,64,4,128,1,float16,float16,0,93.402880859375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,64,8,128,1,float16,float16,0,92.45225219726562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,64,4,128,1,float16,float16,0,22.41802215576172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,64,8,128,1,float16,float16,0,45.203662109375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,64,1,128,1,float16,float16,0,10.950707244873048
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,64,64,128,1,float16,float16,0,12.339046478271484
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,64,2,128,1,float16,float16,0,10.661785888671876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,64,8,128,1,float16,float16,0,22.348953247070312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,64,4,128,1,float16,float16,0,11.018291473388672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,64,8,128,1,float16,float16,0,11.031752014160157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,64,1,128,1,float16,float16,0,30.438400268554688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,64,2,128,1,float16,float16,0,31.294924926757812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,64,64,128,1,float16,float16,0,35.70426940917969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,64,4,128,1,float16,float16,0,30.985165405273438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,64,1,128,1,float16,float16,0,63.15934448242187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,64,2,128,1,float16,float16,0,63.54119873046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,64,1,128,1,float16,float16,0,14.877850341796876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,64,2,128,1,float16,float16,0,15.14598388671875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,64,4,128,1,float16,float16,0,64.3504638671875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,64,64,128,1,float16,float16,0,17.932083129882812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,64,8,128,1,float16,float16,0,66.09873046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,64,8,128,1,float16,float16,0,31.472076416015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,64,1,128,1,float16,float16,0,7.21233901977539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,64,2,128,1,float16,float16,0,7.350220489501953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,64,64,128,1,float16,float16,0,8.529714965820313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,64,4,128,1,float16,float16,0,7.460249328613282
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,64,8,128,1,float16,float16,0,7.576985931396484
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,64,4,128,1,float16,float16,0,15.625010681152343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,64,8,128,1,float16,float16,0,15.569046020507812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,64,1,128,1,float16,float16,0,40.161688232421874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,64,2,128,1,float16,float16,0,40.683154296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,64,4,128,1,float16,float16,0,40.94182434082031
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,64,64,128,1,float16,float16,0,49.4835205078125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,64,1,128,1,float16,float16,0,20.00122833251953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,64,1,128,1,float16,float16,0,81.7006591796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,64,64,128,1,float16,float16,0,24.012432861328126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,64,2,128,1,float16,float16,0,82.8273681640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,64,2,128,1,float16,float16,0,20.580812072753908
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,64,4,128,1,float16,float16,0,85.62913208007812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,64,8,128,1,float16,float16,0,84.77577514648438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,64,8,128,1,float16,float16,0,41.834597778320315
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,64,1,128,1,float16,float16,0,9.981542205810547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,64,64,128,1,float16,float16,0,11.861555480957032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,64,2,128,1,float16,float16,0,9.693030548095702
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,64,4,128,1,float16,float16,0,10.236466979980468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,64,8,128,1,float16,float16,0,10.093977355957032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,64,4,128,1,float16,float16,0,20.083404541015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,64,8,128,1,float16,float16,0,20.37452087402344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,64,64,128,1,float16,float16,0,5.807302474975586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,64,1,128,1,float16,float16,0,4.54021110534668
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,64,2,128,1,float16,float16,0,4.781875228881836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,64,4,128,1,float16,float16,0,4.842745590209961
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,64,8,128,1,float16,float16,0,4.805836868286133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,64,1,128,1,float16,float16,0,22.981529235839844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,64,2,128,1,float16,float16,0,23.155506896972657
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,64,4,128,1,float16,float16,0,23.351654052734375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,64,64,128,1,float16,float16,0,29.028402709960936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,64,1,128,1,float16,float16,0,47.26942749023438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,64,2,128,1,float16,float16,0,46.36021728515625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,64,4,128,1,float16,float16,0,47.229287719726564
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,64,8,128,1,float16,float16,0,49.05369567871094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,64,64,128,1,float16,float16,0,14.23426513671875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,64,1,128,1,float16,float16,0,11.347551727294922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,64,2,128,1,float16,float16,0,11.635858917236328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,64,4,128,1,float16,float16,0,11.712665557861328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,64,8,128,1,float16,float16,0,24.231782531738283
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,64,64,128,1,float16,float16,0,6.848818969726563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,64,1,128,1,float16,float16,0,5.712998580932617
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,64,8,128,1,float16,float16,0,11.766425323486327
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,64,2,128,1,float16,float16,0,5.214566421508789
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,64,4,128,1,float16,float16,0,5.237452697753906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,64,8,128,1,float16,float16,0,5.70332145690918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,64,64,128,1,float16,float16,0,3.38155517578125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,64,1,128,1,float16,float16,0,2.6779136657714844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,64,4,128,1,float16,float16,0,2.6370512008666993
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,64,2,128,1,float16,float16,0,2.6406911849975585
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,64,8,128,1,float16,float16,0,2.701055908203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,64,2,128,1,float16,float16,0,21.128192138671874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,64,1,128,1,float16,float16,0,21.107302856445312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,64,4,128,1,float16,float16,0,22.264933776855468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,64,64,128,1,float16,float16,0,28.54727783203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,64,1,128,1,float16,float16,0,42.63291015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,64,2,128,1,float16,float16,0,43.412271118164064
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,64,4,128,1,float16,float16,0,43.83006591796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,64,8,128,1,float16,float16,0,46.01185302734375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,64,1,128,1,float16,float16,0,10.685183715820312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,64,64,128,1,float16,float16,0,13.828504943847657
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,64,2,128,1,float16,float16,0,10.746623992919922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,64,64,128,1,float16,float16,0,6.735308837890625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,64,4,128,1,float16,float16,0,10.825472259521485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,64,8,128,1,float16,float16,0,22.287922668457032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,64,8,128,1,float16,float16,0,11.1710205078125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,64,1,128,1,float16,float16,0,5.049190521240234
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,64,2,128,1,float16,float16,0,5.0643455505371096
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,64,8,128,1,float16,float16,0,5.128755187988281
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,64,4,128,1,float16,float16,0,4.938854217529297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,64,1,128,1,float16,float16,0,2.410086441040039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,64,64,128,1,float16,float16,0,3.3968639373779297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,64,2,128,1,float16,float16,0,2.370355224609375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,64,4,128,1,float16,float16,0,2.443212890625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,64,8,128,1,float16,float16,0,2.4824832916259765
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,64,64,128,1,float16,float16,0,1.7049087524414062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,64,1,128,1,float16,float16,0,1.2147199630737304
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,64,2,128,1,float16,float16,0,1.1931648254394531
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,64,4,128,1,float16,float16,0,1.2298239707946776
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,64,8,128,1,float16,float16,0,1.2695039749145507
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,64,1,128,1,float16,float16,0,12.170342254638673
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,64,2,128,1,float16,float16,0,11.886334228515626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,64,4,128,1,float16,float16,0,12.336640167236329
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,64,1,128,1,float16,float16,0,24.520755004882812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,64,64,128,1,float16,float16,0,17.410401916503908
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,64,2,128,1,float16,float16,0,25.09870147705078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,64,4,128,1,float16,float16,0,25.8050048828125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,64,8,128,1,float16,float16,0,26.524417114257812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,64,1,128,1,float16,float16,0,5.777612686157227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,64,2,128,1,float16,float16,0,6.040678405761719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,64,4,128,1,float16,float16,0,5.976422500610352
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,64,64,128,1,float16,float16,0,8.54625244140625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,64,8,128,1,float16,float16,0,6.4323585510253904
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,64,64,128,1,float16,float16,0,4.192102432250977
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,64,8,128,1,float16,float16,0,12.9164794921875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,64,1,128,1,float16,float16,0,2.8719615936279297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,64,64,128,1,float16,float16,0,2.1354496002197267
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,64,8,128,1,float16,float16,0,2.9637632369995117
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,64,2,128,1,float16,float16,0,2.7066368103027343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,64,1,128,1,float16,float16,0,1.3776384353637696
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,64,4,128,1,float16,float16,0,2.7822591781616213
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,64,2,128,1,float16,float16,0,1.3797887802124023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,64,8,128,1,float16,float16,0,1.4728192329406737
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,64,4,128,1,float16,float16,0,1.4269439697265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,64,64,128,1,float16,float16,0,1.0627072334289551
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,64,2,128,1,float16,float16,0,0.7197184085845947
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,64,4,128,1,float16,float16,0,0.7395247936248779
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,64,1,128,1,float16,float16,0,0.7053823947906495
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,64,8,128,1,float16,float16,0,0.7517695903778077
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,64,1,128,1,float16,float16,0,11.518720245361328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,64,2,128,1,float16,float16,0,11.622700500488282
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,64,1,128,1,float16,float16,0,23.02054443359375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,64,4,128,1,float16,float16,0,12.10982437133789
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,64,2,128,1,float16,float16,0,23.74671325683594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,64,4,128,1,float16,float16,0,24.283544921875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,64,64,128,1,float16,float16,0,17.951692199707033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,64,8,128,1,float16,float16,0,25.20704040527344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,64,1,128,1,float16,float16,0,5.482598495483399
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,64,2,128,1,float16,float16,0,5.766656112670899
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,64,4,128,1,float16,float16,0,5.737728118896484
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,64,8,128,1,float16,float16,0,5.908633422851563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,64,64,128,1,float16,float16,0,4.428646469116211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,64,64,128,1,float16,float16,0,8.845665740966798
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,64,8,128,1,float16,float16,0,12.342066955566406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,64,1,128,1,float16,float16,0,2.6283519744873045
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,64,2,128,1,float16,float16,0,2.616217613220215
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,64,8,128,1,float16,float16,0,2.8481023788452147
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,64,4,128,1,float16,float16,0,2.7936767578125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,64,64,128,1,float16,float16,0,2.200831985473633
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,64,2,128,1,float16,float16,0,1.2942336082458497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,64,1,128,1,float16,float16,0,1.28471040725708
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,64,4,128,1,float16,float16,0,1.35982084274292
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,64,8,128,1,float16,float16,0,1.4056447982788085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,64,64,128,1,float16,float16,0,1.1167743682861329
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,64,1,128,1,float16,float16,0,0.6789631843566895
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,64,2,128,1,float16,float16,0,0.6850560188293457
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,64,4,128,1,float16,float16,0,0.7040512084960937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,64,1,128,1,float16,float16,0,0.3460095882415771
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,64,8,128,1,float16,float16,0,0.714086389541626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,64,64,128,1,float16,float16,0,0.5684735774993896
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,64,2,128,1,float16,float16,0,0.35256319046020507
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,64,4,128,1,float16,float16,0,0.35394558906555174
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,64,8,128,1,float16,float16,0,0.3614720106124878
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,64,1,128,1,float16,float16,0,6.599628448486328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,64,1,128,1,float16,float16,0,13.746585083007812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,64,2,128,1,float16,float16,0,13.912576293945312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,64,2,128,1,float16,float16,0,6.543769836425781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,64,4,128,1,float16,float16,0,14.365644836425782
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,64,64,128,1,float16,float16,0,11.463219451904298
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,64,8,128,1,float16,float16,0,14.753330993652344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,64,4,128,1,float16,float16,0,6.882150268554687
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,64,1,128,1,float16,float16,0,3.1559680938720702
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,64,2,128,1,float16,float16,0,3.2031742095947267
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,64,4,128,1,float16,float16,0,3.3427455902099608
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,64,64,128,1,float16,float16,0,5.767679977416992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,64,1,128,1,float16,float16,0,1.5152640342712402
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,64,64,128,1,float16,float16,0,2.881177520751953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,64,8,128,1,float16,float16,0,7.3725440979003904
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,64,8,128,1,float16,float16,0,3.443251037597656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,64,4,128,1,float16,float16,0,1.6110591888427734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,64,2,128,1,float16,float16,0,1.558886432647705
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,64,2,128,1,float16,float16,0,0.8019968032836914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,64,64,128,1,float16,float16,0,1.489151954650879
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,64,1,128,1,float16,float16,0,0.7947264194488526
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,64,8,128,1,float16,float16,0,1.67956485748291
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,64,4,128,1,float16,float16,0,0.8023551940917969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,64,8,128,1,float16,float16,0,0.8409088134765625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,64,64,128,1,float16,float16,0,0.7288320064544678
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,64,1,128,1,float16,float16,0,0.40529918670654297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,64,4,128,1,float16,float16,0,0.425932788848877
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,64,8,128,1,float16,float16,0,0.44226560592651365
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,64,2,128,1,float16,float16,0,0.4076543807983398
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,64,1,128,1,float16,float16,0,0.21867520809173585
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,64,64,128,1,float16,float16,0,0.3758080005645752
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,64,2,128,1,float16,float16,0,0.22036480903625488
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,64,4,128,1,float16,float16,0,0.22661120891571046
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,64,8,128,1,float16,float16,0,0.2298367977142334
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,64,1,128,1,float16,float16,0,6.352435302734375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,64,2,128,1,float16,float16,0,13.585664367675781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,64,1,128,1,float16,float16,0,13.216102600097656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,64,4,128,1,float16,float16,0,13.7438720703125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,64,2,128,1,float16,float16,0,6.708326721191407
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,64,64,128,1,float16,float16,0,12.847564697265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,64,8,128,1,float16,float16,0,14.897817993164063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,64,4,128,1,float16,float16,0,6.664806365966797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,64,1,128,1,float16,float16,0,3.067801666259766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,64,1,128,1,float16,float16,0,1.4983087539672852
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,64,2,128,1,float16,float16,0,3.0000640869140627
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,64,4,128,1,float16,float16,0,3.279462432861328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,64,8,128,1,float16,float16,0,3.484569549560547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,64,64,128,1,float16,float16,0,3.51544303894043
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,64,8,128,1,float16,float16,0,6.952191925048828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,64,64,128,1,float16,float16,0,6.553804779052735
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,64,2,128,1,float16,float16,0,1.4965248107910156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,64,1,128,1,float16,float16,0,0.7831039905548096
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,64,8,128,1,float16,float16,0,1.6901119232177735
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,64,4,128,1,float16,float16,0,1.5401984214782716
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,64,64,128,1,float16,float16,0,1.6424959182739258
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,64,2,128,1,float16,float16,0,0.7734784126281739
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,64,4,128,1,float16,float16,0,0.7888383865356445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,64,8,128,1,float16,float16,0,0.8765952110290527
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,64,64,128,1,float16,float16,0,0.8359935760498047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,64,4,128,1,float16,float16,0,0.42112002372741697
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,64,1,128,1,float16,float16,0,0.405350399017334
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,64,8,128,1,float16,float16,0,0.43438081741333007
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,64,2,128,1,float16,float16,0,0.40181760787963866
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,64,64,128,1,float16,float16,0,0.4419072151184082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,64,1,128,1,float16,float16,0,0.1977247953414917
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,64,2,128,1,float16,float16,0,0.20136959552764894
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,64,4,128,1,float16,float16,0,0.2110975980758667
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,64,8,128,1,float16,float16,0,0.21135358810424804
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,64,1,128,1,float16,float16,0,0.11776000261306763
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,64,64,128,1,float16,float16,0,0.16783360242843628
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,64,2,128,1,float16,float16,0,0.11719679832458496
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,64,4,128,1,float16,float16,0,0.12349439859390259
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,64,8,128,1,float16,float16,0,0.12492799758911133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,64,1,128,1,float16,float16,0,8.258918762207031
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,64,1,128,1,float16,float16,0,3.9581695556640626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,64,2,128,1,float16,float16,0,4.0024574279785154
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,64,2,128,1,float16,float16,0,8.341196441650391
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,64,4,128,1,float16,float16,0,8.68172836303711
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,64,4,128,1,float16,float16,0,4.172851181030273
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,64,8,128,1,float16,float16,0,9.589555358886718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,64,64,128,1,float16,float16,0,12.522905731201172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,64,1,128,1,float16,float16,0,1.8621440887451173
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,64,2,128,1,float16,float16,0,1.9371519088745117
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,64,4,128,1,float16,float16,0,2.064896011352539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,64,8,128,1,float16,float16,0,4.692070388793946
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,64,8,128,1,float16,float16,0,2.264678382873535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,64,64,128,1,float16,float16,0,6.361753463745117
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,64,64,128,1,float16,float16,0,3.327027130126953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,64,1,128,1,float16,float16,0,0.9498623847961426
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,64,2,128,1,float16,float16,0,0.9584128379821777
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,64,4,128,1,float16,float16,0,1.0156031608581544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,64,8,128,1,float16,float16,0,1.1914752006530762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,64,2,128,1,float16,float16,0,0.506060791015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,64,1,128,1,float16,float16,0,0.49346561431884767
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,64,64,128,1,float16,float16,0,1.6696832656860352
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,64,8,128,1,float16,float16,0,0.5650944232940673
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,64,4,128,1,float16,float16,0,0.5212672233581543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,64,64,128,1,float16,float16,0,0.8174079895019531
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,64,1,128,1,float16,float16,0,0.27130560874938964
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,64,2,128,1,float16,float16,0,0.2783744096755981
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,64,4,128,1,float16,float16,0,0.27939839363098146
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,64,64,128,1,float16,float16,0,0.37908480167388914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,64,8,128,1,float16,float16,0,0.2936320066452026
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,64,2,128,1,float16,float16,0,0.13015040159225463
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,64,1,128,1,float16,float16,0,0.12948479652404785
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,64,4,128,1,float16,float16,0,0.13434879779815673
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,64,8,128,1,float16,float16,0,0.13864959478378297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,64,64,128,1,float16,float16,0,0.12999680042266845
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,64,1,128,1,float16,float16,0,0.07649279832839966
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,64,2,128,1,float16,float16,0,0.07813119888305664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,64,8,128,1,float16,float16,0,0.08325120210647582
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,64,4,128,1,float16,float16,0,0.0809391975402832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,64,64,128,1,float16,float16,0,0.07167199850082398
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,64,1,128,1,float16,float16,0,0.046489599347114566
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,64,2,128,1,float16,float16,0,0.05007359981536865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,64,4,128,1,float16,float16,0,0.052223998308181765
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,64,8,128,1,float16,float16,0,0.052112001180648806
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,64,1,128,1,float16,float16,0,3.641548919677734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,64,2,128,1,float16,float16,0,3.7518337249755858
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,64,4,128,1,float16,float16,0,4.082022476196289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,64,1,128,1,float16,float16,0,1.874995231628418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,64,8,128,1,float16,float16,0,4.679014587402344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,64,2,128,1,float16,float16,0,1.8396671295166016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,64,4,128,1,float16,float16,0,1.9770879745483398
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,64,64,128,1,float16,float16,0,6.259711837768554
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,64,8,128,1,float16,float16,0,2.3820287704467775
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,64,2,128,1,float16,float16,0,0.9297920227050781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,64,64,128,1,float16,float16,0,3.2196609497070314
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,64,1,128,1,float16,float16,0,0.9354751586914063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,64,4,128,1,float16,float16,0,1.0141183853149414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,64,1,128,1,float16,float16,0,0.4688896179199219
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,64,8,128,1,float16,float16,0,1.2520943641662599
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,64,64,128,1,float16,float16,0,1.612388801574707
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,64,2,128,1,float16,float16,0,0.4839935779571533
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,64,4,128,1,float16,float16,0,0.5032896041870117
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,64,64,128,1,float16,float16,0,0.8071680068969727
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,64,8,128,1,float16,float16,0,0.6099455833435059
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,64,1,128,1,float16,float16,0,0.22415359020233155
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,64,2,128,1,float16,float16,0,0.22067201137542725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,64,8,128,1,float16,float16,0,0.25912320613861084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,64,4,128,1,float16,float16,0,0.23915519714355468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,64,64,128,1,float16,float16,0,0.372377610206604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,64,1,128,1,float16,float16,0,0.09405440092086792
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,64,2,128,1,float16,float16,0,0.09277439713478089
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,64,4,128,1,float16,float16,0,0.09722880125045777
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,64,64,128,1,float16,float16,0,0.10152959823608398
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,64,1,128,1,float16,float16,0,0.05719040036201477
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,64,8,128,1,float16,float16,0,0.10096640586853027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,64,2,128,1,float16,float16,0,0.05708799958229065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,64,4,128,1,float16,float16,0,0.05857279896736145
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,64,8,128,1,float16,float16,0,0.061952000856399535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,64,2,128,1,float16,float16,0,0.03829759955406189
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,64,1,128,1,float16,float16,0,0.0377344012260437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,64,64,128,1,float16,float16,0,0.05841919779777527
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,64,8,128,1,float16,float16,0,0.03947519958019256
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,64,4,128,1,float16,float16,0,0.03819519877433777
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,64,64,128,1,float16,float16,0,0.035020801424980166
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,64,1,128,1,float16,float16,0,0.025139200687408447
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,64,2,128,1,float16,float16,0,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,64,8,128,1,float16,float16,0,0.02590720057487488
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,64,4,128,1,float16,float16,0,0.02529279887676239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,64,1,128,1,float16,float16,0,1.7772544860839843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,64,2,128,1,float16,float16,0,1.845248031616211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,64,4,128,1,float16,float16,0,2.045439910888672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,64,8,128,1,float16,float16,0,2.357094383239746
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,64,1,128,1,float16,float16,0,0.9041407585144043
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,64,64,128,1,float16,float16,0,3.1462400436401365
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,64,2,128,1,float16,float16,0,0.9084927558898925
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,64,4,128,1,float16,float16,0,0.9939968109130859
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,64,8,128,1,float16,float16,0,1.1810303688049317
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,64,64,128,1,float16,float16,0,1.6067071914672852
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,64,1,128,1,float16,float16,0,0.47052798271179197
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,64,2,128,1,float16,float16,0,0.44768800735473635
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,64,4,128,1,float16,float16,0,0.5357056140899659
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,64,8,128,1,float16,float16,0,0.6144512176513672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,64,64,128,1,float16,float16,0,0.8052224159240723
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,64,1,128,1,float16,float16,0,0.22236158847808837
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,64,4,128,1,float16,float16,0,0.2332159996032715
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,64,2,128,1,float16,float16,0,0.23453919887542723
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,64,64,128,1,float16,float16,0,0.3686896085739136
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,64,8,128,1,float16,float16,0,0.2573312044143677
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,64,1,128,1,float16,float16,0,0.0745967984199524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,64,4,128,1,float16,float16,0,0.07838400006294251
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,64,2,128,1,float16,float16,0,0.07608320116996765
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,64,8,128,1,float16,float16,0,0.08335360288619995
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,64,1,128,1,float16,float16,0,0.04618239998817444
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,64,64,128,1,float16,float16,0,0.09937919974327088
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,64,4,128,1,float16,float16,0,0.048332801461219786
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,64,2,128,1,float16,float16,0,0.04633280038833618
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,64,64,128,1,float16,float16,0,0.052070397138595584
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,64,8,128,1,float16,float16,0,0.05002239942550659
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,64,1,128,1,float16,float16,0,0.03178719878196716
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,64,2,128,1,float16,float16,0,0.031641599535942075
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,64,4,128,1,float16,float16,0,0.031488001346588135
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,64,8,128,1,float16,float16,0,0.03256320059299469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,64,64,128,1,float16,float16,0,0.032204800844192506
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,64,2,128,1,float16,float16,0,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,64,1,128,1,float16,float16,0,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,64,4,128,1,float16,float16,0,0.021657599508762358
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,64,8,128,1,float16,float16,0,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,64,1,128,1,float16,float16,0,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,64,2,128,1,float16,float16,0,0.01950560063123703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,64,64,128,1,float16,float16,0,0.023603199422359465
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,64,4,128,1,float16,float16,0,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,64,8,128,1,float16,float16,0,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,64,2,128,1,float16,float16,0,0.9042943954467774
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,64,1,128,1,float16,float16,0,0.8624640464782715
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,64,4,128,1,float16,float16,0,1.0012672424316407
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,64,8,128,1,float16,float16,0,1.1384832382202148
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,64,1,128,1,float16,float16,0,0.43571200370788576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,64,64,128,1,float16,float16,0,1.5583744049072266
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,64,2,128,1,float16,float16,0,0.44794559478759766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,64,4,128,1,float16,float16,0,0.48291840553283694
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,64,8,128,1,float16,float16,0,0.5598720073699951
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,64,1,128,1,float16,float16,0,0.22108159065246583
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,64,64,128,1,float16,float16,0,0.7914495944976807
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,64,4,128,1,float16,float16,0,0.22978560924530028
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,64,2,128,1,float16,float16,0,0.22558720111846925
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,64,64,128,1,float16,float16,0,0.3664383888244629
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,64,8,128,1,float16,float16,0,0.2797055959701538
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,64,1,128,1,float16,float16,0,0.07367680072784424
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,64,2,128,1,float16,float16,0,0.07475200295448303
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,64,4,128,1,float16,float16,0,0.07874559760093688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,64,8,128,1,float16,float16,0,0.08181759715080261
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,64,64,128,1,float16,float16,0,0.12072960138320923
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,64,1,128,1,float16,float16,0,0.04459519982337952
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,64,2,128,1,float16,float16,0,0.045824000239372255
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,64,8,128,1,float16,float16,0,0.048691201210021975
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,64,4,128,1,float16,float16,0,0.04628480076789856
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,64,1,128,1,float16,float16,0,0.0302592009305954
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,64,64,128,1,float16,float16,0,0.06251519918441772
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,64,4,128,1,float16,float16,0,0.03051519989967346
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,64,8,128,1,float16,float16,0,0.031641599535942075
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,64,2,128,1,float16,float16,0,0.03031040132045746
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,64,64,128,1,float16,float16,0,0.03701759874820709
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,64,1,128,1,float16,float16,0,0.021555200219154358
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,64,2,128,1,float16,float16,0,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,64,4,128,1,float16,float16,0,0.02160319983959198
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,64,8,128,1,float16,float16,0,0.021555200219154358
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,64,64,128,1,float16,float16,0,0.024320000410079957
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,64,4,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,64,2,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,64,1,128,1,float16,float16,0,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,64,8,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,64,64,128,1,float16,float16,0,0.020121599733829498
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,64,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,64,2,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,64,4,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,64,1,128,1,float16,float16,0,0.43494400978088377
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,64,8,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,64,4,128,1,float16,float16,0,0.47728638648986815
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,64,2,128,1,float16,float16,0,0.44431362152099607
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,64,64,128,1,float16,float16,0,0.7947264194488526
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,64,8,128,1,float16,float16,0,0.5343232154846191
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,64,1,128,1,float16,float16,0,0.21283841133117676
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,64,2,128,1,float16,float16,0,0.21765120029449464
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,64,4,128,1,float16,float16,0,0.23331840038299562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,64,64,128,1,float16,float16,0,0.3974656105041504
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,64,8,128,1,float16,float16,0,0.25712640285491944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,64,1,128,1,float16,float16,0,0.07325760126113892
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,64,2,128,1,float16,float16,0,0.07557119727134705
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,64,4,128,1,float16,float16,0,0.07654399871826172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,64,8,128,1,float16,float16,0,0.0834559977054596
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,64,64,128,1,float16,float16,0,0.16701439619064332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,64,1,128,1,float16,float16,0,0.04561919867992401
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,64,2,128,1,float16,float16,0,0.04567039906978607
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,64,4,128,1,float16,float16,0,0.046387198567390445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,64,8,128,1,float16,float16,0,0.0492031991481781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,64,64,128,1,float16,float16,0,0.08934400081634522
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,64,1,128,1,float16,float16,0,0.030156800150871278
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,64,4,128,1,float16,float16,0,0.03041279911994934
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,64,2,128,1,float16,float16,0,0.029951998591423036
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,64,8,128,1,float16,float16,0,0.031488001346588135
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,64,64,128,1,float16,float16,0,0.05017600059509277
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,64,1,128,1,float16,float16,0,0.02109439969062805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,64,2,128,1,float16,float16,0,0.02109439969062805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,64,4,128,1,float16,float16,0,0.02263039946556091
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,64,8,128,1,float16,float16,0,0.021555200219154358
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,64,1,128,1,float16,float16,0,0.018988800048828126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,64,4,128,1,float16,float16,0,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,64,2,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,64,64,128,1,float16,float16,0,0.029996800422668456
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,64,8,128,1,float16,float16,0,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,64,64,128,1,float16,float16,0,0.021196800470352172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,64,1,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,64,2,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,64,4,128,1,float16,float16,0,0.01852799952030182
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,64,64,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,64,8,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,64,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,64,2,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,64,8,128,1,float16,float16,0,0.018479999899864197
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,64,4,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,64,1,128,1,float16,float16,0,0.21329920291900634
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,64,2,128,1,float16,float16,0,0.22359039783477783
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,64,4,128,1,float16,float16,0,0.23664638996124268
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,64,8,128,1,float16,float16,0,0.26019840240478515
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,64,64,128,1,float16,float16,0,0.5727744102478027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,64,1,128,1,float16,float16,0,0.07342079877853394
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,64,2,128,1,float16,float16,0,0.07623680233955384
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,64,4,128,1,float16,float16,0,0.07690240144729614
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,64,8,128,1,float16,float16,0,0.0828927993774414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,64,64,128,1,float16,float16,0,0.26547200679779054
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,64,1,128,1,float16,float16,0,0.045414400100708005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,64,2,128,1,float16,float16,0,0.045875200629234315
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,64,4,128,1,float16,float16,0,0.04705280065536499
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,64,64,128,1,float16,float16,0,0.13772799968719482
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,64,8,128,1,float16,float16,0,0.04961279928684235
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,64,2,128,1,float16,float16,0,0.030156800150871278
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,64,4,128,1,float16,float16,0,0.03061760067939758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,64,1,128,1,float16,float16,0,0.030156800150871278
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,64,8,128,1,float16,float16,0,0.031948798894882204
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,64,64,128,1,float16,float16,0,0.07680000066757202
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,64,4,128,1,float16,float16,0,0.021196800470352172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,64,1,128,1,float16,float16,0,0.02088800072669983
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,64,2,128,1,float16,float16,0,0.021401600539684297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,64,64,128,1,float16,float16,0,0.04387840032577515
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,64,8,128,1,float16,float16,0,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,64,2,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,64,1,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,64,4,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,64,8,128,1,float16,float16,0,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,64,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,64,64,128,1,float16,float16,0,0.02744320034980774
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,64,4,128,1,float16,float16,0,0.01858399957418442
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,64,2,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,64,8,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,64,64,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,64,1,128,1,float16,float16,0,0.018272000551223754
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,64,2,128,1,float16,float16,0,0.020268799364566804
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,64,4,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,64,8,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,64,1,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,64,2,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,64,64,128,1,float16,float16,0,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,64,4,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,64,8,128,1,float16,float16,0,0.018225599825382233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,48,1,128,1,float16,float16,0,59.712152099609376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,48,2,128,1,float16,float16,0,58.61893310546875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,48,4,128,1,float16,float16,0,61.049798583984376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,48,48,128,1,float16,float16,0,67.28514404296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,48,1,128,1,float16,float16,0,29.277542114257812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,48,48,128,1,float16,float16,0,31.888180541992188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,48,2,128,1,float16,float16,0,29.242266845703124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,48,1,128,1,float16,float16,0,118.96458740234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,48,2,128,1,float16,float16,0,120.8058349609375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,48,4,128,1,float16,float16,0,122.46917724609375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,48,8,128,1,float16,float16,0,123.7365234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,48,48,128,1,float16,float16,0,16.01361846923828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,48,8,128,1,float16,float16,0,61.47379150390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,48,1,128,1,float16,float16,0,14.127615356445313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,48,4,128,1,float16,float16,0,29.700555419921876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,48,8,128,1,float16,float16,0,29.49595947265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,48,2,128,1,float16,float16,0,14.353715515136718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,48,4,128,1,float16,float16,0,14.426829528808593
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,48,8,128,1,float16,float16,0,14.892134094238282
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,48,2,128,1,float16,float16,0,33.22700805664063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,48,1,128,1,float16,float16,0,33.36575622558594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,48,4,128,1,float16,float16,0,33.81119995117187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,48,48,128,1,float16,float16,0,38.599371337890624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,48,1,128,1,float16,float16,0,68.95457153320312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,48,2,128,1,float16,float16,0,68.60006103515624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,48,4,128,1,float16,float16,0,69.03065185546875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,48,1,128,1,float16,float16,0,16.21360626220703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,48,48,128,1,float16,float16,0,18.5997314453125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,48,2,128,1,float16,float16,0,16.865023803710937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,48,8,128,1,float16,float16,0,70.34935913085937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,48,8,128,1,float16,float16,0,34.50245056152344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,48,1,128,1,float16,float16,0,7.9853569030761715
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,48,48,128,1,float16,float16,0,9.086720275878907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,48,2,128,1,float16,float16,0,8.106752014160156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,48,4,128,1,float16,float16,0,16.799026489257812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,48,8,128,1,float16,float16,0,17.01934051513672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,48,4,128,1,float16,float16,0,8.058521270751953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,48,8,128,1,float16,float16,0,8.422345733642578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,48,1,128,1,float16,float16,0,23.533824157714843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,48,2,128,1,float16,float16,0,23.550003051757812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,48,48,128,1,float16,float16,0,27.0255615234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,48,4,128,1,float16,float16,0,23.24024353027344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,48,2,128,1,float16,float16,0,47.174502563476565
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,48,1,128,1,float16,float16,0,47.2363525390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,48,4,128,1,float16,float16,0,47.53177490234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,48,8,128,1,float16,float16,0,49.214822387695314
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,48,48,128,1,float16,float16,0,13.236579895019531
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,48,1,128,1,float16,float16,0,11.4840576171875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,48,2,128,1,float16,float16,0,11.252684783935546
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,48,8,128,1,float16,float16,0,24.022271728515626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,48,48,128,1,float16,float16,0,6.670591735839844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,48,1,128,1,float16,float16,0,5.404825592041016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,48,2,128,1,float16,float16,0,5.687551879882813
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,48,4,128,1,float16,float16,0,5.598412704467774
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,48,8,128,1,float16,float16,0,12.003839874267578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,48,4,128,1,float16,float16,0,11.81839370727539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,48,8,128,1,float16,float16,0,5.549977493286133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,48,1,128,1,float16,float16,0,30.30169677734375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,48,2,128,1,float16,float16,0,30.139340209960938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,48,4,128,1,float16,float16,0,30.740786743164062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,48,48,128,1,float16,float16,0,36.33966064453125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,48,1,128,1,float16,float16,0,62.10308837890625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,48,2,128,1,float16,float16,0,61.7923583984375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,48,1,128,1,float16,float16,0,14.813746643066406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,48,48,128,1,float16,float16,0,17.629798889160156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,48,2,128,1,float16,float16,0,15.296049499511719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,48,4,128,1,float16,float16,0,64.72473754882813
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,48,8,128,1,float16,float16,0,65.6142822265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,48,8,128,1,float16,float16,0,31.765811157226562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,48,4,128,1,float16,float16,0,15.44847412109375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,48,2,128,1,float16,float16,0,7.190425872802734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,48,1,128,1,float16,float16,0,7.199334716796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,48,48,128,1,float16,float16,0,8.554137420654296
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,48,4,128,1,float16,float16,0,7.811174774169922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,48,8,128,1,float16,float16,0,7.650918579101562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,48,8,128,1,float16,float16,0,15.453645324707031
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,48,48,128,1,float16,float16,0,4.336383819580078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,48,2,128,1,float16,float16,0,3.5037185668945314
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,48,1,128,1,float16,float16,0,3.5152896881103515
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,48,4,128,1,float16,float16,0,3.510015869140625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,48,8,128,1,float16,float16,0,3.447804641723633
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,48,1,128,1,float16,float16,0,16.920268249511718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,48,2,128,1,float16,float16,0,16.967730712890624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,48,4,128,1,float16,float16,0,17.447372436523438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,48,48,128,1,float16,float16,0,21.707621765136718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,48,1,128,1,float16,float16,0,35.065493774414065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,48,2,128,1,float16,float16,0,34.78681640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,48,4,128,1,float16,float16,0,35.7528564453125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,48,8,128,1,float16,float16,0,37.23709411621094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,48,1,128,1,float16,float16,0,8.230860900878906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,48,48,128,1,float16,float16,0,10.249523162841797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,48,2,128,1,float16,float16,0,8.382924652099609
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,48,4,128,1,float16,float16,0,8.584243011474609
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,48,48,128,1,float16,float16,0,5.149951934814453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,48,8,128,1,float16,float16,0,8.975769805908204
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,48,8,128,1,float16,float16,0,18.247116088867188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,48,1,128,1,float16,float16,0,3.879167938232422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,48,1,128,1,float16,float16,0,1.9847679138183594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,48,48,128,1,float16,float16,0,2.5753087997436523
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,48,2,128,1,float16,float16,0,4.028364944458008
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,48,2,128,1,float16,float16,0,1.9655168533325196
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,48,4,128,1,float16,float16,0,3.970764923095703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,48,8,128,1,float16,float16,0,4.061644744873047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,48,4,128,1,float16,float16,0,2.051686477661133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,48,8,128,1,float16,float16,0,2.0517887115478515
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,48,2,128,1,float16,float16,0,15.978189086914062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,48,1,128,1,float16,float16,0,15.616819763183594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,48,48,128,1,float16,float16,0,20.787660217285158
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,48,4,128,1,float16,float16,0,16.336741638183593
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,48,1,128,1,float16,float16,0,31.56229248046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,48,2,128,1,float16,float16,0,32.59171752929687
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,48,4,128,1,float16,float16,0,33.14227294921875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,48,8,128,1,float16,float16,0,34.58447265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,48,1,128,1,float16,float16,0,7.785984039306641
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,48,48,128,1,float16,float16,0,10.506649780273438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,48,2,128,1,float16,float16,0,7.830271911621094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,48,8,128,1,float16,float16,0,17.01739501953125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,48,4,128,1,float16,float16,0,8.176895904541016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,48,1,128,1,float16,float16,0,3.7935089111328124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,48,48,128,1,float16,float16,0,5.151385498046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,48,8,128,1,float16,float16,0,8.374221038818359
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,48,2,128,1,float16,float16,0,3.6794864654541017
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,48,4,128,1,float16,float16,0,3.876505661010742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,48,48,128,1,float16,float16,0,2.5511423110961915
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,48,2,128,1,float16,float16,0,1.7934335708618163
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,48,1,128,1,float16,float16,0,1.7864704132080078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,48,8,128,1,float16,float16,0,1.917900848388672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,48,8,128,1,float16,float16,0,3.9018497467041016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,48,4,128,1,float16,float16,0,1.823744010925293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,48,48,128,1,float16,float16,0,1.2717568397521972
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,48,1,128,1,float16,float16,0,0.9238528251647949
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,48,8,128,1,float16,float16,0,0.9828288078308105
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,48,4,128,1,float16,float16,0,0.9542655944824219
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,48,2,128,1,float16,float16,0,0.9390015602111816
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,48,2,128,1,float16,float16,0,18.47352294921875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,48,48,128,1,float16,float16,0,13.152922058105469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,48,1,128,1,float16,float16,0,18.24808349609375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,48,1,128,1,float16,float16,0,8.79994888305664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,48,2,128,1,float16,float16,0,9.216973114013673
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,48,4,128,1,float16,float16,0,19.20880584716797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,48,4,128,1,float16,float16,0,9.522637176513673
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,48,8,128,1,float16,float16,0,20.083506774902343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,48,1,128,1,float16,float16,0,4.132147216796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,48,2,128,1,float16,float16,0,4.330854415893555
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,48,48,128,1,float16,float16,0,3.1755775451660155
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,48,4,128,1,float16,float16,0,4.543999862670899
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,48,48,128,1,float16,float16,0,6.388326263427734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,48,8,128,1,float16,float16,0,4.765030288696289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,48,1,128,1,float16,float16,0,2.1131263732910157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,48,8,128,1,float16,float16,0,9.814527893066407
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,48,4,128,1,float16,float16,0,2.1727231979370116
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,48,2,128,1,float16,float16,0,2.051686477661133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,48,8,128,1,float16,float16,0,2.2383615493774416
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,48,2,128,1,float16,float16,0,1.0752511978149415
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,48,48,128,1,float16,float16,0,1.5636992454528809
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,48,1,128,1,float16,float16,0,1.053388786315918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,48,4,128,1,float16,float16,0,1.090611171722412
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,48,8,128,1,float16,float16,0,1.1582464218139648
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,48,1,128,1,float16,float16,0,0.5636608123779296
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,48,4,128,1,float16,float16,0,0.5443071842193603
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,48,48,128,1,float16,float16,0,0.818943977355957
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,48,2,128,1,float16,float16,0,0.5540351867675781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,48,8,128,1,float16,float16,0,0.5753856182098389
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,48,1,128,1,float16,float16,0,16.966860961914062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,48,1,128,1,float16,float16,0,8.06983642578125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,48,2,128,1,float16,float16,0,17.83756866455078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,48,4,128,1,float16,float16,0,18.347981262207032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,48,2,128,1,float16,float16,0,8.790016174316406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,48,4,128,1,float16,float16,0,9.01339340209961
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,48,48,128,1,float16,float16,0,13.351986694335938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,48,8,128,1,float16,float16,0,19.258624267578124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,48,2,128,1,float16,float16,0,4.035071945190429
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,48,4,128,1,float16,float16,0,4.271513748168945
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,48,1,128,1,float16,float16,0,3.9714305877685545
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,48,8,128,1,float16,float16,0,4.3249153137207035
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,48,48,128,1,float16,float16,0,6.554009246826172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,48,48,128,1,float16,float16,0,3.4115585327148437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,48,8,128,1,float16,float16,0,9.457096099853516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,48,1,128,1,float16,float16,0,1.9691007614135743
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,48,2,128,1,float16,float16,0,1.9350015640258789
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,48,4,128,1,float16,float16,0,2.0342784881591798
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,48,48,128,1,float16,float16,0,1.652275276184082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,48,1,128,1,float16,float16,0,0.9934320449829102
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,48,2,128,1,float16,float16,0,1.0140159606933594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,48,8,128,1,float16,float16,0,2.190131187438965
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,48,4,128,1,float16,float16,0,1.0255871772766114
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,48,8,128,1,float16,float16,0,1.1109375953674316
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,48,48,128,1,float16,float16,0,0.8381391525268554
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,48,1,128,1,float16,float16,0,0.5150207996368408
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,48,8,128,1,float16,float16,0,0.5689343929290771
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,48,2,128,1,float16,float16,0,0.5319680213928223
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,48,4,128,1,float16,float16,0,0.5373951911926269
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,48,48,128,1,float16,float16,0,0.4363264083862305
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,48,1,128,1,float16,float16,0,0.28492801189422606
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,48,2,128,1,float16,float16,0,0.27688961029052733
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,48,4,128,1,float16,float16,0,0.28298239707946776
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,48,8,128,1,float16,float16,0,0.2932735919952393
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,48,1,128,1,float16,float16,0,9.94867172241211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,48,1,128,1,float16,float16,0,4.667955017089843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,48,2,128,1,float16,float16,0,10.378752136230469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,48,4,128,1,float16,float16,0,10.49533462524414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,48,2,128,1,float16,float16,0,4.740966415405273
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,48,48,128,1,float16,float16,0,8.499763488769531
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,48,4,128,1,float16,float16,0,5.233099365234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,48,8,128,1,float16,float16,0,11.640882873535157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,48,1,128,1,float16,float16,0,2.2767616271972657
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,48,8,128,1,float16,float16,0,5.48771858215332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,48,2,128,1,float16,float16,0,2.3136768341064453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,48,48,128,1,float16,float16,0,4.350156784057617
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,48,8,128,1,float16,float16,0,2.6571264266967773
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,48,48,128,1,float16,float16,0,2.1595136642456056
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,48,4,128,1,float16,float16,0,2.4590768814086914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,48,1,128,1,float16,float16,0,1.1609600067138672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,48,2,128,1,float16,float16,0,1.1487232208251954
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,48,8,128,1,float16,float16,0,1.3314559936523438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,48,48,128,1,float16,float16,0,1.1349504470825196
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,48,2,128,1,float16,float16,0,0.6166016101837158
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,48,4,128,1,float16,float16,0,1.2141056060791016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,48,1,128,1,float16,float16,0,0.6136320114135743
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,48,4,128,1,float16,float16,0,0.6189055919647217
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,48,8,128,1,float16,float16,0,0.6781951904296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,48,48,128,1,float16,float16,0,0.5575168132781982
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,48,1,128,1,float16,float16,0,0.31359999179840087
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,48,48,128,1,float16,float16,0,0.2747391939163208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,48,4,128,1,float16,float16,0,0.312064003944397
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,48,8,128,1,float16,float16,0,0.3331072092056274
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,48,2,128,1,float16,float16,0,0.30448479652404786
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,48,1,128,1,float16,float16,0,0.18600959777832032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,48,2,128,1,float16,float16,0,0.17966079711914062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,48,4,128,1,float16,float16,0,0.1847808003425598
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,48,8,128,1,float16,float16,0,0.19071999788284302
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,48,1,128,1,float16,float16,0,4.371148681640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,48,1,128,1,float16,float16,0,9.781913757324219
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,48,2,128,1,float16,float16,0,10.28264923095703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,48,4,128,1,float16,float16,0,10.830899047851563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,48,48,128,1,float16,float16,0,9.71294708251953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,48,8,128,1,float16,float16,0,11.569920349121094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,48,2,128,1,float16,float16,0,4.90142707824707
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,48,4,128,1,float16,float16,0,5.153689575195313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,48,1,128,1,float16,float16,0,2.2527488708496093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,48,2,128,1,float16,float16,0,2.345574378967285
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,48,8,128,1,float16,float16,0,5.592985534667969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,48,48,128,1,float16,float16,0,4.857702255249023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,48,8,128,1,float16,float16,0,2.687027168273926
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,48,4,128,1,float16,float16,0,2.420582389831543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,48,1,128,1,float16,float16,0,1.125887966156006
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,48,48,128,1,float16,float16,0,2.639360046386719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,48,2,128,1,float16,float16,0,1.159116840362549
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,48,4,128,1,float16,float16,0,1.2180480003356933
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,48,8,128,1,float16,float16,0,1.3537775993347168
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,48,1,128,1,float16,float16,0,0.5868544101715087
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,48,2,128,1,float16,float16,0,0.5979648113250733
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,48,4,128,1,float16,float16,0,0.6155776023864746
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,48,48,128,1,float16,float16,0,1.3293567657470704
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,48,8,128,1,float16,float16,0,0.6822336196899415
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,48,1,128,1,float16,float16,0,0.30919680595397947
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,48,48,128,1,float16,float16,0,0.624942398071289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,48,2,128,1,float16,float16,0,0.321124792098999
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,48,4,128,1,float16,float16,0,0.3290112018585205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,48,48,128,1,float16,float16,0,0.32138240337371826
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,48,8,128,1,float16,float16,0,0.3474431991577148
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,48,1,128,1,float16,float16,0,0.16542719602584838
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,48,2,128,1,float16,float16,0,0.16424959897994995
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,48,8,128,1,float16,float16,0,0.173363196849823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,48,48,128,1,float16,float16,0,0.12815359830856324
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,48,1,128,1,float16,float16,0,0.1047551989555359
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,48,4,128,1,float16,float16,0,0.1718783974647522
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,48,2,128,1,float16,float16,0,0.1028607964515686
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,48,4,128,1,float16,float16,0,0.10178560018539429
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,48,8,128,1,float16,float16,0,0.10654720067977905
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,48,1,128,1,float16,float16,0,6.199859237670898
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,48,2,128,1,float16,float16,0,6.2794750213623045
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,48,1,128,1,float16,float16,0,2.943129539489746
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,48,4,128,1,float16,float16,0,6.699571228027343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,48,2,128,1,float16,float16,0,2.9742576599121096
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,48,4,128,1,float16,float16,0,3.1536127090454102
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,48,8,128,1,float16,float16,0,7.665561676025391
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,48,48,128,1,float16,float16,0,9.400115203857421
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,48,8,128,1,float16,float16,0,3.7471233367919923
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,48,1,128,1,float16,float16,0,1.458022403717041
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,48,2,128,1,float16,float16,0,1.495910358428955
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,48,4,128,1,float16,float16,0,1.5649215698242187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,48,8,128,1,float16,float16,0,2.026291275024414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,48,48,128,1,float16,float16,0,4.765439987182617
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,48,48,128,1,float16,float16,0,2.426982307434082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,48,1,128,1,float16,float16,0,0.7457280158996582
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,48,2,128,1,float16,float16,0,0.759500789642334
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,48,4,128,1,float16,float16,0,0.8028160095214844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,48,8,128,1,float16,float16,0,0.9119232177734375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,48,1,128,1,float16,float16,0,0.3941375970840454
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,48,48,128,1,float16,float16,0,1.219377613067627
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,48,2,128,1,float16,float16,0,0.3959295988082886
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,48,4,128,1,float16,float16,0,0.41318402290344236
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,48,8,128,1,float16,float16,0,0.4635136127471924
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,48,1,128,1,float16,float16,0,0.20587520599365233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,48,48,128,1,float16,float16,0,0.6051328182220459
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,48,2,128,1,float16,float16,0,0.21544959545135497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,48,4,128,1,float16,float16,0,0.22522881031036376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,48,8,128,1,float16,float16,0,0.2412031888961792
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,48,48,128,1,float16,float16,0,0.2626559972763062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,48,1,128,1,float16,float16,0,0.10516480207443238
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,48,2,128,1,float16,float16,0,0.1063423991203308
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,48,8,128,1,float16,float16,0,0.11499199867248536
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,48,4,128,1,float16,float16,0,0.10920959711074829
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,48,48,128,1,float16,float16,0,0.09702399969100953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,48,1,128,1,float16,float16,0,0.06282240152359009
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,48,4,128,1,float16,float16,0,0.06758400201797485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,48,2,128,1,float16,float16,0,0.06282240152359009
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,48,48,128,1,float16,float16,0,0.06072319746017456
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,48,8,128,1,float16,float16,0,0.06840320229530335
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,48,1,128,1,float16,float16,0,0.04500479996204376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,48,4,128,1,float16,float16,0,0.04567039906978607
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,48,2,128,1,float16,float16,0,0.04444159865379334
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,48,8,128,1,float16,float16,0,0.04736000001430511
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,48,1,128,1,float16,float16,0,2.7205120086669923
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,48,2,128,1,float16,float16,0,2.8784128189086915
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,48,4,128,1,float16,float16,0,3.190015983581543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,48,8,128,1,float16,float16,0,3.8374912261962892
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,48,1,128,1,float16,float16,0,1.3585408210754395
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,48,2,128,1,float16,float16,0,1.4569984436035157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,48,4,128,1,float16,float16,0,1.5557632446289062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,48,48,128,1,float16,float16,0,4.716543960571289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,48,8,128,1,float16,float16,0,1.9583919525146485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,48,1,128,1,float16,float16,0,0.7202303886413575
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,48,48,128,1,float16,float16,0,2.4224767684936523
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,48,2,128,1,float16,float16,0,0.7567359924316406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,48,4,128,1,float16,float16,0,0.7968751907348632
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,48,8,128,1,float16,float16,0,0.9521663665771485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,48,48,128,1,float16,float16,0,1.2070912361145019
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,48,1,128,1,float16,float16,0,0.3531264066696167
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,48,4,128,1,float16,float16,0,0.3737087965011597
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,48,2,128,1,float16,float16,0,0.3703808069229126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,48,8,128,1,float16,float16,0,0.4916224002838135
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,48,48,128,1,float16,float16,0,0.5987840175628663
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,48,1,128,1,float16,float16,0,0.15303679704666137
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,48,4,128,1,float16,float16,0,0.177510404586792
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,48,8,128,1,float16,float16,0,0.20792319774627685
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,48,2,128,1,float16,float16,0,0.1632256031036377
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,48,48,128,1,float16,float16,0,0.2486272096633911
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,48,2,128,1,float16,float16,0,0.07567200064659119
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,48,1,128,1,float16,float16,0,0.07695199847221375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,48,8,128,1,float16,float16,0,0.08472800254821777
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,48,4,128,1,float16,float16,0,0.07823359966278076
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,48,48,128,1,float16,float16,0,0.07874559760093688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,48,1,128,1,float16,float16,0,0.048332801461219786
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,48,2,128,1,float16,float16,0,0.04755840003490448
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,48,4,128,1,float16,float16,0,0.048127999901771544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,48,8,128,1,float16,float16,0,0.05063679814338684
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,48,1,128,1,float16,float16,0,0.03404479920864105
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,48,48,128,1,float16,float16,0,0.048640000820159915
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,48,2,128,1,float16,float16,0,0.03363839983940124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,48,4,128,1,float16,float16,0,0.034508800506591795
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,48,8,128,1,float16,float16,0,0.03614720106124878
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,48,48,128,1,float16,float16,0,0.031027200818061828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,48,1,128,1,float16,float16,0,0.024217599630355836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,48,2,128,1,float16,float16,0,0.024063999950885772
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,48,4,128,1,float16,float16,0,0.024268800020217897
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,48,8,128,1,float16,float16,0,0.0247296005487442
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,48,2,128,1,float16,float16,0,1.4326272010803223
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,48,1,128,1,float16,float16,0,1.3519871711730957
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,48,4,128,1,float16,float16,0,1.580288028717041
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,48,8,128,1,float16,float16,0,1.8886144638061524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,48,1,128,1,float16,float16,0,0.6693376064300537
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,48,2,128,1,float16,float16,0,0.7184383869171143
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,48,48,128,1,float16,float16,0,2.3734272003173826
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,48,4,128,1,float16,float16,0,0.762828779220581
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,48,8,128,1,float16,float16,0,0.9738752365112304
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,48,1,128,1,float16,float16,0,0.35768320560455324
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,48,48,128,1,float16,float16,0,1.2050944328308106
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,48,2,128,1,float16,float16,0,0.3573760032653809
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,48,4,128,1,float16,float16,0,0.37406721115112307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,48,8,128,1,float16,float16,0,0.44805121421813965
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,48,48,128,1,float16,float16,0,0.6502399921417237
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,48,2,128,1,float16,float16,0,0.16619520187377929
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,48,1,128,1,float16,float16,0,0.12062079906463623
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,48,4,128,1,float16,float16,0,0.17198079824447632
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,48,8,128,1,float16,float16,0,0.21560320854187012
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,48,48,128,1,float16,float16,0,0.2441728115081787
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,48,1,128,1,float16,float16,0,0.0609279990196228
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,48,2,128,1,float16,float16,0,0.06272000074386597
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,48,4,128,1,float16,float16,0,0.06594560146331788
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,48,48,128,1,float16,float16,0,0.07060480117797852
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,48,1,128,1,float16,float16,0,0.0394239991903305
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,48,8,128,1,float16,float16,0,0.06993600130081176
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,48,2,128,1,float16,float16,0,0.03937279880046844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,48,4,128,1,float16,float16,0,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,48,8,128,1,float16,float16,0,0.04341759979724884
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,48,48,128,1,float16,float16,0,0.04500479996204376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,48,2,128,1,float16,float16,0,0.030355200171470642
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,48,1,128,1,float16,float16,0,0.030156800150871278
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,48,4,128,1,float16,float16,0,0.03061760067939758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,48,8,128,1,float16,float16,0,0.03307519853115082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,48,48,128,1,float16,float16,0,0.02754560112953186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,48,1,128,1,float16,float16,0,0.020427200198173522
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,48,2,128,1,float16,float16,0,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,48,4,128,1,float16,float16,0,0.02094080001115799
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,48,8,128,1,float16,float16,0,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,48,1,128,1,float16,float16,0,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,48,48,128,1,float16,float16,0,0.022012799978256226
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,48,2,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,48,4,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,48,8,128,1,float16,float16,0,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,48,1,128,1,float16,float16,0,0.6674943923950195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,48,2,128,1,float16,float16,0,0.6904319763183594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,48,4,128,1,float16,float16,0,0.7662000179290771
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,48,8,128,1,float16,float16,0,0.9377280235290527
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,48,48,128,1,float16,float16,0,1.176524829864502
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,48,1,128,1,float16,float16,0,0.32778239250183105
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,48,2,128,1,float16,float16,0,0.3422719955444336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,48,4,128,1,float16,float16,0,0.3778048038482666
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,48,8,128,1,float16,float16,0,0.4471295833587646
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,48,1,128,1,float16,float16,0,0.12615679502487182
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,48,48,128,1,float16,float16,0,0.5799424171447753
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,48,2,128,1,float16,float16,0,0.16957440376281738
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,48,4,128,1,float16,float16,0,0.17766400575637817
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,48,8,128,1,float16,float16,0,0.20029280185699463
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,48,48,128,1,float16,float16,0,0.24970240592956544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,48,1,128,1,float16,float16,0,0.05975040197372437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,48,2,128,1,float16,float16,0,0.061286401748657224
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,48,4,128,1,float16,float16,0,0.06460639834403992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,48,8,128,1,float16,float16,0,0.06906880140304565
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,48,48,128,1,float16,float16,0,0.0880128026008606
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,48,1,128,1,float16,float16,0,0.03803519904613495
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,48,2,128,1,float16,float16,0,0.03819519877433777
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,48,4,128,1,float16,float16,0,0.0392192006111145
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,48,8,128,1,float16,float16,0,0.042444801330566405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,48,1,128,1,float16,float16,0,0.029900801181793214
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,48,48,128,1,float16,float16,0,0.050329601764678954
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,48,2,128,1,float16,float16,0,0.029337599873542786
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,48,4,128,1,float16,float16,0,0.029849600791931153
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,48,8,128,1,float16,float16,0,0.030771198868751525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,48,48,128,1,float16,float16,0,0.03374080061912536
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,48,1,128,1,float16,float16,0,0.020121599733829498
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,48,2,128,1,float16,float16,0,0.020377600193023683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,48,4,128,1,float16,float16,0,0.020838400721549986
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,48,48,128,1,float16,float16,0,0.02160640060901642
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,48,1,128,1,float16,float16,0,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,48,8,128,1,float16,float16,0,0.020633600652217865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,48,2,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,48,4,128,1,float16,float16,0,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,48,8,128,1,float16,float16,0,0.019096000492572783
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,48,48,128,1,float16,float16,0,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,48,1,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,48,2,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,48,8,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,48,4,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,48,1,128,1,float16,float16,0,0.3294719934463501
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,48,2,128,1,float16,float16,0,0.34104321002960203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,48,8,128,1,float16,float16,0,0.42731518745422364
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,48,4,128,1,float16,float16,0,0.36807520389556886
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,48,48,128,1,float16,float16,0,0.5857279777526856
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,48,1,128,1,float16,float16,0,0.1144320011138916
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,48,2,128,1,float16,float16,0,0.14986239671707152
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,48,4,128,1,float16,float16,0,0.1777151942253113
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,48,8,128,1,float16,float16,0,0.20485119819641112
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,48,2,128,1,float16,float16,0,0.061440002918243405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,48,48,128,1,float16,float16,0,0.30033919811248777
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,48,1,128,1,float16,float16,0,0.059084802865982056
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,48,4,128,1,float16,float16,0,0.06415359973907471
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,48,8,128,1,float16,float16,0,0.0807424008846283
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,48,48,128,1,float16,float16,0,0.12252160310745239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,48,2,128,1,float16,float16,0,0.03840000033378601
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,48,1,128,1,float16,float16,0,0.03798080086708069
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,48,8,128,1,float16,float16,0,0.048588800430297854
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,48,4,128,1,float16,float16,0,0.0394239991903305
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,48,48,128,1,float16,float16,0,0.06947839856147767
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,48,1,128,1,float16,float16,0,0.029542401432991028
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,48,2,128,1,float16,float16,0,0.029440000653266907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,48,4,128,1,float16,float16,0,0.029235199093818665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,48,8,128,1,float16,float16,0,0.03061760067939758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,48,48,128,1,float16,float16,0,0.040089601278305055
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,48,1,128,1,float16,float16,0,0.019963200390338897
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,48,2,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,48,8,128,1,float16,float16,0,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,48,48,128,1,float16,float16,0,0.028672000765800475
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,48,4,128,1,float16,float16,0,0.020423999428749083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,48,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,48,2,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,48,4,128,1,float16,float16,0,0.018935999274253844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,48,8,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,48,1,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,48,2,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,48,48,128,1,float16,float16,0,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,48,4,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,48,8,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,48,48,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,48,1,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,48,2,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,48,4,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,48,1,128,1,float16,float16,0,0.12646399736404418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,48,8,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,48,4,128,1,float16,float16,0,0.18027520179748535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,48,2,128,1,float16,float16,0,0.146943998336792
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,48,8,128,1,float16,float16,0,0.20787200927734376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,48,48,128,1,float16,float16,0,0.43468799591064455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,48,1,128,1,float16,float16,0,0.06061760187149048
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,48,2,128,1,float16,float16,0,0.061900800466537474
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,48,8,128,1,float16,float16,0,0.08207359910011292
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,48,4,128,1,float16,float16,0,0.07470080256462097
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,48,48,128,1,float16,float16,0,0.19589120149612427
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,48,1,128,1,float16,float16,0,0.038553598523139956
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,48,2,128,1,float16,float16,0,0.03870719969272614
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,48,4,128,1,float16,float16,0,0.045772799849510194
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,48,8,128,1,float16,float16,0,0.048844799399375916
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,48,48,128,1,float16,float16,0,0.10419199466705323
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,48,1,128,1,float16,float16,0,0.029491201043128967
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,48,2,128,1,float16,float16,0,0.029388800263404846
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,48,4,128,1,float16,float16,0,0.029951998591423036
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,48,8,128,1,float16,float16,0,0.030668801069259642
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,48,48,128,1,float16,float16,0,0.060356801748275755
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,48,2,128,1,float16,float16,0,0.0201664000749588
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,48,1,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,48,4,128,1,float16,float16,0,0.020838400721549986
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,48,8,128,1,float16,float16,0,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,48,48,128,1,float16,float16,0,0.035123199224472046
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,48,1,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,48,2,128,1,float16,float16,0,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,48,4,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,48,8,128,1,float16,float16,0,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,48,48,128,1,float16,float16,0,0.02595199942588806
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,48,1,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,48,2,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,48,8,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,48,4,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,48,48,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,48,1,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,48,2,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,48,4,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,48,48,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,48,8,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,48,1,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,48,2,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,48,4,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,48,8,128,1,float16,float16,0,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,40,1,128,1,float16,float16,0,49.09501342773437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,40,2,128,1,float16,float16,0,49.213644409179686
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,40,4,128,1,float16,float16,0,49.4908935546875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,40,40,128,1,float16,float16,0,54.292529296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,40,1,128,1,float16,float16,0,23.607699584960937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,40,40,128,1,float16,float16,0,27.33301696777344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,40,2,128,1,float16,float16,0,24.210226440429686
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,40,1,128,1,float16,float16,0,98.3214599609375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,40,2,128,1,float16,float16,0,100.25523071289062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,40,4,128,1,float16,float16,0,101.7554931640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,40,8,128,1,float16,float16,0,102.2888427734375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,40,8,128,1,float16,float16,0,51.62276000976563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,40,1,128,1,float16,float16,0,11.602381134033203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,40,40,128,1,float16,float16,0,13.205197143554688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,40,2,128,1,float16,float16,0,12.008243560791016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,40,4,128,1,float16,float16,0,12.138495635986327
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,40,4,128,1,float16,float16,0,24.432588195800783
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,40,8,128,1,float16,float16,0,12.568627166748048
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,40,8,128,1,float16,float16,0,25.230335998535157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,40,1,128,1,float16,float16,0,27.043429565429687
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,40,40,128,1,float16,float16,0,31.44320068359375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,40,2,128,1,float16,float16,0,27.970303344726563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,40,4,128,1,float16,float16,0,28.378573608398437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,40,2,128,1,float16,float16,0,56.104705810546875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,40,1,128,1,float16,float16,0,56.97315673828125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,40,4,128,1,float16,float16,0,57.530572509765626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,40,1,128,1,float16,float16,0,13.734042358398437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,40,8,128,1,float16,float16,0,59.247003173828126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,40,2,128,1,float16,float16,0,13.518438720703125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,40,40,128,1,float16,float16,0,15.555584716796876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,40,8,128,1,float16,float16,0,28.428134155273437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,40,40,128,1,float16,float16,0,7.393785858154297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,40,1,128,1,float16,float16,0,6.369996643066406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,40,4,128,1,float16,float16,0,13.720420837402344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,40,2,128,1,float16,float16,0,6.722252655029297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,40,4,128,1,float16,float16,0,6.969036865234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,40,8,128,1,float16,float16,0,14.449459838867188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,40,8,128,1,float16,float16,0,6.868326568603516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,40,1,128,1,float16,float16,0,19.173989868164064
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,40,4,128,1,float16,float16,0,19.456153869628906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,40,2,128,1,float16,float16,0,19.07123260498047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,40,40,128,1,float16,float16,0,22.72337951660156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,40,2,128,1,float16,float16,0,38.576742553710936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,40,1,128,1,float16,float16,0,38.231039428710936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,40,4,128,1,float16,float16,0,39.45062255859375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,40,8,128,1,float16,float16,0,41.10597229003906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,40,1,128,1,float16,float16,0,9.415936279296876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,40,40,128,1,float16,float16,0,11.05269775390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,40,2,128,1,float16,float16,0,9.630258941650391
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,40,8,128,1,float16,float16,0,20.244172668457033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,40,4,128,1,float16,float16,0,9.56451873779297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,40,8,128,1,float16,float16,0,9.590217590332031
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,40,40,128,1,float16,float16,0,5.329049682617187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,40,1,128,1,float16,float16,0,4.665241622924805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,40,2,128,1,float16,float16,0,4.386304092407227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,40,4,128,1,float16,float16,0,4.698931121826172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,40,8,128,1,float16,float16,0,4.829542541503907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,40,2,128,1,float16,float16,0,25.583001708984376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,40,1,128,1,float16,float16,0,25.41460418701172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,40,4,128,1,float16,float16,0,25.930035400390626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,40,40,128,1,float16,float16,0,30.106317138671876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,40,1,128,1,float16,float16,0,50.7884033203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,40,2,128,1,float16,float16,0,50.83023376464844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,40,1,128,1,float16,float16,0,12.29265899658203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,40,4,128,1,float16,float16,0,52.43729858398437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,40,40,128,1,float16,float16,0,14.914816284179688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,40,8,128,1,float16,float16,0,54.07948608398438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,40,2,128,1,float16,float16,0,12.413849639892579
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,40,8,128,1,float16,float16,0,26.9497314453125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,40,4,128,1,float16,float16,0,12.867225646972656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,40,40,128,1,float16,float16,0,7.317401885986328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,40,1,128,1,float16,float16,0,6.0959217071533205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,40,2,128,1,float16,float16,0,6.0554241180419925
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,40,4,128,1,float16,float16,0,5.988761520385742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,40,8,128,1,float16,float16,0,13.063217163085938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,40,8,128,1,float16,float16,0,6.328115081787109
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,40,40,128,1,float16,float16,0,3.5755008697509765
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,40,1,128,1,float16,float16,0,2.9156864166259764
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,40,2,128,1,float16,float16,0,2.8648448944091798
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,40,4,128,1,float16,float16,0,2.893926429748535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,40,8,128,1,float16,float16,0,2.9855743408203126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,40,1,128,1,float16,float16,0,13.93489990234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,40,2,128,1,float16,float16,0,14.20067901611328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,40,4,128,1,float16,float16,0,14.921676635742188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,40,40,128,1,float16,float16,0,17.916773986816406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,40,1,128,1,float16,float16,0,28.644351196289062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,40,2,128,1,float16,float16,0,29.116415405273436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,40,4,128,1,float16,float16,0,29.72907409667969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,40,8,128,1,float16,float16,0,31.205682373046876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,40,1,128,1,float16,float16,0,6.9278205871582035
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,40,40,128,1,float16,float16,0,8.550860595703124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,40,2,128,1,float16,float16,0,7.213816070556641
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,40,40,128,1,float16,float16,0,4.2846721649169925
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,40,4,128,1,float16,float16,0,7.254067230224609
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,40,8,128,1,float16,float16,0,7.538739013671875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,40,8,128,1,float16,float16,0,14.808213806152343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,40,1,128,1,float16,float16,0,3.4475521087646483
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,40,2,128,1,float16,float16,0,3.1644159317016602
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,40,4,128,1,float16,float16,0,3.453286361694336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,40,40,128,1,float16,float16,0,2.152191925048828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,40,1,128,1,float16,float16,0,1.6375295639038085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,40,8,128,1,float16,float16,0,3.50115852355957
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,40,2,128,1,float16,float16,0,1.6440319061279296
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,40,4,128,1,float16,float16,0,1.6739839553833007
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,40,8,128,1,float16,float16,0,1.7558015823364257
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,40,1,128,1,float16,float16,0,12.790937805175782
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,40,2,128,1,float16,float16,0,13.150259399414063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,40,40,128,1,float16,float16,0,17.595233154296874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,40,4,128,1,float16,float16,0,14.059516906738281
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,40,1,128,1,float16,float16,0,26.770892333984374
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,40,2,128,1,float16,float16,0,26.68451843261719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,40,4,128,1,float16,float16,0,28.419891357421875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,40,8,128,1,float16,float16,0,28.896868896484374
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,40,2,128,1,float16,float16,0,6.3288318634033205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,40,1,128,1,float16,float16,0,6.5101264953613285
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,40,40,128,1,float16,float16,0,8.47119369506836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,40,8,128,1,float16,float16,0,14.319973754882813
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,40,4,128,1,float16,float16,0,6.667826843261719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,40,40,128,1,float16,float16,0,4.302848052978516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,40,8,128,1,float16,float16,0,7.089305877685547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,40,1,128,1,float16,float16,0,3.0177791595458983
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,40,2,128,1,float16,float16,0,3.0575103759765625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,40,4,128,1,float16,float16,0,3.1612415313720703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,40,8,128,1,float16,float16,0,3.209830474853516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,40,1,128,1,float16,float16,0,1.5118847846984864
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,40,2,128,1,float16,float16,0,1.5205375671386718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,40,40,128,1,float16,float16,0,2.1518335342407227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,40,4,128,1,float16,float16,0,1.557145595550537
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,40,8,128,1,float16,float16,0,1.6084991455078126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,40,40,128,1,float16,float16,0,1.0845696449279785
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,40,1,128,1,float16,float16,0,0.7625728130340577
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,40,2,128,1,float16,float16,0,0.7526400089263916
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,40,4,128,1,float16,float16,0,0.7882239818572998
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,40,8,128,1,float16,float16,0,0.8368127822875977
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,40,1,128,1,float16,float16,0,15.479653930664062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,40,1,128,1,float16,float16,0,7.445241546630859
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,40,2,128,1,float16,float16,0,15.612518310546875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,40,2,128,1,float16,float16,0,7.609600067138672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,40,40,128,1,float16,float16,0,10.50967025756836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,40,4,128,1,float16,float16,0,16.16046142578125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,40,4,128,1,float16,float16,0,7.948945617675781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,40,8,128,1,float16,float16,0,17.15747833251953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,40,2,128,1,float16,float16,0,3.594188690185547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,40,1,128,1,float16,float16,0,3.506278228759766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,40,40,128,1,float16,float16,0,5.2265472412109375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,40,4,128,1,float16,float16,0,3.784960174560547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,40,8,128,1,float16,float16,0,3.935078430175781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,40,40,128,1,float16,float16,0,2.622719955444336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,40,8,128,1,float16,float16,0,8.145970916748047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,40,1,128,1,float16,float16,0,1.762303924560547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,40,4,128,1,float16,float16,0,1.8498559951782227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,40,2,128,1,float16,float16,0,1.689139175415039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,40,40,128,1,float16,float16,0,1.3238271713256835
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,40,8,128,1,float16,float16,0,1.9233280181884767
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,40,1,128,1,float16,float16,0,0.8803839683532715
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,40,2,128,1,float16,float16,0,0.8924624443054199
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,40,4,128,1,float16,float16,0,0.9337856292724609
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,40,8,128,1,float16,float16,0,0.9893888473510742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,40,40,128,1,float16,float16,0,0.6915040016174316
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,40,4,128,1,float16,float16,0,0.47626237869262694
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,40,2,128,1,float16,float16,0,0.48358402252197263
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,40,1,128,1,float16,float16,0,0.48522238731384276
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,40,8,128,1,float16,float16,0,0.4779520034790039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,40,1,128,1,float16,float16,0,14.096383666992187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,40,2,128,1,float16,float16,0,14.91394500732422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,40,1,128,1,float16,float16,0,6.792704010009766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,40,2,128,1,float16,float16,0,7.306700897216797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,40,4,128,1,float16,float16,0,15.513548278808594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,40,4,128,1,float16,float16,0,7.504332733154297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,40,40,128,1,float16,float16,0,11.020953369140624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,40,8,128,1,float16,float16,0,16.394444274902344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,40,1,128,1,float16,float16,0,3.3287166595458983
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,40,2,128,1,float16,float16,0,3.3675262451171877
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,40,40,128,1,float16,float16,0,5.492326354980468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,40,4,128,1,float16,float16,0,3.5097599029541016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,40,8,128,1,float16,float16,0,7.816294097900391
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,40,8,128,1,float16,float16,0,3.7479934692382812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,40,40,128,1,float16,float16,0,2.762188720703125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,40,1,128,1,float16,float16,0,1.6253952026367187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,40,2,128,1,float16,float16,0,1.62554874420166
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,40,4,128,1,float16,float16,0,1.7517568588256835
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,40,40,128,1,float16,float16,0,1.3971967697143555
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,40,8,128,1,float16,float16,0,1.8691583633422852
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,40,1,128,1,float16,float16,0,0.8434127807617188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,40,2,128,1,float16,float16,0,0.8398336410522461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,40,4,128,1,float16,float16,0,0.8674304008483886
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,40,8,128,1,float16,float16,0,0.9353728294372559
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,40,40,128,1,float16,float16,0,0.7061503887176513
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,40,1,128,1,float16,float16,0,0.4167168140411377
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,40,2,128,1,float16,float16,0,0.4242415904998779
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,40,4,128,1,float16,float16,0,0.43653120994567873
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,40,8,128,1,float16,float16,0,0.4797952175140381
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,40,40,128,1,float16,float16,0,0.3606528043746948
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,40,2,128,1,float16,float16,0,0.23813118934631347
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,40,1,128,1,float16,float16,0,0.23781440258026124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,40,4,128,1,float16,float16,0,0.24929280281066896
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,40,8,128,1,float16,float16,0,0.25553920269012453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,40,1,128,1,float16,float16,0,8.514969635009766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,40,2,128,1,float16,float16,0,8.509337615966796
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,40,1,128,1,float16,float16,0,3.9496192932128906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,40,4,128,1,float16,float16,0,9.264691162109376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,40,40,128,1,float16,float16,0,7.134207916259766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,40,2,128,1,float16,float16,0,4.129945755004883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,40,4,128,1,float16,float16,0,4.075417709350586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,40,8,128,1,float16,float16,0,9.666150665283203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,40,1,128,1,float16,float16,0,1.9268064498901367
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,40,2,128,1,float16,float16,0,1.9843584060668946
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,40,4,128,1,float16,float16,0,2.1204992294311524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,40,40,128,1,float16,float16,0,1.7929727554321289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,40,8,128,1,float16,float16,0,2.270310401916504
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,40,8,128,1,float16,float16,0,4.6993408203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,40,1,128,1,float16,float16,0,0.9568256378173828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,40,40,128,1,float16,float16,0,3.5238399505615234
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,40,2,128,1,float16,float16,0,0.9780223846435547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,40,1,128,1,float16,float16,0,0.5157375812530518
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,40,8,128,1,float16,float16,0,1.1496959686279298
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,40,4,128,1,float16,float16,0,1.0611712455749511
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,40,4,128,1,float16,float16,0,0.5321728229522705
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,40,40,128,1,float16,float16,0,0.9174015998840332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,40,2,128,1,float16,float16,0,0.5242368221282959
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,40,8,128,1,float16,float16,0,0.5807583808898926
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,40,40,128,1,float16,float16,0,0.47518720626831057
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,40,2,128,1,float16,float16,0,0.27095038890838624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,40,1,128,1,float16,float16,0,0.26081280708312987
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,40,4,128,1,float16,float16,0,0.27356159687042236
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,40,8,128,1,float16,float16,0,0.2759167909622192
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,40,1,128,1,float16,float16,0,0.15395840406417846
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,40,40,128,1,float16,float16,0,0.19927040338516236
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,40,2,128,1,float16,float16,0,0.15600639581680298
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,40,4,128,1,float16,float16,0,0.16640000343322753
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,40,8,128,1,float16,float16,0,0.17002719640731812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,40,1,128,1,float16,float16,0,8.173567962646484
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,40,1,128,1,float16,float16,0,3.9012351989746095
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,40,2,128,1,float16,float16,0,8.399616241455078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,40,40,128,1,float16,float16,0,8.02841567993164
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,40,4,128,1,float16,float16,0,8.921234893798829
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,40,8,128,1,float16,float16,0,9.904537963867188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,40,2,128,1,float16,float16,0,3.9713790893554686
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,40,4,128,1,float16,float16,0,4.378521728515625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,40,1,128,1,float16,float16,0,1.917184066772461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,40,2,128,1,float16,float16,0,1.9088895797729493
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,40,40,128,1,float16,float16,0,4.095948791503906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,40,8,128,1,float16,float16,0,4.655718231201172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,40,4,128,1,float16,float16,0,2.104115104675293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,40,8,128,1,float16,float16,0,2.3236608505249023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,40,40,128,1,float16,float16,0,2.0518911361694334
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,40,1,128,1,float16,float16,0,0.9603072166442871
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,40,2,128,1,float16,float16,0,0.9785344123840332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,40,4,128,1,float16,float16,0,1.0218496322631836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,40,8,128,1,float16,float16,0,1.1721728324890137
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,40,2,128,1,float16,float16,0,0.5128704071044922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,40,4,128,1,float16,float16,0,0.5364736080169678
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,40,1,128,1,float16,float16,0,0.4979712009429932
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,40,40,128,1,float16,float16,0,1.083033561706543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,40,8,128,1,float16,float16,0,0.5883903980255127
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,40,40,128,1,float16,float16,0,0.5279168128967285
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,40,1,128,1,float16,float16,0,0.24412159919738768
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,40,8,128,1,float16,float16,0,0.2957823991775513
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,40,2,128,1,float16,float16,0,0.2510335922241211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,40,40,128,1,float16,float16,0,0.2590719938278198
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,40,4,128,1,float16,float16,0,0.265881609916687
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,40,2,128,1,float16,float16,0,0.14617600440979003
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,40,1,128,1,float16,float16,0,0.14069759845733643
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,40,4,128,1,float16,float16,0,0.14602240324020385
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,40,8,128,1,float16,float16,0,0.15006719827651976
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,40,1,128,1,float16,float16,0,0.09077759981155395
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,40,2,128,1,float16,float16,0,0.09041919708251953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,40,40,128,1,float16,float16,0,0.11596800088882446
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,40,4,128,1,float16,float16,0,0.0972208023071289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,40,8,128,1,float16,float16,0,0.10019840002059936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,40,1,128,1,float16,float16,0,4.978579330444336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,40,2,128,1,float16,float16,0,5.368678283691406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,40,1,128,1,float16,float16,0,2.390732765197754
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,40,4,128,1,float16,float16,0,5.696051025390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,40,2,128,1,float16,float16,0,2.523494338989258
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,40,4,128,1,float16,float16,0,2.7932159423828127
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,40,8,128,1,float16,float16,0,6.564608001708985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,40,40,128,1,float16,float16,0,7.887872314453125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,40,8,128,1,float16,float16,0,3.399628829956055
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,40,1,128,1,float16,float16,0,1.2051456451416016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,40,2,128,1,float16,float16,0,1.256447982788086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,40,4,128,1,float16,float16,0,1.4078463554382323
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,40,8,128,1,float16,float16,0,1.68985595703125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,40,40,128,1,float16,float16,0,4.0301567077636715
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,40,40,128,1,float16,float16,0,2.18710880279541
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,40,1,128,1,float16,float16,0,0.6180863857269288
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,40,2,128,1,float16,float16,0,0.6196735858917236
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,40,4,128,1,float16,float16,0,0.6923776149749756
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,40,40,128,1,float16,float16,0,1.0168319702148438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,40,1,128,1,float16,float16,0,0.3374592065811157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,40,8,128,1,float16,float16,0,0.8091648101806641
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,40,2,128,1,float16,float16,0,0.33909759521484373
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,40,4,128,1,float16,float16,0,0.35773439407348634
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,40,8,128,1,float16,float16,0,0.40616960525512696
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,40,40,128,1,float16,float16,0,0.4886015892028809
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,40,1,128,1,float16,float16,0,0.1600000023841858
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,40,2,128,1,float16,float16,0,0.1662976026535034
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,40,4,128,1,float16,float16,0,0.17392640113830565
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,40,8,128,1,float16,float16,0,0.20602879524230958
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,40,40,128,1,float16,float16,0,0.2095616102218628
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,40,1,128,1,float16,float16,0,0.09190400242805481
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,40,2,128,1,float16,float16,0,0.0932864010334015
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,40,4,128,1,float16,float16,0,0.09763839840888977
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,40,8,128,1,float16,float16,0,0.10178560018539429
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,40,40,128,1,float16,float16,0,0.08499199748039246
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,40,1,128,1,float16,float16,0,0.055193597078323366
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,40,2,128,1,float16,float16,0,0.05811200141906738
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,40,4,128,1,float16,float16,0,0.05985280275344849
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,40,8,128,1,float16,float16,0,0.06231039762496948
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,40,40,128,1,float16,float16,0,0.0536575973033905
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,40,1,128,1,float16,float16,0,0.04080640077590943
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,40,2,128,1,float16,float16,0,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,40,4,128,1,float16,float16,0,0.04162560105323791
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,40,8,128,1,float16,float16,0,0.04300000071525574
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,40,1,128,1,float16,float16,0,2.2763519287109375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,40,2,128,1,float16,float16,0,2.3856639862060547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,40,4,128,1,float16,float16,0,2.787225532531738
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,40,8,128,1,float16,float16,0,3.4003456115722654
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,40,1,128,1,float16,float16,0,1.1193344116210937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,40,2,128,1,float16,float16,0,1.1965439796447754
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,40,4,128,1,float16,float16,0,1.3947903633117675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,40,40,128,1,float16,float16,0,3.9524288177490234
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,40,8,128,1,float16,float16,0,1.732659149169922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,40,1,128,1,float16,float16,0,0.5950463771820068
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,40,40,128,1,float16,float16,0,2.013644790649414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,40,2,128,1,float16,float16,0,0.6416895866394043
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,40,4,128,1,float16,float16,0,0.6872064113616944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,40,8,128,1,float16,float16,0,0.8475135803222656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,40,40,128,1,float16,float16,0,1.0086912155151366
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,40,1,128,1,float16,float16,0,0.298854398727417
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,40,2,128,1,float16,float16,0,0.31912961006164553
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,40,4,128,1,float16,float16,0,0.3191807985305786
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,40,8,128,1,float16,float16,0,0.4334080219268799
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,40,1,128,1,float16,float16,0,0.11166720390319824
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,40,2,128,1,float16,float16,0,0.11530239582061767
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,40,40,128,1,float16,float16,0,0.47861762046813966
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,40,4,128,1,float16,float16,0,0.12584480047225952
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,40,8,128,1,float16,float16,0,0.17304960489273072
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,40,40,128,1,float16,float16,0,0.189900803565979
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,40,1,128,1,float16,float16,0,0.06568959951400757
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,40,4,128,1,float16,float16,0,0.07029759883880615
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,40,2,128,1,float16,float16,0,0.06753280162811279
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,40,8,128,1,float16,float16,0,0.0754688024520874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,40,40,128,1,float16,float16,0,0.06855679750442505
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,40,1,128,1,float16,float16,0,0.04142079949378967
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,40,2,128,1,float16,float16,0,0.042080000042915344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,40,4,128,1,float16,float16,0,0.044894400238990786
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,40,8,128,1,float16,float16,0,0.04617919921875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,40,40,128,1,float16,float16,0,0.04618239998817444
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,40,1,128,1,float16,float16,0,0.03307519853115082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,40,2,128,1,float16,float16,0,0.03333120048046112
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,40,4,128,1,float16,float16,0,0.03358719944953918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,40,8,128,1,float16,float16,0,0.034815999865531924
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,40,1,128,1,float16,float16,0,0.02380799949169159
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,40,40,128,1,float16,float16,0,0.02969599962234497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,40,2,128,1,float16,float16,0,0.024166400730609893
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,40,4,128,1,float16,float16,0,0.024422399699687958
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,40,8,128,1,float16,float16,0,0.0247296005487442
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,40,1,128,1,float16,float16,0,1.1385343551635743
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,40,2,128,1,float16,float16,0,1.1924991607666016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,40,4,128,1,float16,float16,0,1.3901823997497558
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,40,8,128,1,float16,float16,0,1.672755241394043
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,40,40,128,1,float16,float16,0,1.9726335525512695
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,40,1,128,1,float16,float16,0,0.5659647941589355
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,40,2,128,1,float16,float16,0,0.5805007934570312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,40,4,128,1,float16,float16,0,0.659609603881836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,40,8,128,1,float16,float16,0,0.8277503967285156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,40,1,128,1,float16,float16,0,0.2958847999572754
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,40,40,128,1,float16,float16,0,1.0045951843261718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,40,4,128,1,float16,float16,0,0.34447360038757324
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,40,2,128,1,float16,float16,0,0.2934783935546875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,40,8,128,1,float16,float16,0,0.39613440036773684
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,40,40,128,1,float16,float16,0,0.48102397918701173
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,40,1,128,1,float16,float16,0,0.08975359797477722
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,40,2,128,1,float16,float16,0,0.09502720236778259
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,40,4,128,1,float16,float16,0,0.10639359951019287
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,40,8,128,1,float16,float16,0,0.16650240421295165
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,40,40,128,1,float16,float16,0,0.17786879539489747
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,40,1,128,1,float16,float16,0,0.053913599252700804
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,40,2,128,1,float16,float16,0,0.05560320019721985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,40,4,128,1,float16,float16,0,0.05877760052680969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,40,8,128,1,float16,float16,0,0.06328319907188415
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,40,40,128,1,float16,float16,0,0.06379520297050476
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,40,1,128,1,float16,float16,0,0.03706879913806915
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,40,4,128,1,float16,float16,0,0.03895359933376312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,40,2,128,1,float16,float16,0,0.037427198886871335
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,40,40,128,1,float16,float16,0,0.04208639860153198
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,40,8,128,1,float16,float16,0,0.04152320027351379
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,40,1,128,1,float16,float16,0,0.02969599962234497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,40,2,128,1,float16,float16,0,0.029747200012207032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,40,40,128,1,float16,float16,0,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,40,8,128,1,float16,float16,0,0.030976000428199767
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,40,4,128,1,float16,float16,0,0.029440000653266907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,40,1,128,1,float16,float16,0,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,40,4,128,1,float16,float16,0,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,40,2,128,1,float16,float16,0,0.0203232005238533
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,40,8,128,1,float16,float16,0,0.020735999941825865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,40,40,128,1,float16,float16,0,0.02160640060901642
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,40,1,128,1,float16,float16,0,0.020582400262355804
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,40,2,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,40,4,128,1,float16,float16,0,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,40,8,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,40,1,128,1,float16,float16,0,0.5526016235351563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,40,2,128,1,float16,float16,0,0.5886447906494141
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,40,4,128,1,float16,float16,0,0.6738431930541993
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,40,8,128,1,float16,float16,0,0.8437760353088379
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,40,40,128,1,float16,float16,0,0.9787903785705566
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,40,1,128,1,float16,float16,0,0.2736639976501465
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,40,2,128,1,float16,float16,0,0.2873856067657471
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,40,8,128,1,float16,float16,0,0.3906048059463501
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,40,4,128,1,float16,float16,0,0.3181040048599243
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,40,40,128,1,float16,float16,0,0.46822400093078614
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,40,1,128,1,float16,float16,0,0.08765439987182617
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,40,2,128,1,float16,float16,0,0.09123839735984803
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,40,4,128,1,float16,float16,0,0.10357439517974854
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,40,8,128,1,float16,float16,0,0.165011203289032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,40,1,128,1,float16,float16,0,0.05324800014495849
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,40,40,128,1,float16,float16,0,0.193996798992157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,40,4,128,1,float16,float16,0,0.056678402423858645
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,40,2,128,1,float16,float16,0,0.053504002094268796
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,40,8,128,1,float16,float16,0,0.06768640279769897
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,40,40,128,1,float16,float16,0,0.07562239766120911
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,40,1,128,1,float16,float16,0,0.03671039938926697
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,40,2,128,1,float16,float16,0,0.037171199917793274
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,40,4,128,1,float16,float16,0,0.038550400733947755
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,40,8,128,1,float16,float16,0,0.04131680130958557
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,40,40,128,1,float16,float16,0,0.04715520143508911
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,40,2,128,1,float16,float16,0,0.028672000765800475
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,40,1,128,1,float16,float16,0,0.02866879999637604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,40,4,128,1,float16,float16,0,0.029175999760627746
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,40,8,128,1,float16,float16,0,0.0304639995098114
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,40,1,128,1,float16,float16,0,0.020372800529003143
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,40,40,128,1,float16,float16,0,0.03189760148525238
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,40,2,128,1,float16,float16,0,0.020377600193023683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,40,4,128,1,float16,float16,0,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,40,8,128,1,float16,float16,0,0.020377600193023683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,40,40,128,1,float16,float16,0,0.02109439969062805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,40,1,128,1,float16,float16,0,0.019097599387168884
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,40,2,128,1,float16,float16,0,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,40,4,128,1,float16,float16,0,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,40,8,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,40,40,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,40,2,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,40,1,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,40,4,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,40,8,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,40,1,128,1,float16,float16,0,0.27822079658508303
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,40,2,128,1,float16,float16,0,0.28764159679412843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,40,4,128,1,float16,float16,0,0.31636478900909426
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,40,8,128,1,float16,float16,0,0.37201919555664065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,40,40,128,1,float16,float16,0,0.4883456230163574
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,40,1,128,1,float16,float16,0,0.08897759914398193
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,40,2,128,1,float16,float16,0,0.09144319891929627
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,40,4,128,1,float16,float16,0,0.11581439971923828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,40,8,128,1,float16,float16,0,0.1752575993537903
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,40,40,128,1,float16,float16,0,0.2461184024810791
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,40,1,128,1,float16,float16,0,0.05294079780578613
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,40,2,128,1,float16,float16,0,0.054118400812149046
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,40,4,128,1,float16,float16,0,0.06231039762496948
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,40,8,128,1,float16,float16,0,0.07910400032997131
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,40,40,128,1,float16,float16,0,0.10828800201416015
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,40,1,128,1,float16,float16,0,0.03671039938926697
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,40,2,128,1,float16,float16,0,0.03947519958019256
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,40,4,128,1,float16,float16,0,0.03865599930286408
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,40,8,128,1,float16,float16,0,0.04715520143508911
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,40,40,128,1,float16,float16,0,0.06062080264091492
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,40,1,128,1,float16,float16,0,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,40,2,128,1,float16,float16,0,0.028415998816490172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,40,4,128,1,float16,float16,0,0.028620800375938414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,40,8,128,1,float16,float16,0,0.03030399978160858
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,40,40,128,1,float16,float16,0,0.03834879994392395
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,40,1,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,40,2,128,1,float16,float16,0,0.0203232005238533
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,40,4,128,1,float16,float16,0,0.020068800449371337
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,40,8,128,1,float16,float16,0,0.020633600652217865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,40,40,128,1,float16,float16,0,0.027187201380729675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,40,1,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,40,4,128,1,float16,float16,0,0.0189423993229866
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,40,2,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,40,8,128,1,float16,float16,0,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,40,40,128,1,float16,float16,0,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,40,1,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,40,2,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,40,4,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,40,40,128,1,float16,float16,0,0.018220800161361694
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,40,8,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,40,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,40,2,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,40,4,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,40,8,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,40,1,128,1,float16,float16,0,0.08929280042648316
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,40,2,128,1,float16,float16,0,0.10019840002059936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,40,4,128,1,float16,float16,0,0.13618240356445313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,40,8,128,1,float16,float16,0,0.18483200073242187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,40,1,128,1,float16,float16,0,0.053504002094268796
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,40,40,128,1,float16,float16,0,0.36085760593414307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,40,2,128,1,float16,float16,0,0.059443199634552
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,40,4,128,1,float16,float16,0,0.07367680072784424
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,40,8,128,1,float16,float16,0,0.08110079765319825
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,40,40,128,1,float16,float16,0,0.17044479846954347
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,40,1,128,1,float16,float16,0,0.037376001477241516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,40,4,128,1,float16,float16,0,0.04480000138282776
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,40,2,128,1,float16,float16,0,0.037478399276733396
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,40,8,128,1,float16,float16,0,0.048537600040435794
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,40,40,128,1,float16,float16,0,0.09333760142326356
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,40,1,128,1,float16,float16,0,0.02903040051460266
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,40,2,128,1,float16,float16,0,0.028518399596214293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,40,8,128,1,float16,float16,0,0.03035680055618286
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,40,4,128,1,float16,float16,0,0.02908160090446472
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,40,1,128,1,float16,float16,0,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,40,40,128,1,float16,float16,0,0.052019202709198
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,40,4,128,1,float16,float16,0,0.020531199872493744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,40,2,128,1,float16,float16,0,0.02001439929008484
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,40,8,128,1,float16,float16,0,0.02073120027780533
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,40,40,128,1,float16,float16,0,0.03409920036792755
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,40,1,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,40,4,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,40,2,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,40,8,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,40,40,128,1,float16,float16,0,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,40,1,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,40,2,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,40,4,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,40,8,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,40,40,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,40,1,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,40,2,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,40,4,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,40,8,128,1,float16,float16,0,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,40,40,128,1,float16,float16,0,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,40,1,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,40,2,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,40,4,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,40,8,128,1,float16,float16,0,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,32,1,128,1,float16,float16,0,38.62691955566406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,32,2,128,1,float16,float16,0,38.48355712890625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,32,4,128,1,float16,float16,0,39.470028686523435
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,32,32,128,1,float16,float16,0,43.22104187011719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,32,32,128,1,float16,float16,0,20.922108459472657
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,32,1,128,1,float16,float16,0,79.36854858398438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,32,2,128,1,float16,float16,0,19.419801330566408
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,32,1,128,1,float16,float16,0,19.2468994140625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,32,2,128,1,float16,float16,0,79.20281372070312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,32,4,128,1,float16,float16,0,80.52469482421876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,32,8,128,1,float16,float16,0,82.5607666015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,32,8,128,1,float16,float16,0,40.05186462402344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,32,1,128,1,float16,float16,0,9.429555511474609
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,32,32,128,1,float16,float16,0,10.457292938232422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,32,2,128,1,float16,float16,0,9.454182434082032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,32,4,128,1,float16,float16,0,9.669273376464844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,32,4,128,1,float16,float16,0,19.382322692871092
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,32,8,128,1,float16,float16,0,10.07104034423828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,32,8,128,1,float16,float16,0,19.854591369628906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,32,1,128,1,float16,float16,0,21.681254577636718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,32,32,128,1,float16,float16,0,24.554644775390624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,32,2,128,1,float16,float16,0,21.83961639404297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,32,4,128,1,float16,float16,0,22.757171630859375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,32,1,128,1,float16,float16,0,44.51328125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,32,2,128,1,float16,float16,0,44.730316162109375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,32,4,128,1,float16,float16,0,45.24897155761719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,32,8,128,1,float16,float16,0,46.73141784667969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,32,1,128,1,float16,float16,0,10.685798645019531
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,32,32,128,1,float16,float16,0,12.159744262695312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,32,2,128,1,float16,float16,0,10.800281524658203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,32,4,128,1,float16,float16,0,10.895104217529298
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,32,8,128,1,float16,float16,0,22.728855895996094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,32,32,128,1,float16,float16,0,6.056550216674805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,32,1,128,1,float16,float16,0,5.310464096069336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,32,2,128,1,float16,float16,0,5.265049743652344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,32,8,128,1,float16,float16,0,11.109375762939454
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,32,4,128,1,float16,float16,0,5.233203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,32,8,128,1,float16,float16,0,5.271091079711914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,32,2,128,1,float16,float16,0,15.156378173828125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,32,1,128,1,float16,float16,0,15.43572540283203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,32,32,128,1,float16,float16,0,17.55663299560547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,32,4,128,1,float16,float16,0,15.736781311035156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,32,1,128,1,float16,float16,0,30.950601196289064
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,32,2,128,1,float16,float16,0,30.551449584960938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,32,4,128,1,float16,float16,0,31.5272705078125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,32,8,128,1,float16,float16,0,32.611532592773436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,32,1,128,1,float16,float16,0,7.379046630859375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,32,32,128,1,float16,float16,0,8.754892730712891
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,32,2,128,1,float16,float16,0,7.353241729736328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,32,8,128,1,float16,float16,0,15.92821807861328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,32,4,128,1,float16,float16,0,7.543193817138672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,32,32,128,1,float16,float16,0,4.253696060180664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,32,1,128,1,float16,float16,0,3.4713088989257814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,32,8,128,1,float16,float16,0,7.686758422851563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,32,2,128,1,float16,float16,0,3.5866622924804688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,32,4,128,1,float16,float16,0,3.7350910186767576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,32,8,128,1,float16,float16,0,3.748659133911133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,32,1,128,1,float16,float16,0,20.0716796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,32,2,128,1,float16,float16,0,19.96037139892578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,32,4,128,1,float16,float16,0,20.5501953125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,32,32,128,1,float16,float16,0,23.543244934082033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,32,1,128,1,float16,float16,0,39.963442993164065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,32,2,128,1,float16,float16,0,40.59729919433594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,32,4,128,1,float16,float16,0,41.85958251953125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,32,8,128,1,float16,float16,0,42.302362060546876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,32,1,128,1,float16,float16,0,9.594009399414062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,32,2,128,1,float16,float16,0,9.90417938232422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,32,32,128,1,float16,float16,0,11.524658966064454
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,32,8,128,1,float16,float16,0,21.3749755859375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,32,4,128,1,float16,float16,0,10.062335968017578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,32,32,128,1,float16,float16,0,5.795737457275391
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,32,1,128,1,float16,float16,0,4.6181377410888675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,32,8,128,1,float16,float16,0,10.583602905273438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,32,2,128,1,float16,float16,0,4.760934448242187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,32,4,128,1,float16,float16,0,4.832153701782227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,32,8,128,1,float16,float16,0,4.867635345458984
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,32,32,128,1,float16,float16,0,2.8713424682617186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,32,1,128,1,float16,float16,0,2.295859146118164
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,32,2,128,1,float16,float16,0,2.3055871963500976
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,32,8,128,1,float16,float16,0,2.4300975799560547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,32,4,128,1,float16,float16,0,2.3254016876220702
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,32,1,128,1,float16,float16,0,11.011430358886718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,32,2,128,1,float16,float16,0,11.535001373291015
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,32,4,128,1,float16,float16,0,11.592294311523437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,32,1,128,1,float16,float16,0,22.601011657714842
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,32,32,128,1,float16,float16,0,13.872998046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,32,2,128,1,float16,float16,0,23.099136352539062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,32,4,128,1,float16,float16,0,23.59198760986328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,32,8,128,1,float16,float16,0,24.547225952148438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,32,1,128,1,float16,float16,0,5.253939056396485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,32,2,128,1,float16,float16,0,5.483827209472656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,32,32,128,1,float16,float16,0,3.3302017211914063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,32,4,128,1,float16,float16,0,5.694822311401367
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,32,8,128,1,float16,float16,0,5.816115188598633
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,32,32,128,1,float16,float16,0,6.903961944580078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,32,8,128,1,float16,float16,0,12.031897735595702
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,32,1,128,1,float16,float16,0,2.6798080444335937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,32,1,128,1,float16,float16,0,1.3634559631347656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,32,32,128,1,float16,float16,0,1.7275392532348632
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,32,2,128,1,float16,float16,0,1.3388799667358398
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,32,2,128,1,float16,float16,0,2.580326461791992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,32,4,128,1,float16,float16,0,2.6528255462646486
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,32,8,128,1,float16,float16,0,2.785331153869629
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,32,4,128,1,float16,float16,0,1.3722623825073241
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,32,8,128,1,float16,float16,0,1.4231040000915527
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,32,1,128,1,float16,float16,0,10.287872314453125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,32,2,128,1,float16,float16,0,10.345932769775391
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,32,32,128,1,float16,float16,0,13.68775634765625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,32,4,128,1,float16,float16,0,10.985215759277343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,32,1,128,1,float16,float16,0,20.829592895507812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,32,2,128,1,float16,float16,0,21.70496063232422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,32,4,128,1,float16,float16,0,22.299288940429687
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,32,8,128,1,float16,float16,0,23.29932098388672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,32,32,128,1,float16,float16,0,6.762342071533203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,32,1,128,1,float16,float16,0,5.0844158172607425
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,32,2,128,1,float16,float16,0,5.046732711791992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,32,4,128,1,float16,float16,0,5.2003326416015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,32,8,128,1,float16,float16,0,11.022489929199219
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,32,32,128,1,float16,float16,0,3.364249420166016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,32,8,128,1,float16,float16,0,5.6463359832763675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,32,1,128,1,float16,float16,0,2.4044031143188476
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,32,2,128,1,float16,float16,0,2.4229888916015625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,32,4,128,1,float16,float16,0,2.442956733703613
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,32,8,128,1,float16,float16,0,2.627529525756836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,32,32,128,1,float16,float16,0,1.7126399993896484
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,32,2,128,1,float16,float16,0,1.2430335998535156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,32,1,128,1,float16,float16,0,1.2127743721008302
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,32,4,128,1,float16,float16,0,1.2730367660522461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,32,8,128,1,float16,float16,0,1.3273088455200195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,32,32,128,1,float16,float16,0,0.8720895767211914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,32,1,128,1,float16,float16,0,0.6330368041992187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,32,2,128,1,float16,float16,0,0.6614528179168702
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,32,4,128,1,float16,float16,0,0.6441472053527832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,32,8,128,1,float16,float16,0,0.6661119937896729
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,32,1,128,1,float16,float16,0,11.844096374511718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,32,2,128,1,float16,float16,0,12.471398162841798
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,32,1,128,1,float16,float16,0,5.508659362792969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,32,4,128,1,float16,float16,0,12.876031494140625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,32,2,128,1,float16,float16,0,5.705363082885742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,32,4,128,1,float16,float16,0,6.127718353271485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,32,32,128,1,float16,float16,0,8.366438293457032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,32,8,128,1,float16,float16,0,13.71136016845703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,32,32,128,1,float16,float16,0,4.194865417480469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,32,1,128,1,float16,float16,0,2.7437055587768553
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,32,4,128,1,float16,float16,0,2.9447168350219726
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,32,2,128,1,float16,float16,0,2.7968000411987304
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,32,8,128,1,float16,float16,0,6.562713623046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,32,8,128,1,float16,float16,0,3.1615999221801756
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,32,32,128,1,float16,float16,0,2.0947967529296876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,32,1,128,1,float16,float16,0,1.4240768432617188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,32,2,128,1,float16,float16,0,1.3896191596984864
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,32,4,128,1,float16,float16,0,1.4929408073425292
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,32,8,128,1,float16,float16,0,1.5537664413452148
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,32,32,128,1,float16,float16,0,1.061631965637207
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,32,1,128,1,float16,float16,0,0.7249407768249512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,32,2,128,1,float16,float16,0,0.7291391849517822
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,32,4,128,1,float16,float16,0,0.744755220413208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,32,8,128,1,float16,float16,0,0.8073216438293457
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,32,1,128,1,float16,float16,0,0.38712320327758787
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,32,32,128,1,float16,float16,0,0.552396821975708
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,32,2,128,1,float16,float16,0,0.3968480110168457
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,32,4,128,1,float16,float16,0,0.40458240509033205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,32,8,128,1,float16,float16,0,0.42132000923156737
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,32,1,128,1,float16,float16,0,11.215103912353516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,32,2,128,1,float16,float16,0,11.732787322998046
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,32,1,128,1,float16,float16,0,5.294591903686523
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,32,2,128,1,float16,float16,0,5.614540863037109
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,32,4,128,1,float16,float16,0,5.949388885498047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,32,32,128,1,float16,float16,0,8.766407775878907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,32,4,128,1,float16,float16,0,12.511385345458985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,32,8,128,1,float16,float16,0,13.436773681640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,32,1,128,1,float16,float16,0,2.541823959350586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,32,2,128,1,float16,float16,0,2.592153549194336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,32,8,128,1,float16,float16,0,6.439218902587891
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,32,32,128,1,float16,float16,0,4.352511978149414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,32,4,128,1,float16,float16,0,2.812723159790039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,32,32,128,1,float16,float16,0,2.2359039306640627
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,32,8,128,1,float16,float16,0,3.033139228820801
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,32,1,128,1,float16,float16,0,1.3351424217224122
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,32,2,128,1,float16,float16,0,1.3137408256530763
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,32,4,128,1,float16,float16,0,1.3716992378234862
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,32,8,128,1,float16,float16,0,1.5219712257385254
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,32,32,128,1,float16,float16,0,1.132595157623291
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,32,1,128,1,float16,float16,0,0.6798336029052734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,32,4,128,1,float16,float16,0,0.7206895828247071
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,32,2,128,1,float16,float16,0,0.7012847900390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,32,8,128,1,float16,float16,0,0.7799759864807129
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,32,32,128,1,float16,float16,0,0.5753344058990478
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,32,8,128,1,float16,float16,0,0.375654411315918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,32,4,128,1,float16,float16,0,0.3546623945236206
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,32,1,128,1,float16,float16,0,0.3469280004501343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,32,2,128,1,float16,float16,0,0.34554879665374755
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,32,32,128,1,float16,float16,0,0.2579967975616455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,32,1,128,1,float16,float16,0,0.20402240753173828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,32,2,128,1,float16,float16,0,0.20464639663696288
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,32,4,128,1,float16,float16,0,0.21288959980010985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,32,8,128,1,float16,float16,0,0.21683199405670167
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,32,1,128,1,float16,float16,0,3.0848512649536133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,32,1,128,1,float16,float16,0,6.704122924804688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,32,32,128,1,float16,float16,0,5.730867385864258
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,32,2,128,1,float16,float16,0,6.931455993652344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,32,4,128,1,float16,float16,0,7.2535041809082035
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,32,8,128,1,float16,float16,0,7.7972465515136715
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,32,2,128,1,float16,float16,0,3.2125953674316405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,32,4,128,1,float16,float16,0,3.4338817596435547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,32,8,128,1,float16,float16,0,3.7856239318847655
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,32,1,128,1,float16,float16,0,1.5725567817687989
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,32,2,128,1,float16,float16,0,1.579520034790039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,32,32,128,1,float16,float16,0,2.8791296005249025
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,32,4,128,1,float16,float16,0,1.6592384338378907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,32,8,128,1,float16,float16,0,1.8559999465942383
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,32,32,128,1,float16,float16,0,1.437235164642334
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,32,1,128,1,float16,float16,0,0.7878143787384033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,32,2,128,1,float16,float16,0,0.8130047798156739
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,32,4,128,1,float16,float16,0,0.8748543739318848
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,32,8,128,1,float16,float16,0,0.9343999862670899
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,32,32,128,1,float16,float16,0,0.7487472057342529
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,32,1,128,1,float16,float16,0,0.4099584102630615
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,32,2,128,1,float16,float16,0,0.42342400550842285
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,32,4,128,1,float16,float16,0,0.4430848121643066
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,32,8,128,1,float16,float16,0,0.4777872085571289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,32,32,128,1,float16,float16,0,0.3812351942062378
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,32,1,128,1,float16,float16,0,0.22016000747680664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,32,2,128,1,float16,float16,0,0.22768640518188477
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,32,4,128,1,float16,float16,0,0.22978560924530028
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,32,8,128,1,float16,float16,0,0.23597118854522706
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,32,1,128,1,float16,float16,0,0.13178880214691163
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,32,32,128,1,float16,float16,0,0.1617408037185669
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,32,2,128,1,float16,float16,0,0.13537280559539794
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,32,4,128,1,float16,float16,0,0.1392639994621277
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,32,8,128,1,float16,float16,0,0.1425920009613037
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,32,1,128,1,float16,float16,0,3.0585344314575194
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,32,1,128,1,float16,float16,0,6.552063751220703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,32,2,128,1,float16,float16,0,6.4648689270019535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,32,4,128,1,float16,float16,0,6.9607421875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,32,32,128,1,float16,float16,0,6.353100967407227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,32,8,128,1,float16,float16,0,7.9265281677246096
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,32,2,128,1,float16,float16,0,3.2076736450195313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,32,4,128,1,float16,float16,0,3.448729705810547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,32,2,128,1,float16,float16,0,1.5791104316711426
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,32,1,128,1,float16,float16,0,1.488588809967041
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,32,8,128,1,float16,float16,0,3.8649856567382814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,32,32,128,1,float16,float16,0,3.260006332397461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,32,4,128,1,float16,float16,0,1.6886272430419922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,32,8,128,1,float16,float16,0,1.9474431991577148
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,32,32,128,1,float16,float16,0,1.639014434814453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,32,1,128,1,float16,float16,0,0.7800320148468017
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,32,2,128,1,float16,float16,0,0.7839744091033936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,32,4,128,1,float16,float16,0,0.862822437286377
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,32,32,128,1,float16,float16,0,0.9087488174438476
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,32,8,128,1,float16,float16,0,0.9456128120422364
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,32,1,128,1,float16,float16,0,0.4024831771850586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,32,2,128,1,float16,float16,0,0.4191232204437256
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,32,4,128,1,float16,float16,0,0.4446720123291016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,32,8,128,1,float16,float16,0,0.4921855926513672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,32,32,128,1,float16,float16,0,0.42772479057312013
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,32,1,128,1,float16,float16,0,0.20106239318847657
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,32,4,128,1,float16,float16,0,0.21125121116638185
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,32,2,128,1,float16,float16,0,0.21206560134887695
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,32,8,128,1,float16,float16,0,0.23434240818023683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,32,1,128,1,float16,float16,0,0.11729919910430908
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,32,32,128,1,float16,float16,0,0.17367039918899535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,32,2,128,1,float16,float16,0,0.12339199781417846
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,32,4,128,1,float16,float16,0,0.12590080499649048
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,32,8,128,1,float16,float16,0,0.13260480165481567
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,32,2,128,1,float16,float16,0,0.07854080200195312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,32,1,128,1,float16,float16,0,0.07301120162010193
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,32,32,128,1,float16,float16,0,0.10132479667663574
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,32,8,128,1,float16,float16,0,0.08806399703025818
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,32,4,128,1,float16,float16,0,0.0889855980873108
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,32,1,128,1,float16,float16,0,4.009830474853516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,32,2,128,1,float16,float16,0,4.105984115600586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,32,1,128,1,float16,float16,0,1.919331169128418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,32,2,128,1,float16,float16,0,2.0057088851928713
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,32,4,128,1,float16,float16,0,4.660070419311523
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,32,4,128,1,float16,float16,0,2.413260841369629
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,32,8,128,1,float16,float16,0,5.75016975402832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,32,32,128,1,float16,float16,0,6.290687942504883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,32,8,128,1,float16,float16,0,2.9435871124267576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,32,1,128,1,float16,float16,0,0.9709568023681641
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,32,2,128,1,float16,float16,0,1.0080767631530763
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,32,32,128,1,float16,float16,0,3.228006362915039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,32,4,128,1,float16,float16,0,1.204428768157959
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,32,8,128,1,float16,float16,0,1.4667776107788086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,32,1,128,1,float16,float16,0,0.506982421875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,32,32,128,1,float16,float16,0,1.7442815780639649
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,32,2,128,1,float16,float16,0,0.527462387084961
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,32,4,128,1,float16,float16,0,0.5645823955535889
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,32,32,128,1,float16,float16,0,0.8174079895019531
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,32,8,128,1,float16,float16,0,0.7825920104980468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,32,2,128,1,float16,float16,0,0.28574719429016116
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,32,1,128,1,float16,float16,0,0.2731008052825928
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,32,4,128,1,float16,float16,0,0.2927615880966187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,32,8,128,1,float16,float16,0,0.37160959243774416
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,32,32,128,1,float16,float16,0,0.37893118858337405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,32,1,128,1,float16,float16,0,0.12974079847335815
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,32,2,128,1,float16,float16,0,0.13624320030212403
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,32,4,128,1,float16,float16,0,0.13890559673309327
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,32,32,128,1,float16,float16,0,0.13306879997253418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,32,8,128,1,float16,float16,0,0.1532415986061096
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,32,1,128,1,float16,float16,0,0.0764415979385376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,32,2,128,1,float16,float16,0,0.07987200021743775
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,32,4,128,1,float16,float16,0,0.08478720188140869
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,32,32,128,1,float16,float16,0,0.07178239822387696
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,32,8,128,1,float16,float16,0,0.09000959992408752
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,32,1,128,1,float16,float16,0,0.04997119903564453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,32,2,128,1,float16,float16,0,0.052275198698043826
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,32,4,128,1,float16,float16,0,0.051148802042007446
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,32,8,128,1,float16,float16,0,0.05432320237159729
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,32,32,128,1,float16,float16,0,0.04377599954605103
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,32,1,128,1,float16,float16,0,0.03399679958820343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,32,2,128,1,float16,float16,0,0.03409920036792755
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,32,4,128,1,float16,float16,0,0.035123199224472046
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,32,8,128,1,float16,float16,0,0.03527680039405823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,32,1,128,1,float16,float16,0,1.8295808792114259
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,32,2,128,1,float16,float16,0,1.9933696746826173
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,32,4,128,1,float16,float16,0,2.3281152725219725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,32,8,128,1,float16,float16,0,2.93493766784668
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,32,32,128,1,float16,float16,0,3.1510528564453124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,32,1,128,1,float16,float16,0,0.9423871994018554
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,32,2,128,1,float16,float16,0,0.9851391792297364
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,32,4,128,1,float16,float16,0,1.1578880310058595
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,32,8,128,1,float16,float16,0,1.4768128395080566
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,32,1,128,1,float16,float16,0,0.48563199043273925
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,32,32,128,1,float16,float16,0,1.6131536483764648
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,32,2,128,1,float16,float16,0,0.5270527839660645
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,32,4,128,1,float16,float16,0,0.6116352081298828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,32,8,128,1,float16,float16,0,0.7195136070251464
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,32,32,128,1,float16,float16,0,0.8134143829345704
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,32,1,128,1,float16,float16,0,0.2332672119140625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,32,2,128,1,float16,float16,0,0.2505215883255005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,32,4,128,1,float16,float16,0,0.2613248109817505
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,32,8,128,1,float16,float16,0,0.35256319046020507
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,32,1,128,1,float16,float16,0,0.09410560131072998
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,32,32,128,1,float16,float16,0,0.3689471960067749
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,32,2,128,1,float16,float16,0,0.09707520008087159
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,32,4,128,1,float16,float16,0,0.10193920135498047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,32,8,128,1,float16,float16,0,0.11622400283813476
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,32,32,128,1,float16,float16,0,0.1087488055229187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,32,1,128,1,float16,float16,0,0.056627202033996585
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,32,2,128,1,float16,float16,0,0.057651197910308837
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,32,4,128,1,float16,float16,0,0.0611840009689331
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,32,8,128,1,float16,float16,0,0.0662015974521637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,32,1,128,1,float16,float16,0,0.038089600205421445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,32,32,128,1,float16,float16,0,0.05857279896736145
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,32,2,128,1,float16,float16,0,0.03850240111351013
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,32,4,128,1,float16,float16,0,0.039526399970054624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,32,8,128,1,float16,float16,0,0.04213759899139404
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,32,32,128,1,float16,float16,0,0.03594239950180054
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,32,1,128,1,float16,float16,0,0.026368001103401185
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,32,2,128,1,float16,float16,0,0.026111999154090883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,32,4,128,1,float16,float16,0,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,32,8,128,1,float16,float16,0,0.027801600098609925
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,32,32,128,1,float16,float16,0,0.028262400627136232
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,32,1,128,1,float16,float16,0,0.024831999838352204
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,32,2,128,1,float16,float16,0,0.023347200453281404
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,32,8,128,1,float16,float16,0,0.02391040027141571
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,32,4,128,1,float16,float16,0,0.023654399812221526
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,32,2,128,1,float16,float16,0,1.0037247657775878
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,32,1,128,1,float16,float16,0,0.9315823554992676
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,32,4,128,1,float16,float16,0,1.1643391609191895
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,32,8,128,1,float16,float16,0,1.4474752426147461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,32,32,128,1,float16,float16,0,1.579263973236084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,32,1,128,1,float16,float16,0,0.466431999206543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,32,2,128,1,float16,float16,0,0.4867584228515625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,32,4,128,1,float16,float16,0,0.5632512092590332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,32,8,128,1,float16,float16,0,0.724838399887085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,32,1,128,1,float16,float16,0,0.23301119804382325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,32,2,128,1,float16,float16,0,0.24519679546356202
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,32,32,128,1,float16,float16,0,0.8025088310241699
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,32,4,128,1,float16,float16,0,0.27648000717163085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,32,8,128,1,float16,float16,0,0.33653759956359863
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,32,32,128,1,float16,float16,0,0.3634687900543213
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,32,1,128,1,float16,float16,0,0.07541760206222534
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,32,2,128,1,float16,float16,0,0.07874559760093688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,32,4,128,1,float16,float16,0,0.08391680121421814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,32,32,128,1,float16,float16,0,0.10496000051498414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,32,8,128,1,float16,float16,0,0.10060800313949585
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,32,1,128,1,float16,float16,0,0.046587198972702026
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,32,2,128,1,float16,float16,0,0.04756479859352112
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,32,4,128,1,float16,float16,0,0.05027840137481689
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,32,8,128,1,float16,float16,0,0.05605599880218506
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,32,32,128,1,float16,float16,0,0.05186560153961182
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,32,1,128,1,float16,float16,0,0.031795200705528257
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,32,2,128,1,float16,float16,0,0.031488001346588135
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,32,8,128,1,float16,float16,0,0.03594239950180054
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,32,4,128,1,float16,float16,0,0.033024001121521
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,32,32,128,1,float16,float16,0,0.032051199674606325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,32,1,128,1,float16,float16,0,0.023193599283695222
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,32,4,128,1,float16,float16,0,0.02314240038394928
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,32,8,128,1,float16,float16,0,0.02451840043067932
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,32,2,128,1,float16,float16,0,0.02247679978609085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,32,32,128,1,float16,float16,0,0.024063999950885772
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,32,1,128,1,float16,float16,0,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,32,2,128,1,float16,float16,0,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,32,4,128,1,float16,float16,0,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,32,8,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,32,32,128,1,float16,float16,0,0.02088959962129593
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,32,1,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,32,2,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,32,4,128,1,float16,float16,0,0.01924159973859787
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,32,8,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,32,1,128,1,float16,float16,0,0.4488704204559326
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,32,2,128,1,float16,float16,0,0.49295358657836913
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,32,4,128,1,float16,float16,0,0.5558239936828613
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,32,8,128,1,float16,float16,0,0.7241216182708741
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,32,32,128,1,float16,float16,0,0.7875072002410889
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,32,4,128,1,float16,float16,0,0.2618288040161133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,32,2,128,1,float16,float16,0,0.23347198963165283
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,32,1,128,1,float16,float16,0,0.21821439266204834
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,32,8,128,1,float16,float16,0,0.3309056043624878
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,32,32,128,1,float16,float16,0,0.356710410118103
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,32,1,128,1,float16,float16,0,0.07531520128250122
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,32,2,128,1,float16,float16,0,0.0775168001651764
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,32,4,128,1,float16,float16,0,0.08478720188140869
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,32,8,128,1,float16,float16,0,0.09953280091285706
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,32,1,128,1,float16,float16,0,0.046028798818588255
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,32,32,128,1,float16,float16,0,0.12349439859390259
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,32,2,128,1,float16,float16,0,0.048179200291633605
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,32,4,128,1,float16,float16,0,0.04981760084629059
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,32,32,128,1,float16,float16,0,0.06318079829216003
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,32,8,128,1,float16,float16,0,0.05539839863777161
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,32,1,128,1,float16,float16,0,0.030720001459121703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,32,4,128,1,float16,float16,0,0.032204800844192506
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,32,2,128,1,float16,float16,0,0.030668801069259642
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,32,8,128,1,float16,float16,0,0.03476319909095764
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,32,32,128,1,float16,float16,0,0.03758080005645752
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,32,1,128,1,float16,float16,0,0.02247679978609085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,32,2,128,1,float16,float16,0,0.023347200453281404
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,32,4,128,1,float16,float16,0,0.022732800245285033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,32,8,128,1,float16,float16,0,0.023856000602245332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,32,32,128,1,float16,float16,0,0.02462719976902008
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,32,1,128,1,float16,float16,0,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,32,2,128,1,float16,float16,0,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,32,4,128,1,float16,float16,0,0.01966080069541931
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,32,8,128,1,float16,float16,0,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,32,32,128,1,float16,float16,0,0.020582400262355804
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,32,1,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,32,2,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,32,4,128,1,float16,float16,0,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,32,8,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,32,32,128,1,float16,float16,0,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,32,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,32,2,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,32,4,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,32,8,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,32,1,128,1,float16,float16,0,0.21882879734039307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,32,2,128,1,float16,float16,0,0.23045120239257813
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,32,8,128,1,float16,float16,0,0.3206144094467163
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,32,4,128,1,float16,float16,0,0.2581952095031738
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,32,32,128,1,float16,float16,0,0.40314879417419436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,32,1,128,1,float16,float16,0,0.07480319738388061
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,32,2,128,1,float16,float16,0,0.07869439721107482
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,32,4,128,1,float16,float16,0,0.08320000171661376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,32,8,128,1,float16,float16,0,0.09937919974327088
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,32,32,128,1,float16,float16,0,0.1680896043777466
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,32,1,128,1,float16,float16,0,0.045516800880432126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,32,4,128,1,float16,float16,0,0.04976319968700409
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,32,2,128,1,float16,float16,0,0.04776960015296936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,32,32,128,1,float16,float16,0,0.09011200070381165
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,32,8,128,1,float16,float16,0,0.05493760108947754
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,32,1,128,1,float16,float16,0,0.03061760067939758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,32,2,128,1,float16,float16,0,0.030720001459121703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,32,4,128,1,float16,float16,0,0.031892800331115724
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,32,8,128,1,float16,float16,0,0.03440159857273102
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,32,32,128,1,float16,float16,0,0.051046401262283325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,32,4,128,1,float16,float16,0,0.022681599855422972
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,32,2,128,1,float16,float16,0,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,32,8,128,1,float16,float16,0,0.024063999950885772
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,32,1,128,1,float16,float16,0,0.021958400309085847
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,32,32,128,1,float16,float16,0,0.03261440098285675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,32,1,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,32,2,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,32,4,128,1,float16,float16,0,0.019763199985027312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,32,32,128,1,float16,float16,0,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,32,8,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,32,1,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,32,4,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,32,2,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,32,8,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,32,32,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,32,1,128,1,float16,float16,0,0.01858240067958832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,32,2,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,32,4,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,32,8,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,32,32,128,1,float16,float16,0,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,32,2,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,32,1,128,1,float16,float16,0,0.018220800161361694
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,32,8,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,32,4,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,32,1,128,1,float16,float16,0,0.07704799771308898
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,32,2,128,1,float16,float16,0,0.07859200239181519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,32,8,128,1,float16,float16,0,0.14888960123062134
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,32,4,128,1,float16,float16,0,0.08360959887504578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,32,1,128,1,float16,float16,0,0.04628480076789856
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,32,32,128,1,float16,float16,0,0.263321590423584
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,32,2,128,1,float16,float16,0,0.04746240079402923
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,32,4,128,1,float16,float16,0,0.050329601764678954
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,32,8,128,1,float16,float16,0,0.08104959726333619
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,32,32,128,1,float16,float16,0,0.13634560108184815
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,32,1,128,1,float16,float16,0,0.030873599648475646
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,32,2,128,1,float16,float16,0,0.030822399258613586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,32,4,128,1,float16,float16,0,0.032204800844192506
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,32,8,128,1,float16,float16,0,0.04736000001430511
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,32,32,128,1,float16,float16,0,0.07772160172462464
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,32,2,128,1,float16,float16,0,0.022521600127220154
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,32,1,128,1,float16,float16,0,0.02222079932689667
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,32,4,128,1,float16,float16,0,0.02263039946556091
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,32,8,128,1,float16,float16,0,0.029747200012207032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,32,32,128,1,float16,float16,0,0.04398080110549927
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,32,1,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,32,4,128,1,float16,float16,0,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,32,8,128,1,float16,float16,0,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,32,2,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,32,32,128,1,float16,float16,0,0.027340799570083618
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,32,1,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,32,2,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,32,4,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,32,8,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,32,32,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,32,2,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,32,4,128,1,float16,float16,0,0.019808000326156615
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,32,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,32,8,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,32,32,128,1,float16,float16,0,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,32,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,32,2,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,32,4,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,32,8,128,1,float16,float16,0,0.01806560009717941
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,32,32,128,1,float16,float16,0,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,32,2,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,32,1,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,32,4,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,32,8,128,1,float16,float16,0,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,24,1,128,1,float16,float16,0,28.594638061523437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,24,2,128,1,float16,float16,0,29.055743408203124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,24,4,128,1,float16,float16,0,29.422744750976562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,24,24,128,1,float16,float16,0,32.377752685546874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,24,1,128,1,float16,float16,0,60.04638671875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,24,2,128,1,float16,float16,0,59.74220581054688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,24,24,128,1,float16,float16,0,15.8708740234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,24,4,128,1,float16,float16,0,60.894921875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,24,1,128,1,float16,float16,0,14.1712890625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,24,2,128,1,float16,float16,0,14.544740295410156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,24,8,128,1,float16,float16,0,61.31885986328125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,24,8,128,1,float16,float16,0,30.036785888671876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,24,1,128,1,float16,float16,0,7.084595489501953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,24,24,128,1,float16,float16,0,7.837849426269531
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,24,2,128,1,float16,float16,0,7.011993408203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,24,8,128,1,float16,float16,0,7.417036437988282
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,24,4,128,1,float16,float16,0,7.270654296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,24,4,128,1,float16,float16,0,14.924134826660156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,24,8,128,1,float16,float16,0,15.092735290527344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,24,2,128,1,float16,float16,0,16.409703063964844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,24,1,128,1,float16,float16,0,16.315084838867186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,24,24,128,1,float16,float16,0,18.61565399169922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,24,4,128,1,float16,float16,0,17.0355712890625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,24,1,128,1,float16,float16,0,32.50370483398437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,24,2,128,1,float16,float16,0,33.079962158203124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,24,4,128,1,float16,float16,0,34.72133178710938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,24,8,128,1,float16,float16,0,34.84180603027344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,24,1,128,1,float16,float16,0,7.90645751953125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,24,2,128,1,float16,float16,0,7.8295036315917965
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,24,24,128,1,float16,float16,0,9.216409301757812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,24,4,128,1,float16,float16,0,7.946444702148438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,24,8,128,1,float16,float16,0,17.4466552734375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,24,24,128,1,float16,float16,0,4.38026237487793
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,24,1,128,1,float16,float16,0,3.926630401611328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,24,8,128,1,float16,float16,0,8.488960266113281
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,24,2,128,1,float16,float16,0,3.8086574554443358
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,24,4,128,1,float16,float16,0,3.9232513427734377
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,24,8,128,1,float16,float16,0,4.028108978271485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,24,24,128,1,float16,float16,0,12.955494689941407
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,24,1,128,1,float16,float16,0,11.366860961914062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,24,2,128,1,float16,float16,0,11.352320098876953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,24,1,128,1,float16,float16,0,22.755677795410158
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,24,4,128,1,float16,float16,0,11.865548706054687
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,24,2,128,1,float16,float16,0,23.356159973144532
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,24,4,128,1,float16,float16,0,24.148326110839843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,24,8,128,1,float16,float16,0,24.333106994628906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,24,1,128,1,float16,float16,0,5.2717536926269535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,24,24,128,1,float16,float16,0,6.34976806640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,24,2,128,1,float16,float16,0,5.5478271484375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,24,8,128,1,float16,float16,0,12.051660919189453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,24,24,128,1,float16,float16,0,3.1602176666259765
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,24,1,128,1,float16,float16,0,2.685798454284668
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,24,8,128,1,float16,float16,0,6.0062206268310545
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,24,4,128,1,float16,float16,0,5.466214370727539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,24,2,128,1,float16,float16,0,2.6720767974853517
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,24,4,128,1,float16,float16,0,2.727372741699219
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,24,8,128,1,float16,float16,0,2.7923456192016602
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,24,24,128,1,float16,float16,0,18.12024383544922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,24,1,128,1,float16,float16,0,14.714060974121093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,24,2,128,1,float16,float16,0,14.998477172851562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,24,4,128,1,float16,float16,0,15.425431823730468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,24,1,128,1,float16,float16,0,29.927374267578124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,24,2,128,1,float16,float16,0,30.80693664550781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,24,4,128,1,float16,float16,0,31.359231567382814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,24,8,128,1,float16,float16,0,32.75054016113281
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,24,1,128,1,float16,float16,0,7.152333068847656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,24,24,128,1,float16,float16,0,8.778240203857422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,24,2,128,1,float16,float16,0,7.155916595458985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,24,4,128,1,float16,float16,0,7.500032043457031
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,24,8,128,1,float16,float16,0,16.128353881835938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,24,24,128,1,float16,float16,0,4.305356979370117
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,24,8,128,1,float16,float16,0,7.782911682128907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,24,1,128,1,float16,float16,0,3.3655296325683595
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,24,2,128,1,float16,float16,0,3.487897491455078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,24,4,128,1,float16,float16,0,3.601766586303711
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,24,8,128,1,float16,float16,0,3.6337665557861327
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,24,1,128,1,float16,float16,0,1.7543664932250977
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,24,24,128,1,float16,float16,0,2.117580795288086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,24,2,128,1,float16,float16,0,1.7707008361816405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,24,4,128,1,float16,float16,0,1.7909759521484374
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,24,8,128,1,float16,float16,0,1.87325439453125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,24,1,128,1,float16,float16,0,8.393881225585938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,24,2,128,1,float16,float16,0,8.260403442382813
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,24,24,128,1,float16,float16,0,10.319974517822265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,24,1,128,1,float16,float16,0,17.215231323242186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,24,2,128,1,float16,float16,0,17.17550048828125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,24,4,128,1,float16,float16,0,8.805785369873046
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,24,4,128,1,float16,float16,0,17.824050903320312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,24,8,128,1,float16,float16,0,18.965863037109376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,24,1,128,1,float16,float16,0,4.0194049835205075
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,24,24,128,1,float16,float16,0,5.086771011352539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,24,2,128,1,float16,float16,0,4.0902656555175785
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,24,4,128,1,float16,float16,0,4.144785690307617
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,24,24,128,1,float16,float16,0,2.532454490661621
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,24,8,128,1,float16,float16,0,8.79974365234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,24,8,128,1,float16,float16,0,4.261116790771484
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,24,1,128,1,float16,float16,0,1.9974143981933594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,24,2,128,1,float16,float16,0,1.9538944244384766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,24,4,128,1,float16,float16,0,2.049740791320801
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,24,8,128,1,float16,float16,0,2.181011199951172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,24,24,128,1,float16,float16,0,1.306879997253418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,24,1,128,1,float16,float16,0,1.0894847869873048
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,24,2,128,1,float16,float16,0,1.0281472206115723
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,24,8,128,1,float16,float16,0,1.1160575866699218
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,24,4,128,1,float16,float16,0,1.0258943557739257
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,24,1,128,1,float16,float16,0,7.363430023193359
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,24,2,128,1,float16,float16,0,7.903948974609375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,24,24,128,1,float16,float16,0,10.398361968994141
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,24,4,128,1,float16,float16,0,8.305661010742188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,24,1,128,1,float16,float16,0,15.696794128417968
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,24,2,128,1,float16,float16,0,16.1586181640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,24,4,128,1,float16,float16,0,16.855859375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,24,8,128,1,float16,float16,0,17.674644470214844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,24,1,128,1,float16,float16,0,3.6413440704345703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,24,24,128,1,float16,float16,0,5.074835205078125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,24,4,128,1,float16,float16,0,3.932160186767578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,24,2,128,1,float16,float16,0,3.7242366790771486
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,24,8,128,1,float16,float16,0,8.486041259765624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,24,24,128,1,float16,float16,0,2.5662464141845702
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,24,8,128,1,float16,float16,0,4.186880111694336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,24,1,128,1,float16,float16,0,1.8087936401367188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,24,2,128,1,float16,float16,0,1.7790464401245116
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,24,4,128,1,float16,float16,0,1.9358720779418945
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,24,8,128,1,float16,float16,0,2.0495872497558594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,24,1,128,1,float16,float16,0,0.9365504264831543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,24,2,128,1,float16,float16,0,0.950169563293457
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,24,4,128,1,float16,float16,0,0.9762304306030274
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,24,24,128,1,float16,float16,0,1.2841456413269043
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,24,8,128,1,float16,float16,0,1.045248031616211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,24,24,128,1,float16,float16,0,0.662886381149292
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,24,2,128,1,float16,float16,0,0.5160960197448731
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,24,1,128,1,float16,float16,0,0.5021615982055664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,24,4,128,1,float16,float16,0,0.5185023784637451
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,24,8,128,1,float16,float16,0,0.5135359764099121
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,24,1,128,1,float16,float16,0,8.702361297607421
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,24,2,128,1,float16,float16,0,9.282201385498047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,24,4,128,1,float16,float16,0,9.73864974975586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,24,1,128,1,float16,float16,0,4.231321716308594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,24,2,128,1,float16,float16,0,4.498636627197266
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,24,4,128,1,float16,float16,0,4.536980819702149
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,24,24,128,1,float16,float16,0,6.2607872009277346
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,24,8,128,1,float16,float16,0,10.490982055664062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,24,24,128,1,float16,float16,0,3.117465591430664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,24,8,128,1,float16,float16,0,4.9810432434082035
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,24,1,128,1,float16,float16,0,2.0841983795166015
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,24,2,128,1,float16,float16,0,2.0913663864135743
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,24,4,128,1,float16,float16,0,2.2540800094604494
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,24,8,128,1,float16,float16,0,2.4573904037475587
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,24,24,128,1,float16,float16,0,1.5939071655273438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,24,1,128,1,float16,float16,0,1.0577407836914063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,24,2,128,1,float16,float16,0,1.073100757598877
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,24,4,128,1,float16,float16,0,1.1307519912719726
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,24,24,128,1,float16,float16,0,0.8110591888427734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,24,1,128,1,float16,float16,0,0.5428736209869385
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,24,2,128,1,float16,float16,0,0.5487103939056397
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,24,8,128,1,float16,float16,0,1.2332544326782227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,24,4,128,1,float16,float16,0,0.5739007949829101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,24,8,128,1,float16,float16,0,0.6239151954650879
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,24,24,128,1,float16,float16,0,0.4079103946685791
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,24,1,128,1,float16,float16,0,0.31385600566864014
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,24,2,128,1,float16,float16,0,0.3110399961471558
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,24,4,128,1,float16,float16,0,0.31651840209960935
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,24,8,128,1,float16,float16,0,0.33315839767456057
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,24,1,128,1,float16,float16,0,8.476822662353516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,24,2,128,1,float16,float16,0,8.88821792602539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,24,4,128,1,float16,float16,0,9.447219085693359
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,24,1,128,1,float16,float16,0,3.8988800048828125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,24,2,128,1,float16,float16,0,4.174540710449219
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,24,4,128,1,float16,float16,0,4.566372680664062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,24,24,128,1,float16,float16,0,6.490573120117188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,24,8,128,1,float16,float16,0,10.29345245361328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,24,24,128,1,float16,float16,0,3.3221630096435546
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,24,1,128,1,float16,float16,0,1.912063980102539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,24,2,128,1,float16,float16,0,2.014668846130371
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,24,8,128,1,float16,float16,0,4.975257492065429
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,24,4,128,1,float16,float16,0,2.166579246520996
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,24,24,128,1,float16,float16,0,1.672447967529297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,24,8,128,1,float16,float16,0,2.429030418395996
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,24,1,128,1,float16,float16,0,1.0003456115722655
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,24,2,128,1,float16,float16,0,1.0422783851623536
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,24,4,128,1,float16,float16,0,1.075712013244629
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,24,24,128,1,float16,float16,0,0.8374784469604493
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,24,1,128,1,float16,float16,0,0.5287936210632325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,24,2,128,1,float16,float16,0,0.5302783966064453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,24,4,128,1,float16,float16,0,0.5699584007263183
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,24,8,128,1,float16,float16,0,1.2254719734191895
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,24,8,128,1,float16,float16,0,0.6205440044403077
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,24,24,128,1,float16,float16,0,0.4369391918182373
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,24,1,128,1,float16,float16,0,0.2776576042175293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,24,2,128,1,float16,float16,0,0.28339200019836425
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,24,4,128,1,float16,float16,0,0.29624319076538086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,24,8,128,1,float16,float16,0,0.29506559371948243
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,24,24,128,1,float16,float16,0,0.19415040016174318
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,24,2,128,1,float16,float16,0,0.16993279457092286
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,24,1,128,1,float16,float16,0,0.16870239973068238
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,24,4,128,1,float16,float16,0,0.17914880514144899
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,24,8,128,1,float16,float16,0,0.18339840173721314
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,24,1,128,1,float16,float16,0,4.7851520538330075
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,24,2,128,1,float16,float16,0,5.083750534057617
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,24,4,128,1,float16,float16,0,5.366374588012695
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,24,1,128,1,float16,float16,0,2.317977523803711
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,24,24,128,1,float16,float16,0,4.248831939697266
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,24,8,128,1,float16,float16,0,6.122905731201172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,24,2,128,1,float16,float16,0,2.4281600952148437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,24,4,128,1,float16,float16,0,2.59368953704834
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,24,24,128,1,float16,float16,0,2.1317632675170897
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,24,8,128,1,float16,float16,0,2.931967926025391
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,24,2,128,1,float16,float16,0,1.223526382446289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,24,1,128,1,float16,float16,0,1.1570688247680665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,24,4,128,1,float16,float16,0,1.3323264122009277
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,24,8,128,1,float16,float16,0,1.5072256088256837
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,24,24,128,1,float16,float16,0,1.0880975723266602
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,24,1,128,1,float16,float16,0,0.603443193435669
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,24,2,128,1,float16,float16,0,0.6184447765350342
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,24,4,128,1,float16,float16,0,0.6696447849273681
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,24,24,128,1,float16,float16,0,0.5529088020324707
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,24,1,128,1,float16,float16,0,0.30899200439453123
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,24,2,128,1,float16,float16,0,0.3110912084579468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,24,8,128,1,float16,float16,0,0.7580160140991211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,24,4,128,1,float16,float16,0,0.3304960012435913
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,24,8,128,1,float16,float16,0,0.38046720027923586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,24,24,128,1,float16,float16,0,0.27053279876708985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,24,1,128,1,float16,float16,0,0.17095680236816407
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,24,2,128,1,float16,float16,0,0.1837056040763855
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,24,4,128,1,float16,float16,0,0.18887679576873778
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,24,24,128,1,float16,float16,0,0.13496320247650145
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,24,8,128,1,float16,float16,0,0.19322880506515502
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,24,1,128,1,float16,float16,0,0.116428804397583
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,24,2,128,1,float16,float16,0,0.11811840534210205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,24,4,128,1,float16,float16,0,0.11939840316772461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,24,8,128,1,float16,float16,0,0.12544000148773193
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,24,1,128,1,float16,float16,0,4.711833572387695
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,24,2,128,1,float16,float16,0,4.827033615112304
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,24,4,128,1,float16,float16,0,5.492787170410156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,24,1,128,1,float16,float16,0,2.2865888595581056
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,24,24,128,1,float16,float16,0,4.770560073852539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,24,8,128,1,float16,float16,0,6.44879379272461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,24,2,128,1,float16,float16,0,2.3286272048950196
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,24,4,128,1,float16,float16,0,2.6555904388427733
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,24,8,128,1,float16,float16,0,3.0674943923950195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,24,24,128,1,float16,float16,0,2.4444416046142576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,24,1,128,1,float16,float16,0,1.143500804901123
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,24,2,128,1,float16,float16,0,1.2035072326660157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,24,4,128,1,float16,float16,0,1.3543423652648925
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,24,8,128,1,float16,float16,0,1.5865856170654298
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,24,24,128,1,float16,float16,0,1.2328960418701171
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,24,1,128,1,float16,float16,0,0.6004735946655273
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,24,2,128,1,float16,float16,0,0.6114304065704346
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,24,4,128,1,float16,float16,0,0.6810111999511719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,24,8,128,1,float16,float16,0,0.788479995727539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,24,24,128,1,float16,float16,0,0.6248960018157959
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,24,1,128,1,float16,float16,0,0.3170815944671631
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,24,2,128,1,float16,float16,0,0.3270143985748291
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,24,4,128,1,float16,float16,0,0.34298880100250245
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,24,8,128,1,float16,float16,0,0.40447998046875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,24,24,128,1,float16,float16,0,0.3185663938522339
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,24,1,128,1,float16,float16,0,0.1655295968055725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,24,2,128,1,float16,float16,0,0.1704416036605835
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,24,4,128,1,float16,float16,0,0.17331199645996093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,24,24,128,1,float16,float16,0,0.12851200103759766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,24,8,128,1,float16,float16,0,0.1785312056541443
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,24,1,128,1,float16,float16,0,0.10301439762115479
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,24,2,128,1,float16,float16,0,0.10490880012512208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,24,4,128,1,float16,float16,0,0.10490880012512208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,24,24,128,1,float16,float16,0,0.08303999900817871
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,24,1,128,1,float16,float16,0,0.06471679806709289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,24,8,128,1,float16,float16,0,0.11207679510116578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,24,4,128,1,float16,float16,0,0.06865919828414917
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,24,2,128,1,float16,float16,0,0.06476799845695495
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,24,8,128,1,float16,float16,0,0.0717311978340149
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,24,1,128,1,float16,float16,0,3.483084869384766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,24,2,128,1,float16,float16,0,3.7072383880615236
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,24,4,128,1,float16,float16,0,4.196915054321289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,24,1,128,1,float16,float16,0,1.7083904266357421
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,24,2,128,1,float16,float16,0,1.8383871078491212
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,24,8,128,1,float16,float16,0,5.391001510620117
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,24,4,128,1,float16,float16,0,2.083635139465332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,24,24,128,1,float16,float16,0,5.151078414916992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,24,8,128,1,float16,float16,0,2.6993152618408205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,24,1,128,1,float16,float16,0,0.862003231048584
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,24,24,128,1,float16,float16,0,2.69168643951416
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,24,4,128,1,float16,float16,0,1.0612735748291016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,24,2,128,1,float16,float16,0,0.9259008407592774
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,24,1,128,1,float16,float16,0,0.44631037712097166
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,24,24,128,1,float16,float16,0,1.3733375549316407
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,24,8,128,1,float16,float16,0,1.3668352127075196
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,24,2,128,1,float16,float16,0,0.4731904029846191
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,24,4,128,1,float16,float16,0,0.5338111877441406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,24,8,128,1,float16,float16,0,0.6884352207183838
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,24,24,128,1,float16,float16,0,0.7049680233001709
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,24,2,128,1,float16,float16,0,0.2564608097076416
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,24,1,128,1,float16,float16,0,0.24673280715942383
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,24,8,128,1,float16,float16,0,0.342630410194397
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,24,4,128,1,float16,float16,0,0.2758656024932861
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,24,24,128,1,float16,float16,0,0.32143359184265136
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,24,1,128,1,float16,float16,0,0.1210368037223816
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,24,2,128,1,float16,float16,0,0.12339199781417846
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,24,4,128,1,float16,float16,0,0.1292799949645996
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,24,24,128,1,float16,float16,0,0.11658240556716919
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,24,8,128,1,float16,float16,0,0.14458880424499512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,24,1,128,1,float16,float16,0,0.07137280106544494
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,24,2,128,1,float16,float16,0,0.07449600100517273
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,24,4,128,1,float16,float16,0,0.07731199860572815
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,24,8,128,1,float16,float16,0,0.08304640054702758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,24,24,128,1,float16,float16,0,0.07556480169296265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,24,1,128,1,float16,float16,0,0.045465600490570066
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,24,2,128,1,float16,float16,0,0.045819199085235594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,24,4,128,1,float16,float16,0,0.046438398957252505
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,24,8,128,1,float16,float16,0,0.04961279928684235
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,24,24,128,1,float16,float16,0,0.059494400024414064
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,24,1,128,1,float16,float16,0,0.04469760060310364
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,24,2,128,1,float16,float16,0,0.0477183997631073
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,24,4,128,1,float16,float16,0,0.05022720098495483
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,24,8,128,1,float16,float16,0,0.05314559936523437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,24,1,128,1,float16,float16,0,1.407487964630127
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,24,2,128,1,float16,float16,0,1.5742976188659668
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,24,4,128,1,float16,float16,0,1.9078144073486327
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,24,8,128,1,float16,float16,0,2.4895488739013674
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,24,24,128,1,float16,float16,0,2.372198486328125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,24,1,128,1,float16,float16,0,0.703385591506958
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,24,2,128,1,float16,float16,0,0.7723008155822754
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,24,4,128,1,float16,float16,0,0.9469951629638672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,24,8,128,1,float16,float16,0,1.2600319862365723
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,24,24,128,1,float16,float16,0,1.2099583625793457
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,24,1,128,1,float16,float16,0,0.3680255889892578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,24,2,128,1,float16,float16,0,0.4118527889251709
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,24,4,128,1,float16,float16,0,0.4931072235107422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,24,8,128,1,float16,float16,0,0.6178304195404053
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,24,24,128,1,float16,float16,0,0.5952000141143798
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,24,1,128,1,float16,float16,0,0.156876802444458
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,24,4,128,1,float16,float16,0,0.20741119384765624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,24,2,128,1,float16,float16,0,0.19343359470367433
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,24,8,128,1,float16,float16,0,0.279040002822876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,24,1,128,1,float16,float16,0,0.07680000066757202
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,24,24,128,1,float16,float16,0,0.25021440982818605
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,24,2,128,1,float16,float16,0,0.07833279967308045
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,24,4,128,1,float16,float16,0,0.08411840200424195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,24,8,128,1,float16,float16,0,0.0939520001411438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,24,24,128,1,float16,float16,0,0.07848960161209106
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,24,1,128,1,float16,float16,0,0.04797439873218536
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,24,2,128,1,float16,float16,0,0.04766719937324524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,24,4,128,1,float16,float16,0,0.050892800092697144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,24,8,128,1,float16,float16,0,0.0563696026802063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,24,24,128,1,float16,float16,0,0.0489984005689621
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,24,1,128,1,float16,float16,0,0.0336896002292633
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,24,2,128,1,float16,float16,0,0.03425279855728149
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,24,4,128,1,float16,float16,0,0.03676159977912903
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,24,8,128,1,float16,float16,0,0.03865599930286408
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,24,24,128,1,float16,float16,0,0.032201600074768064
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,24,1,128,1,float16,float16,0,0.024985599517822265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,24,4,128,1,float16,float16,0,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,24,8,128,1,float16,float16,0,0.026675200462341307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,24,2,128,1,float16,float16,0,0.025497600436210632
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,24,24,128,1,float16,float16,0,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,24,1,128,1,float16,float16,0,0.023398399353027344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,24,2,128,1,float16,float16,0,0.023756800591945647
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,24,4,128,1,float16,float16,0,0.023654399812221526
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,24,8,128,1,float16,float16,0,0.023705600202083586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,24,1,128,1,float16,float16,0,0.693555212020874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,24,2,128,1,float16,float16,0,0.7781343936920166
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,24,4,128,1,float16,float16,0,0.9409024238586425
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,24,8,128,1,float16,float16,0,1.2420096397399902
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,24,1,128,1,float16,float16,0,0.3458559989929199
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,24,24,128,1,float16,float16,0,1.1907072067260742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,24,2,128,1,float16,float16,0,0.3731967926025391
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,24,4,128,1,float16,float16,0,0.4491759777069092
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,24,8,128,1,float16,float16,0,0.6141439914703369
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,24,24,128,1,float16,float16,0,0.5959167957305909
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,24,1,128,1,float16,float16,0,0.1363968014717102
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,24,2,128,1,float16,float16,0,0.17464319467544556
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,24,4,128,1,float16,float16,0,0.2081279993057251
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,24,24,128,1,float16,float16,0,0.24714241027832032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,24,8,128,1,float16,float16,0,0.27053759098052976
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,24,1,128,1,float16,float16,0,0.06332960128784179
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,24,2,128,1,float16,float16,0,0.06568959951400757
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,24,4,128,1,float16,float16,0,0.07065600156784058
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,24,8,128,1,float16,float16,0,0.08381440043449402
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,24,24,128,1,float16,float16,0,0.0707584023475647
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,24,1,128,1,float16,float16,0,0.03947519958019256
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,24,2,128,1,float16,float16,0,0.04080640077590943
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,24,4,128,1,float16,float16,0,0.043724799156188966
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,24,8,128,1,float16,float16,0,0.04924960136413574
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,24,24,128,1,float16,float16,0,0.045516800880432126
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,24,1,128,1,float16,float16,0,0.03031040132045746
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,24,2,128,1,float16,float16,0,0.0304639995098114
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,24,4,128,1,float16,float16,0,0.032358399033546446
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,24,8,128,1,float16,float16,0,0.034457600116729735
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,24,24,128,1,float16,float16,0,0.028518399596214293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,24,1,128,1,float16,float16,0,0.021401600539684297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,24,2,128,1,float16,float16,0,0.021503999829292297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,24,4,128,1,float16,float16,0,0.02252800017595291
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,24,8,128,1,float16,float16,0,0.02314240038394928
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,24,24,128,1,float16,float16,0,0.022519999742507936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,24,1,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,24,2,128,1,float16,float16,0,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,24,4,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,24,8,128,1,float16,float16,0,0.020214399695396422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,24,24,128,1,float16,float16,0,0.020019200444221497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,24,1,128,1,float16,float16,0,0.018940800428390504
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,24,2,128,1,float16,float16,0,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,24,4,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,24,8,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,24,1,128,1,float16,float16,0,0.34523839950561525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,24,2,128,1,float16,float16,0,0.37468159198760986
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,24,4,128,1,float16,float16,0,0.44805121421813965
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,24,8,128,1,float16,float16,0,0.6185984134674072
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,24,24,128,1,float16,float16,0,0.5790719985961914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,24,1,128,1,float16,float16,0,0.1566208004951477
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,24,2,128,1,float16,float16,0,0.17208319902420044
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,24,4,128,1,float16,float16,0,0.20449280738830566
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,24,8,128,1,float16,float16,0,0.2713599920272827
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,24,24,128,1,float16,float16,0,0.24862558841705323
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,24,2,128,1,float16,float16,0,0.06435840129852295
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,24,1,128,1,float16,float16,0,0.06164479851722717
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,24,4,128,1,float16,float16,0,0.06973440051078797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,24,8,128,1,float16,float16,0,0.09338880181312562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,24,24,128,1,float16,float16,0,0.08796160221099854
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,24,1,128,1,float16,float16,0,0.03906559944152832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,24,2,128,1,float16,float16,0,0.039987200498580934
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,24,4,128,1,float16,float16,0,0.04264959990978241
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,24,24,128,1,float16,float16,0,0.050995200872421265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,24,8,128,1,float16,float16,0,0.053913599252700804
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,24,1,128,1,float16,float16,0,0.02969599962234497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,24,2,128,1,float16,float16,0,0.030151998996734618
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,24,4,128,1,float16,float16,0,0.03107840120792389
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,24,8,128,1,float16,float16,0,0.03404799997806549
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,24,24,128,1,float16,float16,0,0.0347135990858078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,24,1,128,1,float16,float16,0,0.021401600539684297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,24,2,128,1,float16,float16,0,0.021196800470352172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,24,4,128,1,float16,float16,0,0.021555200219154358
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,24,8,128,1,float16,float16,0,0.023395200073719025
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,24,24,128,1,float16,float16,0,0.02232320010662079
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,24,1,128,1,float16,float16,0,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,24,2,128,1,float16,float16,0,0.019763199985027312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,24,4,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,24,8,128,1,float16,float16,0,0.01991039961576462
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,24,24,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,24,2,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,24,4,128,1,float16,float16,0,0.020627200603485107
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,24,1,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,24,8,128,1,float16,float16,0,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,24,24,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,24,1,128,1,float16,float16,0,0.018275199830532073
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,24,2,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,24,4,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,24,8,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,24,2,128,1,float16,float16,0,0.1793519973754883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,24,4,128,1,float16,float16,0,0.20633599758148194
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,24,1,128,1,float16,float16,0,0.15011839866638182
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,24,8,128,1,float16,float16,0,0.25912320613861084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,24,24,128,1,float16,float16,0,0.30248959064483644
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,24,1,128,1,float16,float16,0,0.0611840009689331
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,24,2,128,1,float16,float16,0,0.06415359973907471
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,24,4,128,1,float16,float16,0,0.08181759715080261
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,24,8,128,1,float16,float16,0,0.09354239702224731
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,24,24,128,1,float16,float16,0,0.1230847954750061
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,24,1,128,1,float16,float16,0,0.03865599930286408
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,24,2,128,1,float16,float16,0,0.04095999896526337
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,24,4,128,1,float16,float16,0,0.04838399887084961
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,24,8,128,1,float16,float16,0,0.05381119847297668
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,24,24,128,1,float16,float16,0,0.07065600156784058
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,24,1,128,1,float16,float16,0,0.02943040132522583
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,24,2,128,1,float16,float16,0,0.030051198601722718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,24,4,128,1,float16,float16,0,0.031027200818061828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,24,24,128,1,float16,float16,0,0.04167680144309997
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,24,8,128,1,float16,float16,0,0.0335312008857727
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,24,1,128,1,float16,float16,0,0.02088959962129593
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,24,2,128,1,float16,float16,0,0.021196800470352172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,24,4,128,1,float16,float16,0,0.021911999583244322
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,24,8,128,1,float16,float16,0,0.02314240038394928
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,24,24,128,1,float16,float16,0,0.0289792001247406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,24,1,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,24,2,128,1,float16,float16,0,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,24,4,128,1,float16,float16,0,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,24,8,128,1,float16,float16,0,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,24,24,128,1,float16,float16,0,0.01944800019264221
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,24,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,24,2,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,24,4,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,24,8,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,24,24,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,24,1,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,24,2,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,24,4,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,24,8,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,24,24,128,1,float16,float16,0,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,24,1,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,24,2,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,24,4,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,24,1,128,1,float16,float16,0,0.061952000856399535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,24,2,128,1,float16,float16,0,0.07551199793815613
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,24,8,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,24,8,128,1,float16,float16,0,0.14223359823226928
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,24,4,128,1,float16,float16,0,0.08284159898757934
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,24,24,128,1,float16,float16,0,0.19512319564819336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,24,1,128,1,float16,float16,0,0.03891200125217438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,24,2,128,1,float16,float16,0,0.046028798818588255
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,24,4,128,1,float16,float16,0,0.04915199875831604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,24,24,128,1,float16,float16,0,0.10388000011444092
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,24,8,128,1,float16,float16,0,0.08038399815559387
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,24,1,128,1,float16,float16,0,0.029798400402069092
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,24,4,128,1,float16,float16,0,0.032051199674606325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,24,2,128,1,float16,float16,0,0.03036159873008728
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,24,24,128,1,float16,float16,0,0.06184639930725098
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,24,8,128,1,float16,float16,0,0.046438398957252505
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,24,1,128,1,float16,float16,0,0.02114560008049011
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,24,2,128,1,float16,float16,0,0.02165600061416626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,24,8,128,1,float16,float16,0,0.029491201043128967
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,24,1,128,1,float16,float16,0,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,24,4,128,1,float16,float16,0,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,24,24,128,1,float16,float16,0,0.03594239950180054
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,24,4,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,24,2,128,1,float16,float16,0,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,24,8,128,1,float16,float16,0,0.020428800582885744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,24,24,128,1,float16,float16,0,0.026265600323677064
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,24,1,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,24,2,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,24,8,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,24,24,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,24,4,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,24,1,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,24,2,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,24,4,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,24,8,128,1,float16,float16,0,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,24,24,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,24,1,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,24,2,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,24,8,128,1,float16,float16,0,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,24,4,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,24,24,128,1,float16,float16,0,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,24,1,128,1,float16,float16,0,0.018171200156211854
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,24,2,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,24,4,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,24,8,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,16,1,128,1,float16,float16,0,19.04261169433594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,16,2,128,1,float16,float16,0,18.894438171386717
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,16,16,128,1,float16,float16,0,20.949656677246093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,16,4,128,1,float16,float16,0,19.681280517578124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,16,1,128,1,float16,float16,0,39.012454223632815
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,16,2,128,1,float16,float16,0,39.118899536132815
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,16,4,128,1,float16,float16,0,40.038656616210936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,16,8,128,1,float16,float16,0,40.41082763671875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,16,1,128,1,float16,float16,0,9.22945785522461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,16,16,128,1,float16,float16,0,10.333235168457032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,16,2,128,1,float16,float16,0,9.307750701904297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,16,8,128,1,float16,float16,0,20.022169494628905
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,16,4,128,1,float16,float16,0,9.639724731445312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,16,1,128,1,float16,float16,0,4.43084487915039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,16,8,128,1,float16,float16,0,9.966284942626952
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,16,4,128,1,float16,float16,0,4.695910263061523
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,16,2,128,1,float16,float16,0,4.638822555541992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,16,16,128,1,float16,float16,0,5.1539966583251955
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,16,8,128,1,float16,float16,0,4.906649780273438
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,16,1,128,1,float16,float16,0,10.75445785522461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,16,16,128,1,float16,float16,0,12.238028717041015
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,16,4,128,1,float16,float16,0,11.36343002319336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,16,2,128,1,float16,float16,0,10.883686065673828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,16,1,128,1,float16,float16,0,21.63512268066406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,16,2,128,1,float16,float16,0,22.39052734375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,16,4,128,1,float16,float16,0,22.3878662109375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,16,8,128,1,float16,float16,0,23.530393981933592
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,16,16,128,1,float16,float16,0,5.8147327423095705
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,16,1,128,1,float16,float16,0,5.185843276977539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,16,2,128,1,float16,float16,0,5.12174072265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,16,4,128,1,float16,float16,0,5.371593475341797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,16,8,128,1,float16,float16,0,11.49486083984375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,16,16,128,1,float16,float16,0,2.9838336944580077
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,16,1,128,1,float16,float16,0,2.558310317993164
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,16,8,128,1,float16,float16,0,5.734297561645508
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,16,2,128,1,float16,float16,0,2.5485824584960937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,16,4,128,1,float16,float16,0,2.653798484802246
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,16,8,128,1,float16,float16,0,2.7312128067016603
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,16,16,128,1,float16,float16,0,8.535654449462891
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,16,1,128,1,float16,float16,0,15.084953308105469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,16,1,128,1,float16,float16,0,7.470130920410156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,16,2,128,1,float16,float16,0,7.478272247314453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,16,4,128,1,float16,float16,0,15.864422607421876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,16,2,128,1,float16,float16,0,15.555430603027343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,16,8,128,1,float16,float16,0,16.297164916992188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,16,4,128,1,float16,float16,0,7.565670776367187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,16,16,128,1,float16,float16,0,4.167577743530273
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,16,1,128,1,float16,float16,0,3.442534255981445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,16,2,128,1,float16,float16,0,3.627665710449219
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,16,8,128,1,float16,float16,0,7.9929344177246096
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,16,4,128,1,float16,float16,0,3.72147216796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,16,8,128,1,float16,float16,0,3.957708740234375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,16,16,128,1,float16,float16,0,2.1541887283325196
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,16,1,128,1,float16,float16,0,1.8256864547729492
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,16,2,128,1,float16,float16,0,1.816985511779785
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,16,4,128,1,float16,float16,0,1.9036672592163086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,16,8,128,1,float16,float16,0,1.9909631729125976
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,16,16,128,1,float16,float16,0,11.635302734375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,16,1,128,1,float16,float16,0,19.84593811035156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,16,2,128,1,float16,float16,0,19.977267456054687
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,16,1,128,1,float16,float16,0,9.800704193115234
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,16,2,128,1,float16,float16,0,9.791180419921876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,16,4,128,1,float16,float16,0,20.606874084472658
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,16,4,128,1,float16,float16,0,10.095257568359376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,16,8,128,1,float16,float16,0,21.992037963867187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,16,1,128,1,float16,float16,0,4.6939647674560545
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,16,16,128,1,float16,float16,0,2.843955230712891
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,16,2,128,1,float16,float16,0,4.733235168457031
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,16,16,128,1,float16,float16,0,5.699937438964843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,16,4,128,1,float16,float16,0,4.73779182434082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,16,8,128,1,float16,float16,0,5.193164825439453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,16,1,128,1,float16,float16,0,2.317875289916992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,16,8,128,1,float16,float16,0,10.72511978149414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,16,2,128,1,float16,float16,0,2.3021568298339843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,16,16,128,1,float16,float16,0,1.4729727745056151
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,16,4,128,1,float16,float16,0,2.369638442993164
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,16,1,128,1,float16,float16,0,1.3165568351745605
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,16,8,128,1,float16,float16,0,2.5592319488525392
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,16,2,128,1,float16,float16,0,1.2802047729492188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,16,4,128,1,float16,float16,0,1.1966976165771483
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,16,8,128,1,float16,float16,0,1.3225407600402832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,16,1,128,1,float16,float16,0,5.256140899658203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,16,2,128,1,float16,float16,0,5.5213569641113285
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,16,16,128,1,float16,float16,0,6.866534423828125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,16,1,128,1,float16,float16,0,11.123814392089844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,16,2,128,1,float16,float16,0,11.296819305419922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,16,4,128,1,float16,float16,0,11.704934692382812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,16,8,128,1,float16,float16,0,12.743373107910156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,16,4,128,1,float16,float16,0,5.741875076293946
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,16,1,128,1,float16,float16,0,2.639718437194824
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,16,16,128,1,float16,float16,0,3.3295360565185548
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,16,8,128,1,float16,float16,0,6.114048004150391
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,16,2,128,1,float16,float16,0,2.6770944595336914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,16,4,128,1,float16,float16,0,2.7313663482666017
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,16,8,128,1,float16,float16,0,2.9197824478149412
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,16,16,128,1,float16,float16,0,1.7227775573730468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,16,1,128,1,float16,float16,0,1.325312042236328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,16,2,128,1,float16,float16,0,1.3548975944519044
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,16,4,128,1,float16,float16,0,1.4011391639709472
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,16,16,128,1,float16,float16,0,0.9025535583496094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,16,8,128,1,float16,float16,0,1.5177215576171874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,16,1,128,1,float16,float16,0,0.7563776016235352
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,16,4,128,1,float16,float16,0,0.7581183910369873
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,16,2,128,1,float16,float16,0,0.775167989730835
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,16,8,128,1,float16,float16,0,0.7613440036773682
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,16,1,128,1,float16,float16,0,4.764774322509766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,16,16,128,1,float16,float16,0,6.7251708984375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,16,2,128,1,float16,float16,0,4.989235305786133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,16,1,128,1,float16,float16,0,10.502859497070313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,16,2,128,1,float16,float16,0,10.459954833984375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,16,4,128,1,float16,float16,0,11.395635223388672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,16,4,128,1,float16,float16,0,5.420646286010742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,16,8,128,1,float16,float16,0,12.146790313720704
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,16,1,128,1,float16,float16,0,2.38852481842041
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,16,2,128,1,float16,float16,0,2.460518455505371
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,16,4,128,1,float16,float16,0,2.6020864486694335
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,16,16,128,1,float16,float16,0,3.260723114013672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,16,8,128,1,float16,float16,0,5.907712173461914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,16,8,128,1,float16,float16,0,2.9119455337524416
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,16,16,128,1,float16,float16,0,1.7175039291381835
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,16,1,128,1,float16,float16,0,1.2140031814575196
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,16,2,128,1,float16,float16,0,1.2327424049377442
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,16,8,128,1,float16,float16,0,1.4443519592285157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,16,4,128,1,float16,float16,0,1.3352959632873536
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,16,16,128,1,float16,float16,0,0.8686079978942871
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,16,1,128,1,float16,float16,0,0.6487535953521728
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,16,2,128,1,float16,float16,0,0.6433792114257812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,16,4,128,1,float16,float16,0,0.6544384002685547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,16,8,128,1,float16,float16,0,0.7469567775726318
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,16,16,128,1,float16,float16,0,0.425216007232666
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,16,1,128,1,float16,float16,0,0.365721607208252
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,16,2,128,1,float16,float16,0,0.362444806098938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,16,4,128,1,float16,float16,0,0.3796479940414429
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,16,8,128,1,float16,float16,0,0.3816368103027344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,16,1,128,1,float16,float16,0,5.896448135375977
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,16,2,128,1,float16,float16,0,6.135398483276367
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,16,1,128,1,float16,float16,0,2.8252159118652345
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,16,2,128,1,float16,float16,0,2.8000255584716798
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,16,4,128,1,float16,float16,0,6.369228744506836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,16,16,128,1,float16,float16,0,4.132249450683593
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,16,4,128,1,float16,float16,0,3.04716796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,16,8,128,1,float16,float16,0,7.023104095458985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,16,8,128,1,float16,float16,0,3.5132400512695314
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,16,16,128,1,float16,float16,0,2.0786176681518556
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,16,1,128,1,float16,float16,0,1.3732352256774902
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,16,2,128,1,float16,float16,0,1.4535167694091797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,16,8,128,1,float16,float16,0,1.7480176925659179
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,16,16,128,1,float16,float16,0,1.0438655853271483
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,16,1,128,1,float16,float16,0,0.7425536155700684
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,16,4,128,1,float16,float16,0,1.558732795715332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,16,2,128,1,float16,float16,0,0.7539711952209472
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,16,4,128,1,float16,float16,0,0.798464012145996
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,16,16,128,1,float16,float16,0,0.5489151954650879
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,16,8,128,1,float16,float16,0,0.8972800254821778
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,16,2,128,1,float16,float16,0,0.40714240074157715
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,16,4,128,1,float16,float16,0,0.3980799913406372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,16,1,128,1,float16,float16,0,0.3970544099807739
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,16,8,128,1,float16,float16,0,0.4306431770324707
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,16,16,128,1,float16,float16,0,0.25830399990081787
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,16,1,128,1,float16,float16,0,0.23173120021820068
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,16,8,128,1,float16,float16,0,0.2514944076538086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,16,2,128,1,float16,float16,0,0.23710720539093016
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,16,4,128,1,float16,float16,0,0.24191999435424805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,16,1,128,1,float16,float16,0,5.591803359985351
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,16,2,128,1,float16,float16,0,5.798348617553711
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,16,4,128,1,float16,float16,0,6.242967987060547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,16,16,128,1,float16,float16,0,4.352153778076172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,16,1,128,1,float16,float16,0,2.5901056289672852
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,16,2,128,1,float16,float16,0,2.7487232208251955
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,16,4,128,1,float16,float16,0,3.0175743103027344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,16,8,128,1,float16,float16,0,7.173887634277344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,16,16,128,1,float16,float16,0,2.1625343322753907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,16,8,128,1,float16,float16,0,3.523271942138672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,16,1,128,1,float16,float16,0,1.325004768371582
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,16,16,128,1,float16,float16,0,1.1106816291809083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,16,4,128,1,float16,float16,0,1.5003135681152344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,16,1,128,1,float16,float16,0,0.7040512084960937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,16,8,128,1,float16,float16,0,1.7414655685424805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,16,2,128,1,float16,float16,0,1.399449634552002
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,16,2,128,1,float16,float16,0,0.7056384086608887
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,16,4,128,1,float16,float16,0,0.7848447799682617
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,16,16,128,1,float16,float16,0,0.5761023998260498
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,16,2,128,1,float16,float16,0,0.3556864023208618
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,16,8,128,1,float16,float16,0,0.8866815567016602
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,16,1,128,1,float16,float16,0,0.3505664110183716
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,16,4,128,1,float16,float16,0,0.3767807960510254
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,16,8,128,1,float16,float16,0,0.4517888069152832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,16,16,128,1,float16,float16,0,0.25676798820495605
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,16,1,128,1,float16,float16,0,0.2047951936721802
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,16,2,128,1,float16,float16,0,0.21155838966369628
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,16,4,128,1,float16,float16,0,0.21826560497283937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,16,1,128,1,float16,float16,0,0.12753920555114745
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,16,16,128,1,float16,float16,0,0.15385600328445434
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,16,8,128,1,float16,float16,0,0.2248703956604004
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,16,2,128,1,float16,float16,0,0.13419519662857055
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,16,4,128,1,float16,float16,0,0.14140959978103637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,16,8,128,1,float16,float16,0,0.14602240324020385
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,16,1,128,1,float16,float16,0,3.148441505432129
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,16,2,128,1,float16,float16,0,3.420364761352539
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,16,4,128,1,float16,float16,0,3.7456897735595702
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,16,16,128,1,float16,float16,0,2.8085248947143553
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,16,8,128,1,float16,float16,0,4.37396469116211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,16,2,128,1,float16,float16,0,1.6328144073486328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,16,1,128,1,float16,float16,0,1.5654911994934082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,16,4,128,1,float16,float16,0,1.8562047958374024
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,16,8,128,1,float16,float16,0,2.18787841796875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,16,16,128,1,float16,float16,0,1.4418944358825683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,16,2,128,1,float16,float16,0,0.8561663627624512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,16,1,128,1,float16,float16,0,0.8027647972106934
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,16,4,128,1,float16,float16,0,0.9263615608215332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,16,8,128,1,float16,float16,0,1.0801152229309081
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,16,1,128,1,float16,float16,0,0.4257232189178467
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,16,16,128,1,float16,float16,0,0.7262703895568847
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,16,2,128,1,float16,float16,0,0.43939838409423826
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,16,4,128,1,float16,float16,0,0.48942079544067385
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,16,8,128,1,float16,float16,0,0.5634047985076904
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,16,16,128,1,float16,float16,0,0.374783992767334
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,16,1,128,1,float16,float16,0,0.21990399360656737
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,16,2,128,1,float16,float16,0,0.23152639865875244
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,16,4,128,1,float16,float16,0,0.23598079681396483
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,16,8,128,1,float16,float16,0,0.2554368019104004
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,16,16,128,1,float16,float16,0,0.16010240316390992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,16,1,128,1,float16,float16,0,0.1366960048675537
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,16,2,128,1,float16,float16,0,0.1385983943939209
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,16,4,128,1,float16,float16,0,0.14161920547485352
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,16,8,128,1,float16,float16,0,0.14888960123062134
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,16,16,128,1,float16,float16,0,0.10444799661636353
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,16,1,128,1,float16,float16,0,0.09410560131072998
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,16,2,128,1,float16,float16,0,0.09425920248031616
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,16,4,128,1,float16,float16,0,0.09835519790649414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,16,8,128,1,float16,float16,0,0.10209280252456665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,16,1,128,1,float16,float16,0,3.104051208496094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,16,2,128,1,float16,float16,0,3.313510513305664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,16,4,128,1,float16,float16,0,3.768627166748047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,16,16,128,1,float16,float16,0,3.194112014770508
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,16,8,128,1,float16,float16,0,4.606924819946289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,16,2,128,1,float16,float16,0,1.65795841217041
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,16,4,128,1,float16,float16,0,1.9239936828613282
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,16,1,128,1,float16,float16,0,1.5353856086730957
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,16,8,128,1,float16,float16,0,2.343065643310547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,16,1,128,1,float16,float16,0,0.789299201965332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,16,16,128,1,float16,float16,0,1.6373247146606444
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,16,2,128,1,float16,float16,0,0.8614399909973145
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,16,4,128,1,float16,float16,0,0.9629695892333985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,16,8,128,1,float16,float16,0,1.1775487899780273
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,16,16,128,1,float16,float16,0,0.8357872009277344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,16,1,128,1,float16,float16,0,0.41451520919799806
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,16,2,128,1,float16,float16,0,0.4390399932861328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,16,4,128,1,float16,float16,0,0.4848639965057373
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,16,8,128,1,float16,float16,0,0.5974527835845947
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,16,16,128,1,float16,float16,0,0.4244991779327393
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,16,1,128,1,float16,float16,0,0.21196799278259276
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,16,2,128,1,float16,float16,0,0.21232318878173828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,16,4,128,1,float16,float16,0,0.23541760444641113
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,16,16,128,1,float16,float16,0,0.1679360032081604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,16,8,128,1,float16,float16,0,0.30253760814666747
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,16,1,128,1,float16,float16,0,0.12257280349731445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,16,2,128,1,float16,float16,0,0.12467199563980103
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,16,8,128,1,float16,float16,0,0.1381376028060913
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,16,16,128,1,float16,float16,0,0.10076160430908203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,16,4,128,1,float16,float16,0,0.13199360370635987
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,16,1,128,1,float16,float16,0,0.07823200225830078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,16,2,128,1,float16,float16,0,0.08335360288619995
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,16,8,128,1,float16,float16,0,0.09057279825210571
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,16,4,128,1,float16,float16,0,0.08729599714279175
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,16,16,128,1,float16,float16,0,0.05934079885482788
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,16,2,128,1,float16,float16,0,0.04976640045642853
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,16,4,128,1,float16,float16,0,0.050995200872421265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,16,1,128,1,float16,float16,0,0.0494592010974884
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,16,8,128,1,float16,float16,0,0.05355520248413086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,16,1,128,1,float16,float16,0,2.3295488357543945
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,16,2,128,1,float16,float16,0,2.5496063232421875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,16,1,128,1,float16,float16,0,1.1667967796325684
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,16,2,128,1,float16,float16,0,1.3000639915466308
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,16,4,128,1,float16,float16,0,3.1310848236083983
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,16,4,128,1,float16,float16,0,1.6089088439941406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,16,16,128,1,float16,float16,0,3.3079807281494142
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,16,8,128,1,float16,float16,0,4.372787094116211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,16,8,128,1,float16,float16,0,2.32488956451416
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,16,1,128,1,float16,float16,0,0.6055424213409424
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,16,4,128,1,float16,float16,0,0.8056320190429688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,16,2,128,1,float16,float16,0,0.6552063941955566
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,16,16,128,1,float16,float16,0,1.8081792831420898
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,16,8,128,1,float16,float16,0,1.1795455932617187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,16,16,128,1,float16,float16,0,0.9378288269042969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,16,1,128,1,float16,float16,0,0.3226624011993408
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,16,2,128,1,float16,float16,0,0.3429375886917114
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,16,4,128,1,float16,float16,0,0.4015103816986084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,16,8,128,1,float16,float16,0,0.5750271797180175
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,16,16,128,1,float16,float16,0,0.4443647861480713
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,16,1,128,1,float16,float16,0,0.16435199975967407
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,16,2,128,1,float16,float16,0,0.16952320337295532
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,16,4,128,1,float16,float16,0,0.1962496042251587
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,16,8,128,1,float16,float16,0,0.2649087905883789
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,16,16,128,1,float16,float16,0,0.18017280101776123
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,16,1,128,1,float16,float16,0,0.09589759707450866
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,16,2,128,1,float16,float16,0,0.09804800152778625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,16,4,128,1,float16,float16,0,0.09953280091285706
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,16,8,128,1,float16,float16,0,0.1129472017288208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,16,16,128,1,float16,float16,0,0.08565760254859925
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,16,1,128,1,float16,float16,0,0.05939199924468994
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,16,4,128,1,float16,float16,0,0.0637440025806427
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,16,2,128,1,float16,float16,0,0.06113280057907104
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,16,8,128,1,float16,float16,0,0.07101439833641052
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,16,16,128,1,float16,float16,0,0.05309439897537231
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,16,1,128,1,float16,float16,0,0.04285120069980621
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,16,2,128,1,float16,float16,0,0.04321280121803284
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,16,4,128,1,float16,float16,0,0.04469760060310364
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,16,8,128,1,float16,float16,0,0.04699679911136627
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,16,16,128,1,float16,float16,0,0.04505600035190582
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,16,2,128,1,float16,float16,0,0.040089601278305055
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,16,4,128,1,float16,float16,0,0.040038400888442995
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,16,1,128,1,float16,float16,0,0.03978239893913269
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,16,8,128,1,float16,float16,0,0.04162560105323791
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,16,1,128,1,float16,float16,0,0.9845248222351074
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,16,2,128,1,float16,float16,0,1.1444735527038574
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,16,4,128,1,float16,float16,0,1.468569564819336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,16,8,128,1,float16,float16,0,2.0551679611206053
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,16,1,128,1,float16,float16,0,0.4874239921569824
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,16,16,128,1,float16,float16,0,1.5902720451354981
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,16,2,128,1,float16,float16,0,0.5524479866027832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,16,4,128,1,float16,float16,0,0.7201791763305664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,16,8,128,1,float16,float16,0,1.032908821105957
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,16,16,128,1,float16,float16,0,0.8082431793212891
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,16,1,128,1,float16,float16,0,0.24642560482025147
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,16,2,128,1,float16,float16,0,0.28083200454711915
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,16,4,128,1,float16,float16,0,0.3353600025177002
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,16,8,128,1,float16,float16,0,0.5049808025360107
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,16,1,128,1,float16,float16,0,0.09507840275764465
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,16,16,128,1,float16,float16,0,0.36771359443664553
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,16,4,128,1,float16,float16,0,0.11642719507217407
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,16,2,128,1,float16,float16,0,0.10183680057525635
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,16,8,128,1,float16,float16,0,0.19993599653244018
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,16,16,128,1,float16,float16,0,0.10337280035018921
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,16,1,128,1,float16,float16,0,0.059084802865982056
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,16,2,128,1,float16,float16,0,0.061286401748657224
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,16,4,128,1,float16,float16,0,0.06589440107345582
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,16,16,128,1,float16,float16,0,0.05862399935722351
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,16,1,128,1,float16,float16,0,0.037990400195121767
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,16,8,128,1,float16,float16,0,0.07649279832839966
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,16,2,128,1,float16,float16,0,0.03937279880046844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,16,4,128,1,float16,float16,0,0.042444801330566405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,16,8,128,1,float16,float16,0,0.04756160080432892
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,16,16,128,1,float16,float16,0,0.03614400029182434
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,16,1,128,1,float16,float16,0,0.026163199543952943
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,16,2,128,1,float16,float16,0,0.026624000072479247
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,16,4,128,1,float16,float16,0,0.027647998929023743
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,16,16,128,1,float16,float16,0,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,16,8,128,1,float16,float16,0,0.030822399258613586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,16,1,128,1,float16,float16,0,0.024371199309825897
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,16,2,128,1,float16,float16,0,0.02462719976902008
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,16,4,128,1,float16,float16,0,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,16,8,128,1,float16,float16,0,0.026163199543952943
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,16,16,128,1,float16,float16,0,0.02503040134906769
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,16,1,128,1,float16,float16,0,0.023296000063419343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,16,2,128,1,float16,float16,0,0.023296000063419343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,16,8,128,1,float16,float16,0,0.023705600202083586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,16,4,128,1,float16,float16,0,0.023244799673557283
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,16,1,128,1,float16,float16,0,0.49152002334594724
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,16,2,128,1,float16,float16,0,0.562278413772583
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,16,4,128,1,float16,float16,0,0.7226367950439453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,16,8,128,1,float16,float16,0,1.0263551712036132
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,16,1,128,1,float16,float16,0,0.23311200141906738
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,16,16,128,1,float16,float16,0,0.800716781616211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,16,2,128,1,float16,float16,0,0.2614255905151367
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,16,4,128,1,float16,float16,0,0.3325952053070068
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,16,8,128,1,float16,float16,0,0.4997119903564453
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,16,16,128,1,float16,float16,0,0.36869120597839355
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,16,1,128,1,float16,float16,0,0.07879679799079894
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,16,2,128,1,float16,float16,0,0.08442879915237426
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,16,4,128,1,float16,float16,0,0.10117119550704956
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,16,8,128,1,float16,float16,0,0.19379199743270875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,16,16,128,1,float16,float16,0,0.09297919869422913
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,16,2,128,1,float16,float16,0,0.051507198810577394
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,16,4,128,1,float16,float16,0,0.05719040036201477
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,16,1,128,1,float16,float16,0,0.04878720045089722
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,16,8,128,1,float16,float16,0,0.06778879761695862
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,16,16,128,1,float16,float16,0,0.0518127977848053
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,16,1,128,1,float16,float16,0,0.03184640109539032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,16,4,128,1,float16,float16,0,0.03609600067138672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,16,2,128,1,float16,float16,0,0.03317759931087494
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,16,8,128,1,float16,float16,0,0.04111360013484955
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,16,16,128,1,float16,float16,0,0.032153600454330446
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,16,1,128,1,float16,float16,0,0.02242559939622879
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,16,2,128,1,float16,float16,0,0.022835199534893037
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,16,4,128,1,float16,float16,0,0.024115200340747833
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,16,1,128,1,float16,float16,0,0.020787200331687926
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,16,8,128,1,float16,float16,0,0.02677760124206543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,16,16,128,1,float16,float16,0,0.024985599517822265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,16,2,128,1,float16,float16,0,0.02104319930076599
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,16,4,128,1,float16,float16,0,0.02094080001115799
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,16,8,128,1,float16,float16,0,0.022579200565814972
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,16,1,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,16,16,128,1,float16,float16,0,0.021196800470352172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,16,4,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,16,2,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,16,8,128,1,float16,float16,0,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,16,16,128,1,float16,float16,0,0.01966080069541931
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,16,1,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,16,2,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,16,4,128,1,float16,float16,0,0.019097599387168884
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,16,8,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,16,1,128,1,float16,float16,0,0.23116800785064698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,16,4,128,1,float16,float16,0,0.33182721138000487
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,16,2,128,1,float16,float16,0,0.2589184045791626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,16,8,128,1,float16,float16,0,0.5041071891784668
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,16,16,128,1,float16,float16,0,0.3658751964569092
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,16,2,128,1,float16,float16,0,0.08366079926490784
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,16,1,128,1,float16,float16,0,0.07838720083236694
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,16,4,128,1,float16,float16,0,0.1013759970664978
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,16,8,128,1,float16,float16,0,0.19486720561981202
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,16,16,128,1,float16,float16,0,0.11873279809951783
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,16,1,128,1,float16,float16,0,0.04746240079402923
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,16,2,128,1,float16,float16,0,0.05017600059509277
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,16,4,128,1,float16,float16,0,0.05575680136680603
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,16,8,128,1,float16,float16,0,0.06732800006866455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,16,1,128,1,float16,float16,0,0.03112959861755371
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,16,16,128,1,float16,float16,0,0.06338559985160827
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,16,2,128,1,float16,float16,0,0.032153600454330446
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,16,4,128,1,float16,float16,0,0.034764799475669864
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,16,8,128,1,float16,float16,0,0.04044800102710724
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,16,16,128,1,float16,float16,0,0.037887999415397645
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,16,1,128,1,float16,float16,0,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,16,2,128,1,float16,float16,0,0.022784000635147093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,16,4,128,1,float16,float16,0,0.02385759949684143
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,16,16,128,1,float16,float16,0,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,16,2,128,1,float16,float16,0,0.020735999941825865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,16,4,128,1,float16,float16,0,0.02252800017595291
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,16,1,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,16,8,128,1,float16,float16,0,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,16,8,128,1,float16,float16,0,0.022168000042438508
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,16,16,128,1,float16,float16,0,0.02094080001115799
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,16,1,128,1,float16,float16,0,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,16,2,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,16,8,128,1,float16,float16,0,0.019763199985027312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,16,16,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,16,4,128,1,float16,float16,0,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,16,1,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,16,2,128,1,float16,float16,0,0.0188400000333786
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,16,4,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,16,1,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,16,8,128,1,float16,float16,0,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,16,16,128,1,float16,float16,0,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,16,4,128,1,float16,float16,0,0.01852799952030182
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,16,2,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,16,8,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,16,1,128,1,float16,float16,0,0.07833600044250488
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,16,4,128,1,float16,float16,0,0.1037824034690857
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,16,1,128,1,float16,float16,0,0.04776960015296936
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,16,2,128,1,float16,float16,0,0.0839680016040802
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,16,8,128,1,float16,float16,0,0.21698238849639892
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,16,16,128,1,float16,float16,0,0.16860159635543823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,16,2,128,1,float16,float16,0,0.05063199996948242
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,16,4,128,1,float16,float16,0,0.05601279735565186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,16,8,128,1,float16,float16,0,0.09180160164833069
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,16,16,128,1,float16,float16,0,0.09149439930915833
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,16,1,128,1,float16,float16,0,0.031027200818061828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,16,2,128,1,float16,float16,0,0.03246079981327057
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,16,8,128,1,float16,float16,0,0.05272960066795349
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,16,16,128,1,float16,float16,0,0.051097601652145386
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,16,4,128,1,float16,float16,0,0.034764799475669864
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,16,2,128,1,float16,float16,0,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,16,1,128,1,float16,float16,0,0.02242559939622879
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,16,4,128,1,float16,float16,0,0.024166400730609893
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,16,8,128,1,float16,float16,0,0.03292160034179688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,16,16,128,1,float16,float16,0,0.031385600566864014
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,16,1,128,1,float16,float16,0,0.020475199818611144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,16,2,128,1,float16,float16,0,0.020374399423599244
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,16,4,128,1,float16,float16,0,0.020787200331687926
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,16,16,128,1,float16,float16,0,0.02160640060901642
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,16,8,128,1,float16,float16,0,0.022681599855422972
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,16,1,128,1,float16,float16,0,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,16,2,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,16,4,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,16,8,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,16,16,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,16,1,128,1,float16,float16,0,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,16,2,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,16,8,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,16,4,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,16,16,128,1,float16,float16,0,0.018328000605106354
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,16,1,128,1,float16,float16,0,0.018275199830532073
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,16,2,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,16,4,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,16,16,128,1,float16,float16,0,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,16,8,128,1,float16,float16,0,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,16,2,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,16,4,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,16,1,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,16,8,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,16,1,128,1,float16,float16,0,0.048230400681495665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,16,2,128,1,float16,float16,0,0.05119839906692505
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,16,4,128,1,float16,float16,0,0.08145920038223267
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,16,16,128,1,float16,float16,0,0.13946880102157594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,16,8,128,1,float16,float16,0,0.13941760063171388
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,16,1,128,1,float16,float16,0,0.031641599535942075
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,16,2,128,1,float16,float16,0,0.032767999172210696
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,16,4,128,1,float16,float16,0,0.047870400547981265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,16,8,128,1,float16,float16,0,0.07997440099716187
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,16,16,128,1,float16,float16,0,0.07823359966278076
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,16,1,128,1,float16,float16,0,0.022886399924755097
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,16,4,128,1,float16,float16,0,0.03031040132045746
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,16,8,128,1,float16,float16,0,0.04628480076789856
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,16,2,128,1,float16,float16,0,0.022988800704479218
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,16,1,128,1,float16,float16,0,0.02067680060863495
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,16,16,128,1,float16,float16,0,0.045465600490570066
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,16,2,128,1,float16,float16,0,0.020479999482631683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,16,4,128,1,float16,float16,0,0.021401600539684297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,16,8,128,1,float16,float16,0,0.029183998703956604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,16,16,128,1,float16,float16,0,0.029798400402069092
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,16,1,128,1,float16,float16,0,0.019244800508022308
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,16,2,128,1,float16,float16,0,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,16,4,128,1,float16,float16,0,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,16,16,128,1,float16,float16,0,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,16,8,128,1,float16,float16,0,0.019916799664497376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,16,2,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,16,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,16,4,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,16,8,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,16,16,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,16,1,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,16,2,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,16,8,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,16,16,128,1,float16,float16,0,0.017764799296855927
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,16,4,128,1,float16,float16,0,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,16,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,16,2,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,16,8,128,1,float16,float16,0,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,16,16,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,16,4,128,1,float16,float16,0,0.017705599963665008
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,16,1,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,16,2,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,16,4,128,1,float16,float16,0,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,16,8,128,1,float16,float16,0,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,12,12,128,1,float16,float16,0,7.764838409423828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,12,1,128,1,float16,float16,0,14.088243103027343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,12,2,128,1,float16,float16,0,14.38970947265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,12,12,128,1,float16,float16,0,16.003736877441405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,12,4,128,1,float16,float16,0,14.9243896484375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,12,1,128,1,float16,float16,0,28.847512817382814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,12,2,128,1,float16,float16,0,28.9512451171875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,12,1,128,1,float16,float16,0,6.953215789794922
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,12,4,128,1,float16,float16,0,29.79210205078125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,12,12,128,1,float16,float16,0,3.8990848541259764
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,12,2,128,1,float16,float16,0,7.053977966308594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,12,4,128,1,float16,float16,0,7.319705963134766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,12,1,128,1,float16,float16,0,3.438227081298828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,12,2,128,1,float16,float16,0,3.427174377441406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,12,4,128,1,float16,float16,0,3.5359729766845702
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,12,1,128,1,float16,float16,0,8.092876434326172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,12,12,128,1,float16,float16,0,9.074227142333985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,12,2,128,1,float16,float16,0,7.898820495605468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,12,4,128,1,float16,float16,0,8.473036956787109
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,12,1,128,1,float16,float16,0,16.086630249023436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,12,12,128,1,float16,float16,0,4.315084838867188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,12,4,128,1,float16,float16,0,17.149746704101563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,12,2,128,1,float16,float16,0,16.53975067138672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,12,12,128,1,float16,float16,0,2.295751953125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,12,1,128,1,float16,float16,0,1.9709936141967774
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,12,1,128,1,float16,float16,0,3.910655975341797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,12,2,128,1,float16,float16,0,3.9352832794189454
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,12,2,128,1,float16,float16,0,2.043340873718262
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,12,4,128,1,float16,float16,0,3.8851585388183594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,12,4,128,1,float16,float16,0,2.0335615158081053
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,12,1,128,1,float16,float16,0,11.407817840576172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,12,12,128,1,float16,float16,0,6.519956970214844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,12,1,128,1,float16,float16,0,5.462937545776367
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,12,2,128,1,float16,float16,0,5.548646545410156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,12,4,128,1,float16,float16,0,5.709107208251953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,12,12,128,1,float16,float16,0,3.0738431930541994
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,12,2,128,1,float16,float16,0,11.56126708984375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,12,4,128,1,float16,float16,0,12.018278503417969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,12,1,128,1,float16,float16,0,2.6677248001098635
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,12,12,128,1,float16,float16,0,1.6386560440063476
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,12,2,128,1,float16,float16,0,2.7295743942260744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,12,4,128,1,float16,float16,0,2.8330495834350584
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,12,1,128,1,float16,float16,0,1.5048704147338867
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,12,2,128,1,float16,float16,0,1.5358927726745606
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,12,4,128,1,float16,float16,0,1.4227968215942384
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,12,1,128,1,float16,float16,0,6.944972991943359
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,12,12,128,1,float16,float16,0,4.167935943603515
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,12,1,128,1,float16,float16,0,14.777139282226562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,12,12,128,1,float16,float16,0,8.751103973388672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,12,2,128,1,float16,float16,0,7.422515106201172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,12,4,128,1,float16,float16,0,7.403059387207032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,12,2,128,1,float16,float16,0,15.522297668457032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,12,4,128,1,float16,float16,0,15.798118591308594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,12,12,128,1,float16,float16,0,2.1518335342407227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,12,2,128,1,float16,float16,0,3.535103988647461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,12,1,128,1,float16,float16,0,3.45794563293457
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,12,4,128,1,float16,float16,0,3.770675277709961
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,12,1,128,1,float16,float16,0,1.7390592575073243
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,12,2,128,1,float16,float16,0,1.7764816284179688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,12,4,128,1,float16,float16,0,1.895475196838379
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,12,12,128,1,float16,float16,0,1.123840045928955
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,12,1,128,1,float16,float16,0,0.971827220916748
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,12,2,128,1,float16,float16,0,0.9639424324035645
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,12,4,128,1,float16,float16,0,0.9977855682373047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,12,1,128,1,float16,float16,0,3.847014236450195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,12,12,128,1,float16,float16,0,5.171200180053711
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,12,1,128,1,float16,float16,0,8.275302124023437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,12,2,128,1,float16,float16,0,8.699391937255859
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,12,4,128,1,float16,float16,0,8.964710235595703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,12,2,128,1,float16,float16,0,4.136806488037109
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,12,12,128,1,float16,float16,0,2.4774143218994142
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,12,4,128,1,float16,float16,0,4.33070068359375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,12,1,128,1,float16,float16,0,1.9717632293701173
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,12,2,128,1,float16,float16,0,2.002022361755371
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,12,4,128,1,float16,float16,0,2.149168014526367
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,12,12,128,1,float16,float16,0,1.3033984184265137
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,12,2,128,1,float16,float16,0,1.026251220703125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,12,12,128,1,float16,float16,0,0.6732800006866455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,12,1,128,1,float16,float16,0,1.0671584129333496
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,12,4,128,1,float16,float16,0,1.0959856033325195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,12,1,128,1,float16,float16,0,0.587827205657959
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,12,2,128,1,float16,float16,0,0.5925375938415527
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,12,4,128,1,float16,float16,0,0.5907968044281006
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,12,12,128,1,float16,float16,0,4.879462432861328
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,12,1,128,1,float16,float16,0,3.6438526153564452
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,12,2,128,1,float16,float16,0,7.879057312011719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,12,1,128,1,float16,float16,0,7.4388481140136715
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,12,4,128,1,float16,float16,0,8.550962829589844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,12,2,128,1,float16,float16,0,3.767603302001953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,12,12,128,1,float16,float16,0,2.520217514038086
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,12,4,128,1,float16,float16,0,4.161843109130859
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,12,1,128,1,float16,float16,0,1.7909759521484374
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,12,12,128,1,float16,float16,0,1.276313591003418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,12,2,128,1,float16,float16,0,1.8496000289916992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,12,1,128,1,float16,float16,0,0.950220775604248
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,12,2,128,1,float16,float16,0,0.9593855857849121
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,12,4,128,1,float16,float16,0,2.022086334228516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,12,4,128,1,float16,float16,0,1.0478079795837403
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,12,12,128,1,float16,float16,0,0.6681087970733642
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,12,1,128,1,float16,float16,0,0.5240320205688477
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,12,2,128,1,float16,float16,0,0.5119999885559082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,12,12,128,1,float16,float16,0,0.32445440292358396
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,12,1,128,1,float16,float16,0,0.2958336114883423
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,12,4,128,1,float16,float16,0,0.5138432025909424
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,12,4,128,1,float16,float16,0,0.30766079425811765
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,12,2,128,1,float16,float16,0,0.3042304039001465
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,12,1,128,1,float16,float16,0,4.35676155090332
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,12,1,128,1,float16,float16,0,2.0567039489746093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,12,2,128,1,float16,float16,0,4.542105484008789
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,12,12,128,1,float16,float16,0,3.1245311737060546
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,12,2,128,1,float16,float16,0,2.1578176498413084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,12,4,128,1,float16,float16,0,4.920729446411133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,12,4,128,1,float16,float16,0,2.4191999435424805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,12,12,128,1,float16,float16,0,1.5617024421691894
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,12,1,128,1,float16,float16,0,1.0573823928833008
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,12,2,128,1,float16,float16,0,1.1120639801025392
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,12,4,128,1,float16,float16,0,1.2076031684875488
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,12,1,128,1,float16,float16,0,0.5461503982543945
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,12,2,128,1,float16,float16,0,0.573798418045044
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,12,12,128,1,float16,float16,0,0.7965695858001709
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,12,4,128,1,float16,float16,0,0.631603193283081
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,12,12,128,1,float16,float16,0,0.40371198654174806
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,12,1,128,1,float16,float16,0,0.3110912084579468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,12,2,128,1,float16,float16,0,0.32419838905334475
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,12,12,128,1,float16,float16,0,0.21580801010131836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,12,1,128,1,float16,float16,0,0.19496959447860718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,12,4,128,1,float16,float16,0,0.32865281105041505
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,12,4,128,1,float16,float16,0,0.20894720554351806
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,12,2,128,1,float16,float16,0,0.20177919864654542
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,12,1,128,1,float16,float16,0,4.1274879455566404
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,12,2,128,1,float16,float16,0,4.363468933105469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,12,1,128,1,float16,float16,0,1.9502592086791992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,12,2,128,1,float16,float16,0,2.163711929321289
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,12,12,128,1,float16,float16,0,3.2697822570800783
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,12,4,128,1,float16,float16,0,2.3938560485839844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,12,4,128,1,float16,float16,0,4.904959869384766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,12,12,128,1,float16,float16,0,1.6132095336914063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,12,1,128,1,float16,float16,0,1.009663963317871
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,12,2,128,1,float16,float16,0,1.0808832168579101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,12,12,128,1,float16,float16,0,0.8404992103576661
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,12,4,128,1,float16,float16,0,1.1870719909667968
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,12,1,128,1,float16,float16,0,0.5347839832305908
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,12,2,128,1,float16,float16,0,0.5598720073699951
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,12,4,128,1,float16,float16,0,0.6278143882751465
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,12,12,128,1,float16,float16,0,0.4377583980560303
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,12,2,128,1,float16,float16,0,0.29071359634399413
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,12,1,128,1,float16,float16,0,0.2830847978591919
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,12,4,128,1,float16,float16,0,0.2954751968383789
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,12,1,128,1,float16,float16,0,0.1701375961303711
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,12,12,128,1,float16,float16,0,0.19353599548339845
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,12,2,128,1,float16,float16,0,0.17929760217666627
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,12,4,128,1,float16,float16,0,0.18227200508117675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,12,12,128,1,float16,float16,0,0.12933119535446166
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,12,1,128,1,float16,float16,0,0.1146880030632019
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,12,4,128,1,float16,float16,0,0.12226560115814208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,12,2,128,1,float16,float16,0,0.11653120517730713
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,12,1,128,1,float16,float16,0,2.3755775451660157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,12,2,128,1,float16,float16,0,2.534347152709961
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,12,12,128,1,float16,float16,0,2.126233673095703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,12,4,128,1,float16,float16,0,2.9627904891967773
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,12,1,128,1,float16,float16,0,1.1733455657958984
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,12,2,128,1,float16,float16,0,1.28788480758667
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,12,4,128,1,float16,float16,0,1.4840831756591797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,12,12,128,1,float16,float16,0,1.0743712425231933
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,12,1,128,1,float16,float16,0,0.6219776153564454
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,12,2,128,1,float16,float16,0,0.6593535900115967
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,12,4,128,1,float16,float16,0,0.7523839950561524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,12,12,128,1,float16,float16,0,0.5643263816833496
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,12,1,128,1,float16,float16,0,0.3156480073928833
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,12,2,128,1,float16,float16,0,0.32578558921813966
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,12,4,128,1,float16,float16,0,0.3857408046722412
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,12,1,128,1,float16,float16,0,0.18355200290679932
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,12,12,128,1,float16,float16,0,0.26869759559631345
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,12,2,128,1,float16,float16,0,0.19020800590515136
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,12,12,128,1,float16,float16,0,0.1358847975730896
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,12,4,128,1,float16,float16,0,0.19020800590515136
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,12,1,128,1,float16,float16,0,0.11658240556716919
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,12,4,128,1,float16,float16,0,0.12646399736404418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,12,2,128,1,float16,float16,0,0.11939840316772461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,12,1,128,1,float16,float16,0,0.07270399928092956
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,12,12,128,1,float16,float16,0,0.08427519798278808
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,12,2,128,1,float16,float16,0,0.07470080256462097
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,12,4,128,1,float16,float16,0,0.07649279832839966
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,12,1,128,1,float16,float16,0,2.657792091369629
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,12,2,128,1,float16,float16,0,2.9079023361206056
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,12,4,128,1,float16,float16,0,3.4131969451904296
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,12,1,128,1,float16,float16,0,1.3410816192626953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,12,12,128,1,float16,float16,0,2.6437631607055665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,12,2,128,1,float16,float16,0,1.4648320198059082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,12,4,128,1,float16,float16,0,1.7087999343872071
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,12,12,128,1,float16,float16,0,1.3279744148254395
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,12,1,128,1,float16,float16,0,0.679475212097168
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,12,1,128,1,float16,float16,0,0.3653120040893555
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,12,4,128,1,float16,float16,0,0.876796817779541
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,12,12,128,1,float16,float16,0,0.7160831928253174
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,12,2,128,1,float16,float16,0,0.7406079769134521
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,12,2,128,1,float16,float16,0,0.3842560052871704
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,12,4,128,1,float16,float16,0,0.443236780166626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,12,12,128,1,float16,float16,0,0.37862401008605956
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,12,1,128,1,float16,float16,0,0.18227200508117675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,12,2,128,1,float16,float16,0,0.18780159950256348
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,12,4,128,1,float16,float16,0,0.2038271903991699
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,12,12,128,1,float16,float16,0,0.15334399938583373
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,12,1,128,1,float16,float16,0,0.1112064003944397
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,12,2,128,1,float16,float16,0,0.11294080018997192
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,12,4,128,1,float16,float16,0,0.11678719520568848
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,12,12,128,1,float16,float16,0,0.09763839840888977
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,12,1,128,1,float16,float16,0,0.06507520079612732
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,12,2,128,1,float16,float16,0,0.06941919922828674
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,12,4,128,1,float16,float16,0,0.07203840017318726
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,12,1,128,1,float16,float16,0,0.050995200872421265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,12,12,128,1,float16,float16,0,0.07106559872627258
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,12,2,128,1,float16,float16,0,0.05642240047454834
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,12,4,128,1,float16,float16,0,0.06154239773750305
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,12,1,128,1,float16,float16,0,1.8214399337768554
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,12,2,128,1,float16,float16,0,2.0732927322387695
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,12,4,128,1,float16,float16,0,2.7868160247802733
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,12,12,128,1,float16,float16,0,2.591436767578125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,12,2,128,1,float16,float16,0,1.0505727767944335
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,12,1,128,1,float16,float16,0,0.9177087783813477
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,12,4,128,1,float16,float16,0,1.4126591682434082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,12,1,128,1,float16,float16,0,0.47631359100341797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,12,12,128,1,float16,float16,0,1.285529613494873
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,12,2,128,1,float16,float16,0,0.5327360153198242
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,12,4,128,1,float16,float16,0,0.7055871963500977
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,12,12,128,1,float16,float16,0,0.7051775932312012
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,12,1,128,1,float16,float16,0,0.25779199600219727
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,12,2,128,1,float16,float16,0,0.2825727939605713
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,12,4,128,1,float16,float16,0,0.3453439950942993
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,12,1,128,1,float16,float16,0,0.12989439964294433
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,12,12,128,1,float16,float16,0,0.31887359619140626
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,12,2,128,1,float16,float16,0,0.13414399623870848
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,12,4,128,1,float16,float16,0,0.15764479637145995
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,12,12,128,1,float16,float16,0,0.11576319932937622
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,12,1,128,1,float16,float16,0,0.07910400032997131
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,12,2,128,1,float16,float16,0,0.08304640054702758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,12,12,128,1,float16,float16,0,0.07249919772148132
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,12,4,128,1,float16,float16,0,0.08883200287818908
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,12,1,128,1,float16,float16,0,0.052019202709198
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,12,2,128,1,float16,float16,0,0.05294079780578613
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,12,4,128,1,float16,float16,0,0.056883198022842404
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,12,12,128,1,float16,float16,0,0.05939199924468994
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,12,1,128,1,float16,float16,0,0.04766719937324524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,12,2,128,1,float16,float16,0,0.048332801461219786
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,12,4,128,1,float16,float16,0,0.053504002094268796
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,12,12,128,1,float16,float16,0,0.038955199718475345
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,12,1,128,1,float16,float16,0,0.03558399975299835
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,12,2,128,1,float16,float16,0,0.036248001456260684
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,12,4,128,1,float16,float16,0,0.03671039938926697
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,12,1,128,1,float16,float16,0,0.7724991798400879
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,12,2,128,1,float16,float16,0,0.9285120010375977
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,12,4,128,1,float16,float16,0,1.2532735824584962
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,12,12,128,1,float16,float16,0,1.1877887725830079
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,12,1,128,1,float16,float16,0,0.3748863935470581
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,12,2,128,1,float16,float16,0,0.45311999320983887
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,12,4,128,1,float16,float16,0,0.6184447765350342
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,12,12,128,1,float16,float16,0,0.5933568000793457
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,12,1,128,1,float16,float16,0,0.18350080251693726
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,12,2,128,1,float16,float16,0,0.2076159954071045
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,12,4,128,1,float16,float16,0,0.27607040405273436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,12,12,128,1,float16,float16,0,0.24596478939056396
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,12,1,128,1,float16,float16,0,0.07823359966278076
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,12,2,128,1,float16,float16,0,0.08427519798278808
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,12,12,128,1,float16,float16,0,0.07838720083236694
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,12,4,128,1,float16,float16,0,0.09410560131072998
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,12,1,128,1,float16,float16,0,0.04787200093269348
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,12,4,128,1,float16,float16,0,0.05626879930496216
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,12,2,128,1,float16,float16,0,0.05079039931297302
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,12,12,128,1,float16,float16,0,0.04894720017910004
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,12,1,128,1,float16,float16,0,0.03404799997806549
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,12,4,128,1,float16,float16,0,0.039526399970054624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,12,2,128,1,float16,float16,0,0.03583360016345978
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,12,1,128,1,float16,float16,0,0.02529279887676239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,12,12,128,1,float16,float16,0,0.032153600454330446
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,12,4,128,1,float16,float16,0,0.027083200216293336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,12,2,128,1,float16,float16,0,0.025651198625564576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,12,12,128,1,float16,float16,0,0.027289599180221558
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,12,2,128,1,float16,float16,0,0.02467840015888214
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,12,1,128,1,float16,float16,0,0.024217599630355836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,12,1,128,1,float16,float16,0,0.022988800704479218
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,12,12,128,1,float16,float16,0,0.024268800020217897
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,12,4,128,1,float16,float16,0,0.025088000297546386
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,12,2,128,1,float16,float16,0,0.023449599742889404
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,12,4,128,1,float16,float16,0,0.023449599742889404
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,12,1,128,1,float16,float16,0,0.37780160903930665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,12,2,128,1,float16,float16,0,0.45552639961242675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,12,1,128,1,float16,float16,0,0.17607680559158326
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,12,12,128,1,float16,float16,0,0.5896192073822022
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,12,4,128,1,float16,float16,0,0.6156288146972656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,12,2,128,1,float16,float16,0,0.2027008056640625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,12,4,128,1,float16,float16,0,0.2706943988800049
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,12,12,128,1,float16,float16,0,0.2431999921798706
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,12,1,128,1,float16,float16,0,0.06609920263290406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,12,2,128,1,float16,float16,0,0.07147520184516906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,12,4,128,1,float16,float16,0,0.08238080143928528
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,12,12,128,1,float16,float16,0,0.07167999744415283
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,12,1,128,1,float16,float16,0,0.04111360013484955
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,12,2,128,1,float16,float16,0,0.04403199851512909
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,12,4,128,1,float16,float16,0,0.04971520006656647
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,12,12,128,1,float16,float16,0,0.045465600490570066
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,12,1,128,1,float16,float16,0,0.03107840120792389
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,12,2,128,1,float16,float16,0,0.03225600123405457
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,12,4,128,1,float16,float16,0,0.03680320084095001
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,12,12,128,1,float16,float16,0,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,12,4,128,1,float16,float16,0,0.023347200453281404
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,12,1,128,1,float16,float16,0,0.021555200219154358
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,12,2,128,1,float16,float16,0,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,12,12,128,1,float16,float16,0,0.023296000063419343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,12,1,128,1,float16,float16,0,0.020582400262355804
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,12,2,128,1,float16,float16,0,0.02088959962129593
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,12,4,128,1,float16,float16,0,0.02099200040102005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,12,12,128,1,float16,float16,0,0.020428800582885744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,12,2,128,1,float16,float16,0,0.019811199605464937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,12,1,128,1,float16,float16,0,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,12,4,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,12,12,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,12,1,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,12,2,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,12,4,128,1,float16,float16,0,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,12,1,128,1,float16,float16,0,0.17274880409240723
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,12,2,128,1,float16,float16,0,0.2066431999206543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,12,4,128,1,float16,float16,0,0.27612159252166746
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,12,12,128,1,float16,float16,0,0.25036799907684326
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,12,1,128,1,float16,float16,0,0.06405119895935059
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,12,2,128,1,float16,float16,0,0.07044640183448792
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,12,4,128,1,float16,float16,0,0.0927232027053833
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,12,12,128,1,float16,float16,0,0.0889855980873108
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,12,1,128,1,float16,float16,0,0.04039680063724518
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,12,2,128,1,float16,float16,0,0.042905598878860474
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,12,4,128,1,float16,float16,0,0.05437440276145935
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,12,1,128,1,float16,float16,0,0.02999520003795624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,12,2,128,1,float16,float16,0,0.03133440017700195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,12,4,128,1,float16,float16,0,0.03480960130691528
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,12,12,128,1,float16,float16,0,0.051097601652145386
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,12,12,128,1,float16,float16,0,0.034867200255393985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,12,1,128,1,float16,float16,0,0.021401600539684297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,12,2,128,1,float16,float16,0,0.021503999829292297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,12,12,128,1,float16,float16,0,0.023244799673557283
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,12,4,128,1,float16,float16,0,0.023296000063419343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,12,1,128,1,float16,float16,0,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,12,2,128,1,float16,float16,0,0.020377600193023683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,12,12,128,1,float16,float16,0,0.020019200444221497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,12,4,128,1,float16,float16,0,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,12,1,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,12,2,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,12,4,128,1,float16,float16,0,0.01966080069541931
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,12,12,128,1,float16,float16,0,0.01873600035905838
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,12,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,12,2,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,12,4,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,12,12,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,12,1,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,12,2,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,12,4,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,12,1,128,1,float16,float16,0,0.06502400040626526
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,12,2,128,1,float16,float16,0,0.0822272002696991
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,12,4,128,1,float16,float16,0,0.09492480158805847
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,12,12,128,1,float16,float16,0,0.12298239469528198
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,12,1,128,1,float16,float16,0,0.040243199467658995
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,12,2,128,1,float16,float16,0,0.04915199875831604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,12,4,128,1,float16,float16,0,0.0547327995300293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,12,12,128,1,float16,float16,0,0.071424001455307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,12,4,128,1,float16,float16,0,0.03430399894714355
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,12,1,128,1,float16,float16,0,0.030000001192092896
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,12,2,128,1,float16,float16,0,0.031231999397277832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,12,12,128,1,float16,float16,0,0.041771200299263
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,12,1,128,1,float16,float16,0,0.021350400149822236
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,12,2,128,1,float16,float16,0,0.02242559939622879
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,12,4,128,1,float16,float16,0,0.023705600202083586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,12,12,128,1,float16,float16,0,0.029849600791931153
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,12,1,128,1,float16,float16,0,0.02037599980831146
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,12,2,128,1,float16,float16,0,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,12,4,128,1,float16,float16,0,0.020428800582885744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,12,2,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,12,12,128,1,float16,float16,0,0.019865599274635316
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,12,1,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,12,4,128,1,float16,float16,0,0.019092799723148347
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,12,12,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,12,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,12,2,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,12,4,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,12,12,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,12,1,128,1,float16,float16,0,0.01873439997434616
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,12,2,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,12,12,128,1,float16,float16,0,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,12,4,128,1,float16,float16,0,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,12,1,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,12,2,128,1,float16,float16,0,0.01812320053577423
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,12,4,128,1,float16,float16,0,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,12,1,128,1,float16,float16,0,0.04715520143508911
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,12,12,128,1,float16,float16,0,0.10577759742736817
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,12,2,128,1,float16,float16,0,0.04940800070762634
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,12,4,128,1,float16,float16,0,0.08084480166435241
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,12,1,128,1,float16,float16,0,0.0304639995098114
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,12,2,128,1,float16,float16,0,0.031990399956703185
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,12,4,128,1,float16,float16,0,0.04746240079402923
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,12,12,128,1,float16,float16,0,0.06220800280570984
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,12,1,128,1,float16,float16,0,0.022169600427150726
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,12,2,128,1,float16,float16,0,0.022784000635147093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,12,4,128,1,float16,float16,0,0.029900801181793214
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,12,1,128,1,float16,float16,0,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,12,12,128,1,float16,float16,0,0.03701759874820709
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,12,2,128,1,float16,float16,0,0.020428800582885744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,12,4,128,1,float16,float16,0,0.021350400149822236
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,12,12,128,1,float16,float16,0,0.027033600211143493
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,12,1,128,1,float16,float16,0,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,12,2,128,1,float16,float16,0,0.019198399782180787
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,12,12,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,12,2,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,12,4,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,12,1,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,12,4,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,12,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,12,12,128,1,float16,float16,0,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,12,2,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,12,1,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,12,12,128,1,float16,float16,0,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,12,4,128,1,float16,float16,0,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,12,2,128,1,float16,float16,0,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,12,4,128,1,float16,float16,0,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,12,12,128,1,float16,float16,0,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,12,1,128,1,float16,float16,0,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,12,2,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,12,4,128,1,float16,float16,0,0.017763200402259826
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,8,1,128,1,float16,float16,0,9.357721710205078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,8,2,128,1,float16,float16,0,9.508403015136718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,8,8,128,1,float16,float16,0,10.351615905761719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,8,8,128,1,float16,float16,0,4.960307312011719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,8,4,128,1,float16,float16,0,9.847296142578125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,8,1,128,1,float16,float16,0,19.216998291015624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,8,2,128,1,float16,float16,0,19.446885681152345
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,8,4,128,1,float16,float16,0,20.143463134765625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,8,1,128,1,float16,float16,0,4.476467132568359
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,8,1,128,1,float16,float16,0,2.40133113861084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,8,2,128,1,float16,float16,0,4.509337615966797
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,8,8,128,1,float16,float16,0,2.588057518005371
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,8,4,128,1,float16,float16,0,4.814233779907227
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,8,2,128,1,float16,float16,0,2.441676712036133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,8,4,128,1,float16,float16,0,2.5601024627685547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,8,1,128,1,float16,float16,0,5.30145263671875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,8,8,128,1,float16,float16,0,5.802803039550781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,8,2,128,1,float16,float16,0,5.362636947631836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,8,1,128,1,float16,float16,0,10.794692993164062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,8,8,128,1,float16,float16,0,2.9138431549072266
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,8,2,128,1,float16,float16,0,10.740889739990234
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,8,4,128,1,float16,float16,0,5.330534362792969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,8,4,128,1,float16,float16,0,11.400396728515625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,8,1,128,1,float16,float16,0,2.56046085357666
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,8,2,128,1,float16,float16,0,2.7042816162109373
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,8,8,128,1,float16,float16,0,1.557248020172119
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,8,4,128,1,float16,float16,0,2.7548160552978516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,8,1,128,1,float16,float16,0,1.501747226715088
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,8,2,128,1,float16,float16,0,1.486899185180664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,8,4,128,1,float16,float16,0,1.49104642868042
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,8,1,128,1,float16,float16,0,7.461119842529297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,8,1,128,1,float16,float16,0,3.5777023315429686
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,8,2,128,1,float16,float16,0,3.6888065338134766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,8,8,128,1,float16,float16,0,4.239769744873047
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,8,2,128,1,float16,float16,0,7.583641815185547
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,8,8,128,1,float16,float16,0,2.105958366394043
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,8,4,128,1,float16,float16,0,7.95868148803711
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,8,4,128,1,float16,float16,0,3.890687942504883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,8,1,128,1,float16,float16,0,1.7856000900268554
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,8,2,128,1,float16,float16,0,1.8431999206542968
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,8,1,128,1,float16,float16,0,1.0043392181396484
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,8,8,128,1,float16,float16,0,1.1438591957092286
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,8,2,128,1,float16,float16,0,1.0655743598937988
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,8,4,128,1,float16,float16,0,1.024665641784668
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,8,4,128,1,float16,float16,0,1.9743215560913085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,8,1,128,1,float16,float16,0,9.669068908691406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,8,2,128,1,float16,float16,0,9.812735748291015
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,8,2,128,1,float16,float16,0,4.699238586425781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,8,1,128,1,float16,float16,0,4.643174362182617
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,8,8,128,1,float16,float16,0,2.8151296615600585
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,8,8,128,1,float16,float16,0,5.479475021362305
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,8,4,128,1,float16,float16,0,4.9688575744628904
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,8,4,128,1,float16,float16,0,10.693785858154296
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,8,1,128,1,float16,float16,0,2.383001518249512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,8,2,128,1,float16,float16,0,2.3640064239501952
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,8,4,128,1,float16,float16,0,2.521651268005371
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,8,8,128,1,float16,float16,0,1.458022403717041
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,8,2,128,1,float16,float16,0,1.1989503860473634
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,8,1,128,1,float16,float16,0,1.2762111663818358
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,8,8,128,1,float16,float16,0,0.754636812210083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,8,4,128,1,float16,float16,0,1.3188608169555665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,8,1,128,1,float16,float16,0,0.7150496006011963
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,8,2,128,1,float16,float16,0,0.703436803817749
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,8,4,128,1,float16,float16,0,0.7102975845336914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,8,8,128,1,float16,float16,0,3.3512958526611327
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,8,1,128,1,float16,float16,0,2.6390527725219726
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,8,1,128,1,float16,float16,0,5.381119918823242
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,8,2,128,1,float16,float16,0,5.499135971069336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,8,2,128,1,float16,float16,0,2.6389503479003906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,8,4,128,1,float16,float16,0,6.248191833496094
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,8,4,128,1,float16,float16,0,2.9660160064697267
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,8,8,128,1,float16,float16,0,1.6883712768554688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,8,1,128,1,float16,float16,0,1.3507583618164063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,8,2,128,1,float16,float16,0,1.433852767944336
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,8,8,128,1,float16,float16,0,0.8958975791931152
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,8,4,128,1,float16,float16,0,1.4968832015991211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,8,1,128,1,float16,float16,0,0.7803391933441162
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,8,2,128,1,float16,float16,0,0.7676415920257569
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,8,4,128,1,float16,float16,0,0.7578112125396729
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,8,8,128,1,float16,float16,0,0.4517888069152832
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,8,1,128,1,float16,float16,0,0.41850881576538085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,8,2,128,1,float16,float16,0,0.4364287853240967
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,8,4,128,1,float16,float16,0,0.44257278442382814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,8,1,128,1,float16,float16,0,5.003724670410156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,8,8,128,1,float16,float16,0,3.299737548828125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,8,2,128,1,float16,float16,0,5.232486343383789
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,8,4,128,1,float16,float16,0,5.775564956665039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,8,1,128,1,float16,float16,0,2.4153600692749024
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,8,2,128,1,float16,float16,0,2.55098876953125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,8,8,128,1,float16,float16,0,1.6550912857055664
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,8,4,128,1,float16,float16,0,2.859462356567383
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,8,1,128,1,float16,float16,0,1.2256768226623536
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,8,2,128,1,float16,float16,0,1.2925439834594727
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,8,4,128,1,float16,float16,0,1.4195712089538575
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,8,8,128,1,float16,float16,0,0.8756735801696778
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,8,1,128,1,float16,float16,0,0.6640128135681153
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,8,2,128,1,float16,float16,0,0.6500271797180176
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,8,4,128,1,float16,float16,0,0.7360511779785156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,8,8,128,1,float16,float16,0,0.4254720211029053
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,8,1,128,1,float16,float16,0,0.36567039489746095
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,8,2,128,1,float16,float16,0,0.37411839962005616
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,8,8,128,1,float16,float16,0,0.2511359930038452
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,8,4,128,1,float16,float16,0,0.38348801136016847
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,8,2,128,1,float16,float16,0,0.24104959964752198
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,8,1,128,1,float16,float16,0,0.22553598880767822
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,8,4,128,1,float16,float16,0,0.24678399562835693
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,8,1,128,1,float16,float16,0,2.8185087203979493
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,8,2,128,1,float16,float16,0,3.029964828491211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,8,4,128,1,float16,float16,0,3.3880577087402344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,8,1,128,1,float16,float16,0,1.4359552383422851
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,8,8,128,1,float16,float16,0,2.034022331237793
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,8,2,128,1,float16,float16,0,1.5218688011169434
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,8,4,128,1,float16,float16,0,1.7181184768676758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,8,8,128,1,float16,float16,0,1.0605055809020996
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,8,1,128,1,float16,float16,0,0.73405442237854
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,8,2,128,1,float16,float16,0,0.7816703796386719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,8,4,128,1,float16,float16,0,0.8853504180908203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,8,8,128,1,float16,float16,0,0.5584896087646485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,8,2,128,1,float16,float16,0,0.40657920837402345
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,8,8,128,1,float16,float16,0,0.25835518836975097
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,8,1,128,1,float16,float16,0,0.40140318870544434
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,8,4,128,1,float16,float16,0,0.42444801330566406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,8,1,128,1,float16,float16,0,0.237824010848999
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,8,2,128,1,float16,float16,0,0.2440704107284546
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,8,4,128,1,float16,float16,0,0.24924159049987793
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,8,8,128,1,float16,float16,0,0.1726464033126831
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,8,1,128,1,float16,float16,0,0.157695996761322
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,8,2,128,1,float16,float16,0,0.15964159965515137
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,8,4,128,1,float16,float16,0,0.16440320014953613
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,8,1,128,1,float16,float16,0,2.7016191482543945
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,8,2,128,1,float16,float16,0,2.8996608734130858
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,8,4,128,1,float16,float16,0,3.442534255981445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,8,1,128,1,float16,float16,0,1.3840383529663085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,8,8,128,1,float16,float16,0,2.1622783660888674
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,8,2,128,1,float16,float16,0,1.4915583610534668
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,8,4,128,1,float16,float16,0,1.7210880279541017
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,8,8,128,1,float16,float16,0,1.1017215728759766
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,8,1,128,1,float16,float16,0,0.6957568168640137
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,8,2,128,1,float16,float16,0,0.7554048061370849
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,8,8,128,1,float16,float16,0,0.5727744102478027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,8,1,128,1,float16,float16,0,0.35665919780731203
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,8,4,128,1,float16,float16,0,0.8876543998718261
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,8,2,128,1,float16,float16,0,0.3788288116455078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,8,4,128,1,float16,float16,0,0.44810237884521487
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,8,8,128,1,float16,float16,0,0.255948805809021
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,8,1,128,1,float16,float16,0,0.21206879615783691
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,8,2,128,1,float16,float16,0,0.21765120029449464
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,8,4,128,1,float16,float16,0,0.22476799488067628
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,8,8,128,1,float16,float16,0,0.15155199766159058
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,8,1,128,1,float16,float16,0,0.13598719835281373
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,8,2,128,1,float16,float16,0,0.14218239784240722
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,8,4,128,1,float16,float16,0,0.14520319700241088
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,8,8,128,1,float16,float16,0,0.09097920060157776
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,8,1,128,1,float16,float16,0,0.08171520233154297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,8,2,128,1,float16,float16,0,0.08115199804306031
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,8,4,128,1,float16,float16,0,0.08616960048675537
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,8,1,128,1,float16,float16,0,1.8002944946289063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,8,2,128,1,float16,float16,0,1.9978752136230469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,8,4,128,1,float16,float16,0,2.3571456909179687
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,8,8,128,1,float16,float16,0,1.5465984344482422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,8,1,128,1,float16,float16,0,0.9142784118652344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,8,2,128,1,float16,float16,0,0.9987520217895508
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,8,4,128,1,float16,float16,0,1.1743743896484375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,8,8,128,1,float16,float16,0,0.7907328128814697
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,8,1,128,1,float16,float16,0,0.47815680503845215
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,8,2,128,1,float16,float16,0,0.5228032112121582
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,8,4,128,1,float16,float16,0,0.6231040000915528
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,8,8,128,1,float16,float16,0,0.425216007232666
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,8,1,128,1,float16,float16,0,0.248524808883667
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,8,8,128,1,float16,float16,0,0.16993279457092286
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,8,4,128,1,float16,float16,0,0.30458879470825195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,8,2,128,1,float16,float16,0,0.25528318881988527
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,8,1,128,1,float16,float16,0,0.14566400051116943
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,8,2,128,1,float16,float16,0,0.14545919895172119
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,8,4,128,1,float16,float16,0,0.1541632056236267
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,8,8,128,1,float16,float16,0,0.09917439818382263
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,8,2,128,1,float16,float16,0,0.09564160108566284
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,8,1,128,1,float16,float16,0,0.09323520064353943
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,8,4,128,1,float16,float16,0,0.0919535994529724
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,8,8,128,1,float16,float16,0,0.08023040294647217
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,8,1,128,1,float16,float16,0,0.059699201583862306
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,8,2,128,1,float16,float16,0,0.06896640062332153
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,8,4,128,1,float16,float16,0,0.0754688024520874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,8,1,128,1,float16,float16,0,1.834649658203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,8,2,128,1,float16,float16,0,2.0767232894897463
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,8,4,128,1,float16,float16,0,2.524671936035156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,8,2,128,1,float16,float16,0,1.0494976043701172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,8,1,128,1,float16,float16,0,0.9279999732971191
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,8,8,128,1,float16,float16,0,1.724825668334961
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,8,4,128,1,float16,float16,0,1.2777471542358398
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,8,8,128,1,float16,float16,0,0.8869888305664062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,8,1,128,1,float16,float16,0,0.48849921226501464
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,8,2,128,1,float16,float16,0,0.5463039875030518
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,8,1,128,1,float16,float16,0,0.2506239891052246
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,8,8,128,1,float16,float16,0,0.47457280158996584
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,8,4,128,1,float16,float16,0,0.6602752208709717
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,8,2,128,1,float16,float16,0,0.27688961029052733
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,8,4,128,1,float16,float16,0,0.3473328113555908
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,8,8,128,1,float16,float16,0,0.22072319984436034
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,8,2,128,1,float16,float16,0,0.14673919677734376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,8,1,128,1,float16,float16,0,0.14228479862213134
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,8,4,128,1,float16,float16,0,0.15800319910049437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,8,8,128,1,float16,float16,0,0.11084799766540528
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,8,1,128,1,float16,float16,0,0.09599999785423279
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,8,4,128,1,float16,float16,0,0.09953280091285706
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,8,2,128,1,float16,float16,0,0.09748479723930359
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,8,8,128,1,float16,float16,0,0.07214080095291138
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,8,2,128,1,float16,float16,0,0.06425600051879883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,8,1,128,1,float16,float16,0,0.06108160018920898
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,8,4,128,1,float16,float16,0,0.06686720252037048
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,8,8,128,1,float16,float16,0,0.054318398237228394
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,8,1,128,1,float16,float16,0,0.04802559912204742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,8,4,128,1,float16,float16,0,0.0492031991481781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,8,2,128,1,float16,float16,0,0.04802559912204742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,8,1,128,1,float16,float16,0,1.3109248161315918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,8,2,128,1,float16,float16,0,1.6336896896362305
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,8,8,128,1,float16,float16,0,1.6973312377929688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,8,4,128,1,float16,float16,0,2.2598640441894533
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,8,1,128,1,float16,float16,0,0.6681087970733642
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,8,2,128,1,float16,float16,0,0.7921152114868164
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,8,4,128,1,float16,float16,0,1.1240447998046874
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,8,8,128,1,float16,float16,0,0.8497152328491211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,8,1,128,1,float16,float16,0,0.3474431991577148
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,8,2,128,1,float16,float16,0,0.410368013381958
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,8,4,128,1,float16,float16,0,0.5895679950714111
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,8,1,128,1,float16,float16,0,0.1748479962348938
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,8,8,128,1,float16,float16,0,0.45496320724487305
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,8,4,128,1,float16,float16,0,0.2688512086868286
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,8,2,128,1,float16,float16,0,0.2016767978668213
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,8,8,128,1,float16,float16,0,0.18688000440597535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,8,1,128,1,float16,float16,0,0.10106879472732544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,8,2,128,1,float16,float16,0,0.10434559583663941
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,8,4,128,1,float16,float16,0,0.11576319932937622
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,8,8,128,1,float16,float16,0,0.0871936023235321
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,8,1,128,1,float16,float16,0,0.062054401636123656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,8,2,128,1,float16,float16,0,0.06528000235557556
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,8,8,128,1,float16,float16,0,0.05324800014495849
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,8,4,128,1,float16,float16,0,0.07316480278968811
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,8,1,128,1,float16,float16,0,0.043007999658584595
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,8,2,128,1,float16,float16,0,0.04469760060310364
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,8,4,128,1,float16,float16,0,0.04741120040416717
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,8,8,128,1,float16,float16,0,0.0447488009929657
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,8,1,128,1,float16,float16,0,0.03993600010871887
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,8,2,128,1,float16,float16,0,0.040191999077796935
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,8,4,128,1,float16,float16,0,0.04142079949378967
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,8,1,128,1,float16,float16,0,0.03358719944953918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,8,8,128,1,float16,float16,0,0.03527520000934601
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,8,2,128,1,float16,float16,0,0.03338240087032318
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,8,4,128,1,float16,float16,0,0.03430399894714355
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,8,1,128,1,float16,float16,0,0.5610400199890136
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,8,2,128,1,float16,float16,0,0.7192575931549072
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,8,2,128,1,float16,float16,0,0.3327984094619751
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,8,4,128,1,float16,float16,0,1.0316800117492675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,8,8,128,1,float16,float16,0,0.8075263977050782
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,8,1,128,1,float16,float16,0,0.2624000072479248
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,8,8,128,1,float16,float16,0,0.3719167947769165
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,8,4,128,1,float16,float16,0,0.502016019821167
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,8,1,128,1,float16,float16,0,0.10398720502853394
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,8,2,128,1,float16,float16,0,0.11668479442596436
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,8,4,128,1,float16,float16,0,0.2006527900695801
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,8,8,128,1,float16,float16,0,0.10383360385894776
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,8,1,128,1,float16,float16,0,0.06077439785003662
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,8,4,128,1,float16,float16,0,0.07680000066757202
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,8,2,128,1,float16,float16,0,0.06661120057106018
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,8,8,128,1,float16,float16,0,0.05836799740791321
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,8,1,128,1,float16,float16,0,0.04162560105323791
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,8,2,128,1,float16,float16,0,0.04254719913005829
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,8,4,128,1,float16,float16,0,0.04782080054283142
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,8,8,128,1,float16,float16,0,0.03594239950180054
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,8,1,128,1,float16,float16,0,0.02687999904155731
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,8,2,128,1,float16,float16,0,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,8,4,128,1,float16,float16,0,0.030720001459121703
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,8,1,128,1,float16,float16,0,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,8,8,128,1,float16,float16,0,0.029235199093818665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,8,2,128,1,float16,float16,0,0.024883200228214265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,8,4,128,1,float16,float16,0,0.027033600211143493
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,8,8,128,1,float16,float16,0,0.026060798764228822
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,8,2,128,1,float16,float16,0,0.024268800020217897
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,8,1,128,1,float16,float16,0,0.024063999950885772
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,8,4,128,1,float16,float16,0,0.02452480047941208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,8,8,128,1,float16,float16,0,0.023552000522613525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,8,1,128,1,float16,float16,0,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,8,2,128,1,float16,float16,0,0.02314240038394928
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,8,4,128,1,float16,float16,0,0.023552000522613525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,8,1,128,1,float16,float16,0,0.2606591939926147
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,8,2,128,1,float16,float16,0,0.3325952053070068
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,8,4,128,1,float16,float16,0,0.501043176651001
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,8,1,128,1,float16,float16,0,0.08529279828071594
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,8,8,128,1,float16,float16,0,0.3662336111068726
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,8,2,128,1,float16,float16,0,0.10183680057525635
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,8,4,128,1,float16,float16,0,0.1965567946434021
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,8,8,128,1,float16,float16,0,0.09850879907608032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,8,1,128,1,float16,float16,0,0.0517632007598877
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,8,2,128,1,float16,float16,0,0.05713919997215271
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,8,4,128,1,float16,float16,0,0.06783999800682068
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,8,8,128,1,float16,float16,0,0.05242879986763001
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,8,1,128,1,float16,float16,0,0.03368000090122223
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,8,2,128,1,float16,float16,0,0.0361952006816864
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,8,8,128,1,float16,float16,0,0.032400000095367434
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,8,4,128,1,float16,float16,0,0.04152320027351379
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,8,1,128,1,float16,float16,0,0.022988800704479218
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,8,4,128,1,float16,float16,0,0.027238398790359497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,8,2,128,1,float16,float16,0,0.024831999838352204
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,8,8,128,1,float16,float16,0,0.025088000297546386
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,8,1,128,1,float16,float16,0,0.020838400721549986
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,8,4,128,1,float16,float16,0,0.022681599855422972
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,8,2,128,1,float16,float16,0,0.02094080001115799
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,8,8,128,1,float16,float16,0,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,8,2,128,1,float16,float16,0,0.020633600652217865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,8,1,128,1,float16,float16,0,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,8,4,128,1,float16,float16,0,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,8,8,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,8,1,128,1,float16,float16,0,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,8,2,128,1,float16,float16,0,0.01966080069541931
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,8,8,128,1,float16,float16,0,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,8,4,128,1,float16,float16,0,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,8,1,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,8,2,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,8,4,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,8,1,128,1,float16,float16,0,0.08591359853744507
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,8,4,128,1,float16,float16,0,0.19860479831695557
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,8,2,128,1,float16,float16,0,0.10536960363388062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,8,1,128,1,float16,float16,0,0.050379198789596555
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,8,8,128,1,float16,float16,0,0.11612160205841064
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,8,2,128,1,float16,float16,0,0.055961602926254274
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,8,4,128,1,float16,float16,0,0.06702079772949218
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,8,8,128,1,float16,float16,0,0.06384639739990235
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,8,1,128,1,float16,float16,0,0.03292160034179688
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,8,2,128,1,float16,float16,0,0.035729598999023435
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,8,4,128,1,float16,float16,0,0.040550398826599124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,8,1,128,1,float16,float16,0,0.023296000063419343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,8,2,128,1,float16,float16,0,0.024985599517822265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,8,8,128,1,float16,float16,0,0.03809280097484589
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,8,4,128,1,float16,float16,0,0.02682879865169525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,8,8,128,1,float16,float16,0,0.026009601354599
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,8,1,128,1,float16,float16,0,0.020735999941825865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,8,2,128,1,float16,float16,0,0.02114560008049011
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,8,4,128,1,float16,float16,0,0.02227199971675873
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,8,8,128,1,float16,float16,0,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,8,1,128,1,float16,float16,0,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,8,2,128,1,float16,float16,0,0.020633600652217865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,8,4,128,1,float16,float16,0,0.020531199872493744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,8,8,128,1,float16,float16,0,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,8,1,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,8,2,128,1,float16,float16,0,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,8,8,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,8,4,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,8,1,128,1,float16,float16,0,0.018326400220394133
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,8,4,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,8,2,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,8,8,128,1,float16,float16,0,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,8,1,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,8,2,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,8,4,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,8,1,128,1,float16,float16,0,0.050995200872421265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,8,2,128,1,float16,float16,0,0.056576001644134524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,8,4,128,1,float16,float16,0,0.09241600036621093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,8,2,128,1,float16,float16,0,0.03614720106124878
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,8,1,128,1,float16,float16,0,0.032767999172210696
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,8,8,128,1,float16,float16,0,0.09031680226325989
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,8,4,128,1,float16,float16,0,0.053299200534820554
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,8,8,128,1,float16,float16,0,0.051507198810577394
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,8,1,128,1,float16,float16,0,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,8,2,128,1,float16,float16,0,0.024422399699687958
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,8,4,128,1,float16,float16,0,0.03297280073165894
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,8,8,128,1,float16,float16,0,0.032051199674606325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,8,2,128,1,float16,float16,0,0.02109439969062805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,8,1,128,1,float16,float16,0,0.020787200331687926
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,8,4,128,1,float16,float16,0,0.023289600014686586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,8,8,128,1,float16,float16,0,0.02227199971675873
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,8,1,128,1,float16,float16,0,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,8,2,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,8,8,128,1,float16,float16,0,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,8,4,128,1,float16,float16,0,0.020172800123691558
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,8,1,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,8,2,128,1,float16,float16,0,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,8,4,128,1,float16,float16,0,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,8,8,128,1,float16,float16,0,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,8,1,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,8,2,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,8,4,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,8,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,8,2,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,8,8,128,1,float16,float16,0,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,8,8,128,1,float16,float16,0,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,8,4,128,1,float16,float16,0,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,8,1,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,8,2,128,1,float16,float16,0,0.018275199830532073
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,8,4,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,8,1,128,1,float16,float16,0,0.033939200639724734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,8,8,128,1,float16,float16,0,0.07915520071983337
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,8,2,128,1,float16,float16,0,0.04910080134868622
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,8,4,128,1,float16,float16,0,0.08125439882278443
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,8,1,128,1,float16,float16,0,0.023647999763488768
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,8,2,128,1,float16,float16,0,0.030822399258613586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,8,8,128,1,float16,float16,0,0.04622560143470764
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,8,4,128,1,float16,float16,0,0.04659200012683869
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,8,1,128,1,float16,float16,0,0.021299199759960176
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,8,2,128,1,float16,float16,0,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,8,4,128,1,float16,float16,0,0.029481598734855653
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,8,8,128,1,float16,float16,0,0.02887359857559204
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,8,1,128,1,float16,float16,0,0.020735999941825865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,8,2,128,1,float16,float16,0,0.019968000054359437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,8,4,128,1,float16,float16,0,0.021299199759960176
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,8,8,128,1,float16,float16,0,0.019763199985027312
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,8,1,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,8,2,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,8,4,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,8,8,128,1,float16,float16,0,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,8,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,8,2,128,1,float16,float16,0,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,8,4,128,1,float16,float16,0,0.018321600556373597
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,8,8,128,1,float16,float16,0,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,8,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,8,2,128,1,float16,float16,0,0.01801919937133789
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,8,4,128,1,float16,float16,0,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,8,8,128,1,float16,float16,0,0.017603200674057008
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,8,1,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,8,2,128,1,float16,float16,0,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,8,4,128,1,float16,float16,0,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,8,8,128,1,float16,float16,0,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,8,1,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,8,2,128,1,float16,float16,0,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,8,4,128,1,float16,float16,0,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,4,1,128,1,float16,float16,0,4.66165771484375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,4,4,128,1,float16,float16,0,4.980326461791992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,4,4,128,1,float16,float16,0,2.581504058837891
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,4,1,128,1,float16,float16,0,2.4057344436645507
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,4,1,128,1,float16,float16,0,9.534361267089844
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,4,2,128,1,float16,float16,0,4.686028671264649
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,4,2,128,1,float16,float16,0,2.4919551849365233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,4,2,128,1,float16,float16,0,9.689651489257812
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,4,1,128,1,float16,float16,0,1.3099519729614257
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,4,4,128,1,float16,float16,0,1.3574655532836915
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,4,2,128,1,float16,float16,0,1.3932543754577638
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,4,1,128,1,float16,float16,0,5.265910339355469
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,4,1,128,1,float16,float16,0,2.6280448913574217
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,4,4,128,1,float16,float16,0,2.868172836303711
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,4,2,128,1,float16,float16,0,5.540249633789062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,4,2,128,1,float16,float16,0,2.721433639526367
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,4,1,128,1,float16,float16,0,1.4656000137329102
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,4,2,128,1,float16,float16,0,1.4322688102722168
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,4,4,128,1,float16,float16,0,1.5399423599243165
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,4,4,128,1,float16,float16,0,0.8349184036254883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,4,1,128,1,float16,float16,0,0.8115712165832519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,4,2,128,1,float16,float16,0,0.8539135932922364
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,4,1,128,1,float16,float16,0,3.785932922363281
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,4,2,128,1,float16,float16,0,3.744406509399414
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,4,4,128,1,float16,float16,0,2.120550346374512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,4,2,128,1,float16,float16,0,1.965260887145996
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,4,1,128,1,float16,float16,0,1.863167953491211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,4,4,128,1,float16,float16,0,1.1493887901306152
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,4,1,128,1,float16,float16,0,1.0380288124084474
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,4,2,128,1,float16,float16,0,1.0819071769714355
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,4,4,128,1,float16,float16,0,0.6395391941070556
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,4,2,128,1,float16,float16,0,0.6116352081298828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,4,1,128,1,float16,float16,0,0.6071296215057373
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,4,1,128,1,float16,float16,0,4.661759948730468
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,4,2,128,1,float16,float16,0,5.1231742858886715
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,4,1,128,1,float16,float16,0,2.348748779296875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,4,4,128,1,float16,float16,0,2.7664384841918945
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,4,2,128,1,float16,float16,0,2.459903907775879
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,4,4,128,1,float16,float16,0,1.4246399879455567
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,4,2,128,1,float16,float16,0,1.3042688369750977
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,4,1,128,1,float16,float16,0,1.1909631729125976
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,4,4,128,1,float16,float16,0,0.7374335765838623
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,4,1,128,1,float16,float16,0,0.714188814163208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,4,2,128,1,float16,float16,0,0.7057407855987549
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,4,4,128,1,float16,float16,0,0.44820480346679686
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,4,1,128,1,float16,float16,0,0.41318402290344236
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,4,2,128,1,float16,float16,0,0.43279361724853516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,4,1,128,1,float16,float16,0,2.691584014892578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,4,2,128,1,float16,float16,0,2.9046783447265625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,4,1,128,1,float16,float16,0,1.3917695999145507
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,4,4,128,1,float16,float16,0,1.6780288696289063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,4,2,128,1,float16,float16,0,1.4937583923339843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,4,4,128,1,float16,float16,0,0.8905728340148926
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,4,1,128,1,float16,float16,0,0.7825407981872559
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,4,2,128,1,float16,float16,0,0.7449600219726562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,4,4,128,1,float16,float16,0,0.4521471977233887
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,4,1,128,1,float16,float16,0,0.4386303901672363
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,4,4,128,1,float16,float16,0,0.28999040126800535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,4,1,128,1,float16,float16,0,0.2794496059417725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,4,2,128,1,float16,float16,0,0.2832895994186401
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,4,2,128,1,float16,float16,0,0.44605441093444825
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,4,1,128,1,float16,float16,0,2.494361686706543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,4,2,128,1,float16,float16,0,2.770534324645996
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,4,4,128,1,float16,float16,0,1.6733695983886718
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,4,2,128,1,float16,float16,0,0.7382527828216553
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,4,1,128,1,float16,float16,0,0.6548992156982422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,4,2,128,1,float16,float16,0,1.4298080444335937
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,4,4,128,1,float16,float16,0,0.8642047882080078
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,4,1,128,1,float16,float16,0,1.3054464340209961
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,4,4,128,1,float16,float16,0,0.4196864128112793
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,4,1,128,1,float16,float16,0,0.37893118858337405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,4,2,128,1,float16,float16,0,0.3810816049575806
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,4,4,128,1,float16,float16,0,0.2521087884902954
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,4,1,128,1,float16,float16,0,0.24053759574890138
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,4,4,128,1,float16,float16,0,0.1557952046394348
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,4,1,128,1,float16,float16,0,0.15293439626693725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,4,2,128,1,float16,float16,0,0.24580800533294678
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,4,2,128,1,float16,float16,0,0.1523200035095215
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,4,1,128,1,float16,float16,0,1.63768310546875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,4,2,128,1,float16,float16,0,1.8222080230712892
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,4,4,128,1,float16,float16,0,1.1133952140808105
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,4,1,128,1,float16,float16,0,0.8499711990356446
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,4,2,128,1,float16,float16,0,0.9298432350158692
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,4,4,128,1,float16,float16,0,0.5874688148498535
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,4,1,128,1,float16,float16,0,0.4246528148651123
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,4,2,128,1,float16,float16,0,0.4730832099914551
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,4,4,128,1,float16,float16,0,0.26946558952331545
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,4,1,128,1,float16,float16,0,0.25175039768218993
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,4,4,128,1,float16,float16,0,0.15708160400390625
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,4,2,128,1,float16,float16,0,0.2535936117172241
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,4,1,128,1,float16,float16,0,0.15201280117034913
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,4,2,128,1,float16,float16,0,0.15211360454559325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,4,4,128,1,float16,float16,0,0.11514879465103149
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,4,1,128,1,float16,float16,0,0.09651200175285339
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,4,2,128,1,float16,float16,0,0.0989184021949768
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,4,1,128,1,float16,float16,0,1.6184303283691406
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,4,4,128,1,float16,float16,0,1.181644821166992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,4,1,128,1,float16,float16,0,0.8308688163757324
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,4,2,128,1,float16,float16,0,1.8476032257080077
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,4,2,128,1,float16,float16,0,0.9449983596801758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,4,4,128,1,float16,float16,0,0.6174208164215088
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,4,1,128,1,float16,float16,0,0.4220928192138672
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,4,2,128,1,float16,float16,0,0.4933119773864746
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,4,4,128,1,float16,float16,0,0.3084752082824707
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,4,1,128,1,float16,float16,0,0.23685119152069092
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,4,2,128,1,float16,float16,0,0.2411520004272461
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,4,4,128,1,float16,float16,0,0.1637887954711914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,4,1,128,1,float16,float16,0,0.1506816029548645
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,4,2,128,1,float16,float16,0,0.15308799743652343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,4,4,128,1,float16,float16,0,0.10803200006484985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,4,1,128,1,float16,float16,0,0.09779199957847595
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,4,2,128,1,float16,float16,0,0.10393600463867188
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,4,4,128,1,float16,float16,0,0.071424001455307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,4,1,128,1,float16,float16,0,0.06742240190505981
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,4,2,128,1,float16,float16,0,0.06778879761695862
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,4,1,128,1,float16,float16,0,1.0249728202819823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,4,4,128,1,float16,float16,0,0.7922688007354737
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,4,1,128,1,float16,float16,0,0.5253119945526123
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,4,2,128,1,float16,float16,0,0.616703987121582
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,4,2,128,1,float16,float16,0,1.1914239883422852
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,4,4,128,1,float16,float16,0,0.4200448036193848
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,4,2,128,1,float16,float16,0,0.30883839130401614
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,4,1,128,1,float16,float16,0,0.26024959087371824
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,4,4,128,1,float16,float16,0,0.18472959995269775
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,4,1,128,1,float16,float16,0,0.16112639904022216
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,4,4,128,1,float16,float16,0,0.11089919805526734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,4,2,128,1,float16,float16,0,0.16240639686584474
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,4,1,128,1,float16,float16,0,0.10158079862594604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,4,2,128,1,float16,float16,0,0.10808320045471191
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,4,4,128,1,float16,float16,0,0.08053759932518005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,4,2,128,1,float16,float16,0,0.07557119727134705
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,4,4,128,1,float16,float16,0,0.051046401262283325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,4,2,128,1,float16,float16,0,0.04935680031776428
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,4,1,128,1,float16,float16,0,0.04797439873218536
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,4,1,128,1,float16,float16,0,0.06901760101318359
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,4,1,128,1,float16,float16,0,1.0643391609191895
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,4,2,128,1,float16,float16,0,1.295257568359375
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,4,1,128,1,float16,float16,0,0.5483520030975342
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,4,4,128,1,float16,float16,0,0.4686768054962158
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,4,4,128,1,float16,float16,0,0.8903679847717285
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,4,2,128,1,float16,float16,0,0.6643199920654297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,4,2,128,1,float16,float16,0,0.3441152095794678
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,4,1,128,1,float16,float16,0,0.27514560222625734
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,4,1,128,1,float16,float16,0,0.15349760055541992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,4,4,128,1,float16,float16,0,0.22865920066833495
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,4,2,128,1,float16,float16,0,0.1633792042732239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,4,2,128,1,float16,float16,0,0.107315194606781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,4,1,128,1,float16,float16,0,0.10618879795074462
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,4,4,128,1,float16,float16,0,0.11473920345306396
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,4,4,128,1,float16,float16,0,0.07193599939346314
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,4,1,128,1,float16,float16,0,0.06364160180091857
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,4,2,128,1,float16,float16,0,0.06671199798583985
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,4,4,128,1,float16,float16,0,0.05231680274009705
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,4,1,128,1,float16,float16,0,0.048127999901771544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,4,2,128,1,float16,float16,0,0.04956159889698029
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,4,4,128,1,float16,float16,0,0.040038400888442995
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,4,1,128,1,float16,float16,0,0.03804160058498383
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,4,2,128,1,float16,float16,0,0.03860319852828979
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,4,1,128,1,float16,float16,0,0.8041983604431152
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,4,2,128,1,float16,float16,0,1.1502592086791992
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,4,4,128,1,float16,float16,0,0.8536064147949218
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,4,1,128,1,float16,float16,0,0.40560641288757326
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,4,2,128,1,float16,float16,0,0.5522431850433349
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,4,4,128,1,float16,float16,0,0.42024960517883303
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,4,1,128,1,float16,float16,0,0.20060160160064697
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,4,2,128,1,float16,float16,0,0.2673664093017578
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,4,1,128,1,float16,float16,0,0.10593279600143432
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,4,4,128,1,float16,float16,0,0.18959360122680663
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,4,2,128,1,float16,float16,0,0.11781120300292969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,4,4,128,1,float16,float16,0,0.08734719753265381
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,4,1,128,1,float16,float16,0,0.07045120000839233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,4,2,128,1,float16,float16,0,0.0745967984199524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,4,4,128,1,float16,float16,0,0.05304319858551025
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,4,1,128,1,float16,float16,0,0.04485119879245758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,4,2,128,1,float16,float16,0,0.04756479859352112
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,4,4,128,1,float16,float16,0,0.04473919868469238
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,4,2,128,1,float16,float16,0,0.04152320027351379
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,4,1,128,1,float16,float16,0,0.04028800129890442
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,4,4,128,1,float16,float16,0,0.035524800419807434
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,4,1,128,1,float16,float16,0,0.03343360126018524
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,4,2,128,1,float16,float16,0,0.03415040075778961
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,4,4,128,1,float16,float16,0,0.02903040051460266
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,4,1,128,1,float16,float16,0,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,4,2,128,1,float16,float16,0,0.028518399596214293
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,4,2,128,1,float16,float16,0,0.4991456031799316
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,4,4,128,1,float16,float16,0,0.37144639492034914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,4,1,128,1,float16,float16,0,0.33622400760650634
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,4,1,128,1,float16,float16,0,0.12467199563980103
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,4,2,128,1,float16,float16,0,0.20198400020599366
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,4,4,128,1,float16,float16,0,0.10567679405212402
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,4,1,128,1,float16,float16,0,0.06691840291023254
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,4,2,128,1,float16,float16,0,0.07674880027770996
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,4,4,128,1,float16,float16,0,0.059084802865982056
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,4,1,128,1,float16,float16,0,0.042752000689506534
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,4,2,128,1,float16,float16,0,0.04782080054283142
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,4,4,128,1,float16,float16,0,0.03589119911193848
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,4,1,128,1,float16,float16,0,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,4,2,128,1,float16,float16,0,0.03061760067939758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,4,4,128,1,float16,float16,0,0.02913280129432678
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,4,1,128,1,float16,float16,0,0.024883200228214265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,4,2,128,1,float16,float16,0,0.026368001103401185
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,4,4,128,1,float16,float16,0,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,4,1,128,1,float16,float16,0,0.024164800345897675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,4,2,128,1,float16,float16,0,0.02452480047941208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,4,4,128,1,float16,float16,0,0.024320000410079957
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,4,2,128,1,float16,float16,0,0.024268800020217897
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,4,1,128,1,float16,float16,0,0.02391040027141571
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,4,4,128,1,float16,float16,0,0.023244799673557283
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,4,1,128,1,float16,float16,0,0.023188799619674683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,4,2,128,1,float16,float16,0,0.023244799673557283
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,4,2,128,1,float16,float16,0,0.19793920516967772
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,4,1,128,1,float16,float16,0,0.10465279817581177
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,4,1,128,1,float16,float16,0,0.057651197910308837
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,4,4,128,1,float16,float16,0,0.09640960097312927
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,4,2,128,1,float16,float16,0,0.06814720034599304
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,4,1,128,1,float16,float16,0,0.03671039938926697
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,4,4,128,1,float16,float16,0,0.052223998308181765
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,4,2,128,1,float16,float16,0,0.04183039963245392
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,4,4,128,1,float16,float16,0,0.03261440098285675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,4,4,128,1,float16,float16,0,0.02534399926662445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,4,1,128,1,float16,float16,0,0.024673600494861603
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,4,2,128,1,float16,float16,0,0.02739199995994568
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,4,1,128,1,float16,float16,0,0.02109439969062805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,4,2,128,1,float16,float16,0,0.022726400196552275
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,4,4,128,1,float16,float16,0,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,4,1,128,1,float16,float16,0,0.022470399737358093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,4,2,128,1,float16,float16,0,0.020838400721549986
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,4,4,128,1,float16,float16,0,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,4,2,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,4,1,128,1,float16,float16,0,0.02011840045452118
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,4,4,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,4,1,128,1,float16,float16,0,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,4,4,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,4,2,128,1,float16,float16,0,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,4,2,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,4,1,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,4,1,128,1,float16,float16,0,0.05637120008468628
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,4,2,128,1,float16,float16,0,0.06727679967880248
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,4,4,128,1,float16,float16,0,0.0645632028579712
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,4,1,128,1,float16,float16,0,0.03624959886074066
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,4,2,128,1,float16,float16,0,0.04106239974498749
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,4,4,128,1,float16,float16,0,0.03824639916419983
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,4,1,128,1,float16,float16,0,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,4,2,128,1,float16,float16,0,0.027136000990867614
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,4,4,128,1,float16,float16,0,0.025753599405288697
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,4,1,128,1,float16,float16,0,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,4,2,128,1,float16,float16,0,0.022579200565814972
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,4,4,128,1,float16,float16,0,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,4,1,128,1,float16,float16,0,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,4,2,128,1,float16,float16,0,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,4,4,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,4,1,128,1,float16,float16,0,0.02011680006980896
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,4,2,128,1,float16,float16,0,0.020479999482631683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,4,4,128,1,float16,float16,0,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,4,1,128,1,float16,float16,0,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,4,2,128,1,float16,float16,0,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,4,4,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,4,2,128,1,float16,float16,0,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,4,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,4,4,128,1,float16,float16,0,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,4,1,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,4,2,128,1,float16,float16,0,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,4,2,128,1,float16,float16,0,0.054579198360443115
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,4,1,128,1,float16,float16,0,0.03635199964046478
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,4,4,128,1,float16,float16,0,0.05237759947776795
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,4,1,128,1,float16,float16,0,0.02529279887676239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,4,2,128,1,float16,float16,0,0.03363839983940124
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,4,4,128,1,float16,float16,0,0.032204800844192506
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,4,1,128,1,float16,float16,0,0.021401600539684297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,4,2,128,1,float16,float16,0,0.023296000063419343
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,4,4,128,1,float16,float16,0,0.022784000635147093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,4,1,128,1,float16,float16,0,0.020479999482631683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,4,2,128,1,float16,float16,0,0.020633600652217865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,4,4,128,1,float16,float16,0,0.020121599733829498
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,4,1,128,1,float16,float16,0,0.020019200444221497
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,4,2,128,1,float16,float16,0,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,4,4,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,4,1,128,1,float16,float16,0,0.0189423993229866
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,4,2,128,1,float16,float16,0,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,4,4,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,4,1,128,1,float16,float16,0,0.01867839992046356
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,4,2,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,4,1,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,4,4,128,1,float16,float16,0,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,4,1,128,1,float16,float16,0,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,4,2,128,1,float16,float16,0,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,4,4,128,1,float16,float16,0,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,4,1,128,1,float16,float16,0,0.032051199674606325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,4,2,128,1,float16,float16,0,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,4,2,128,1,float16,float16,0,0.04802559912204742
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,4,4,128,1,float16,float16,0,0.04664320051670075
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,4,1,128,1,float16,float16,0,0.022784000635147093
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,4,1,128,1,float16,float16,0,0.020633600652217865
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,4,2,128,1,float16,float16,0,0.030054399371147157
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,4,4,128,1,float16,float16,0,0.029440000653266907
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,4,2,128,1,float16,float16,0,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,4,4,128,1,float16,float16,0,0.02099200040102005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,4,1,128,1,float16,float16,0,0.019916799664497376
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,4,2,128,1,float16,float16,0,0.020168000459671022
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,4,4,128,1,float16,float16,0,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,4,1,128,1,float16,float16,0,0.01847680062055588
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,4,4,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,4,2,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,4,1,128,1,float16,float16,0,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,4,4,128,1,float16,float16,0,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,4,2,128,1,float16,float16,0,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,4,1,128,1,float16,float16,0,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,4,4,128,1,float16,float16,0,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,4,2,128,1,float16,float16,0,0.017505599558353423
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,4,1,128,1,float16,float16,0,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,4,1,128,1,float16,float16,0,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,4,4,128,1,float16,float16,0,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,4,2,128,1,float16,float16,0,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,4,2,128,1,float16,float16,0,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16384,2,1,128,1,float16,float16,0,4.61322250366211
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,2,2,128,1,float16,float16,0,1.4586352348327636
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,2,2,128,1,float16,float16,0,2.54965763092041
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,2,2,128,1,float16,float16,0,0.8495615959167481
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,2,1,128,1,float16,float16,0,0.7943168163299561
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,2,1,128,1,float16,float16,0,2.554470443725586
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,2,1,128,1,float16,float16,0,1.4029312133789062
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,12288,2,1,128,1,float16,float16,0,2.8392959594726563
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,2,2,128,1,float16,float16,0,1.5370240211486816
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,2,2,128,1,float16,float16,0,0.8340479850769043
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,2,1,128,1,float16,float16,0,1.4957568168640136
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,2,2,128,1,float16,float16,0,0.5186048030853272
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,2,1,128,1,float16,float16,0,0.521830415725708
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,2,1,128,1,float16,float16,0,0.8461824417114258
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,10240,2,1,128,1,float16,float16,0,2.03540153503418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,2,2,128,1,float16,float16,0,1.2179967880249023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,2,1,128,1,float16,float16,0,1.0053119659423828
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,2,2,128,1,float16,float16,0,0.6244863986968994
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,2,1,128,1,float16,float16,0,0.38932480812072756
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,2,1,128,1,float16,float16,0,0.6187520027160645
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,2,2,128,1,float16,float16,0,0.3876352071762085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,8192,2,1,128,1,float16,float16,0,2.478540802001953
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,2,2,128,1,float16,float16,0,1.4144512176513673
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,2,1,128,1,float16,float16,0,1.3938639640808106
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,2,2,128,1,float16,float16,0,0.7409152030944824
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,2,1,128,1,float16,float16,0,0.723199987411499
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,2,2,128,1,float16,float16,0,0.44661760330200195
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,2,2,128,1,float16,float16,0,0.2800640106201172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,2,1,128,1,float16,float16,0,0.43565921783447265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,2,1,128,1,float16,float16,0,0.2799616098403931
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,6144,2,1,128,1,float16,float16,0,1.5712767601013184
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,2,2,128,1,float16,float16,0,0.9259008407592774
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,2,2,128,1,float16,float16,0,0.4734464168548584
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,2,1,128,1,float16,float16,0,0.8210432052612304
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,2,1,128,1,float16,float16,0,0.4472320079803467
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,2,2,128,1,float16,float16,0,0.26705920696258545
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,2,1,128,1,float16,float16,0,0.26362879276275636
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,2,2,128,1,float16,float16,0,0.16389119625091553
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,2,1,128,1,float16,float16,0,0.1570304036140442
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,2,2,128,1,float16,float16,0,0.9175552368164063
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,4096,2,1,128,1,float16,float16,0,1.5262207984924316
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,2,2,128,1,float16,float16,0,0.4708303928375244
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,2,1,128,1,float16,float16,0,0.7981056213378906
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,2,1,128,1,float16,float16,0,0.4088319778442383
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,2,2,128,1,float16,float16,0,0.26086399555206297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,2,1,128,1,float16,float16,0,0.25185279846191405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,2,2,128,1,float16,float16,0,0.17817599773406984
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,2,1,128,1,float16,float16,0,0.1745919942855835
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,2,2,128,1,float16,float16,0,0.1075711965560913
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,2,1,128,1,float16,float16,0,0.10542080402374268
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,3072,2,1,128,1,float16,float16,0,0.95283203125
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,2,2,128,1,float16,float16,0,0.5989376068115234
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,2,1,128,1,float16,float16,0,0.49008641242980955
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,2,2,128,1,float16,float16,0,0.29409279823303225
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,2,1,128,1,float16,float16,0,0.264192008972168
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,2,2,128,1,float16,float16,0,0.16711679697036744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,2,1,128,1,float16,float16,0,0.16440320014953613
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,2,2,128,1,float16,float16,0,0.11550719738006592
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,2,1,128,1,float16,float16,0,0.10562560558319092
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,2,2,128,1,float16,float16,0,0.07045120000839233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,2,1,128,1,float16,float16,0,0.06804479956626892
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,2048,2,1,128,1,float16,float16,0,0.9622528076171875
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,2,2,128,1,float16,float16,0,0.31129601001739504
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,2,1,128,1,float16,float16,0,0.5031936168670654
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,2,2,128,1,float16,float16,0,0.6195712089538574
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,2,1,128,1,float16,float16,0,0.25241119861602784
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,2,1,128,1,float16,float16,0,0.16117759943008422
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,2,2,128,1,float16,float16,0,0.10792959928512573
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,2,2,128,1,float16,float16,0,0.1710592031478882
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,2,1,128,1,float16,float16,0,0.10424319505691529
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,2,1,128,1,float16,float16,0,0.06783999800682068
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,2,2,128,1,float16,float16,0,0.048281601071357726
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,2,2,128,1,float16,float16,0,0.0711679995059967
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,2,1,128,1,float16,float16,0,0.04684799909591675
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,2,2,128,1,float16,float16,0,0.41994237899780273
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1536,2,1,128,1,float16,float16,0,0.6169040203094482
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,2,1,128,1,float16,float16,0,0.31201279163360596
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,2,2,128,1,float16,float16,0,0.18867199420928954
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,2,1,128,1,float16,float16,0,0.16936960220336914
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,2,2,128,1,float16,float16,0,0.11842559576034546
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,2,1,128,1,float16,float16,0,0.11054079532623291
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,2,2,128,1,float16,float16,0,0.08053759932518005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,2,1,128,1,float16,float16,0,0.0741375982761383
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,2,2,128,1,float16,float16,0,0.050995200872421265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,2,1,128,1,float16,float16,0,0.0492031991481781
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,2,2,128,1,float16,float16,0,0.039526399970054624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,2,1,128,1,float16,float16,0,0.03834879994392395
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,1024,2,1,128,1,float16,float16,0,0.6768640041351318
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,2,1,128,1,float16,float16,0,0.35055840015411377
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,2,2,128,1,float16,float16,0,0.4680191993713379
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,2,2,128,1,float16,float16,0,0.22655999660491943
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,2,1,128,1,float16,float16,0,0.16558079719543456
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,2,1,128,1,float16,float16,0,0.10798079967498779
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,2,2,128,1,float16,float16,0,0.11786240339279175
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,2,2,128,1,float16,float16,0,0.07219200134277344
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,2,1,128,1,float16,float16,0,0.06666240096092224
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,2,1,128,1,float16,float16,0,0.04975839853286743
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,2,2,128,1,float16,float16,0,0.05258240103721619
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,2,2,128,1,float16,float16,0,0.039934399724006656
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,2,1,128,1,float16,float16,0,0.03865599930286408
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,2,2,128,1,float16,float16,0,0.03358719944953918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,2,1,128,1,float16,float16,0,0.03266560137271881
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,512,2,1,128,1,float16,float16,0,0.5560832023620605
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,2,2,128,1,float16,float16,0,0.41881442070007324
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,2,1,128,1,float16,float16,0,0.2726383924484253
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,2,2,128,1,float16,float16,0,0.18421759605407714
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,2,1,128,1,float16,float16,0,0.11791360378265381
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,2,2,128,1,float16,float16,0,0.08954880237579346
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,2,1,128,1,float16,float16,0,0.07464960217475891
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,2,2,128,1,float16,float16,0,0.05294079780578613
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,2,1,128,1,float16,float16,0,0.04782080054283142
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,2,2,128,1,float16,float16,0,0.04510720074176788
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,2,1,128,1,float16,float16,0,0.04177919924259186
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,2,2,128,1,float16,float16,0,0.03547680079936981
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,2,1,128,1,float16,float16,0,0.034201601147651674
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,2,1,128,1,float16,float16,0,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,2,2,128,1,float16,float16,0,0.029183998703956604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,2,2,128,1,float16,float16,0,0.02677760124206543
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,2,1,128,1,float16,float16,0,0.02687999904155731
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,256,2,1,128,1,float16,float16,0,0.2006527900695801
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,2,2,128,1,float16,float16,0,0.11048959493637085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,2,1,128,1,float16,float16,0,0.07761920094490052
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,2,2,128,1,float16,float16,0,0.05877760052680969
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,2,1,128,1,float16,float16,0,0.04792320132255554
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,2,2,128,1,float16,float16,0,0.03614720106124878
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,2,1,128,1,float16,float16,0,0.030976000428199767
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,2,2,128,1,float16,float16,0,0.02913280129432678
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,2,1,128,1,float16,float16,0,0.026675200462341307
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,2,2,128,1,float16,float16,0,0.026465600728988646
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,2,1,128,1,float16,float16,0,0.024831999838352204
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,2,2,128,1,float16,float16,0,0.02452480047941208
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,2,1,128,1,float16,float16,0,0.024166400730609893
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,2,2,128,1,float16,float16,0,0.024063999950885772
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,2,2,128,1,float16,float16,0,0.023756800591945647
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,2,1,128,1,float16,float16,0,0.024320000410079957
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,2,1,128,1,float16,float16,0,0.023343999683856965
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,128,2,1,128,1,float16,float16,0,0.06988800168037415
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,2,2,128,1,float16,float16,0,0.05278720259666443
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,2,2,128,1,float16,float16,0,0.03287039995193482
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,2,1,128,1,float16,float16,0,0.042444801330566405
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,2,1,128,1,float16,float16,0,0.027596798539161683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,2,1,128,1,float16,float16,0,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,2,2,128,1,float16,float16,0,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,2,2,128,1,float16,float16,0,0.02247679978609085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,2,1,128,1,float16,float16,0,0.020838400721549986
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,2,2,128,1,float16,float16,0,0.020582400262355804
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,2,1,128,1,float16,float16,0,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,2,2,128,1,float16,float16,0,0.020479999482631683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,2,1,128,1,float16,float16,0,0.020428800582885744
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,2,2,128,1,float16,float16,0,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,2,1,128,1,float16,float16,0,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,2,2,128,1,float16,float16,0,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,2,1,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,64,2,1,128,1,float16,float16,0,0.04223999977111816
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,2,2,128,1,float16,float16,0,0.03916800022125244
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,2,1,128,1,float16,float16,0,0.028415998816490172
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,2,2,128,1,float16,float16,0,0.02600640058517456
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,2,1,128,1,float16,float16,0,0.022886399924755097
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,2,2,128,1,float16,float16,0,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,2,1,128,1,float16,float16,0,0.02114560008049011
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,2,2,128,1,float16,float16,0,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,2,2,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,2,1,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,2,1,128,1,float16,float16,0,0.019968000054359437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,2,2,128,1,float16,float16,0,0.01863519996404648
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,2,1,128,1,float16,float16,0,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,2,1,128,1,float16,float16,0,0.01867839992046356
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,2,2,128,1,float16,float16,0,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,2,2,128,1,float16,float16,0,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,2,1,128,1,float16,float16,0,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,32,2,1,128,1,float16,float16,0,0.034764799475669864
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,2,2,128,1,float16,float16,0,0.03296479880809784
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,2,1,128,1,float16,float16,0,0.02380799949169159
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,2,2,128,1,float16,float16,0,0.022835199534893037
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,2,1,128,1,float16,float16,0,0.021137599647045136
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,2,2,128,1,float16,float16,0,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,2,1,128,1,float16,float16,0,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,2,2,128,1,float16,float16,0,0.019555200636386872
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,2,1,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,2,2,128,1,float16,float16,0,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,2,1,128,1,float16,float16,0,0.018780800700187682
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,2,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,2,2,128,1,float16,float16,0,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,2,1,128,1,float16,float16,0,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,2,2,128,1,float16,float16,0,0.017455999553203583
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,2,2,128,1,float16,float16,0,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,2,1,128,1,float16,float16,0,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,256,16,2,1,128,1,float16,float16,0,0.03118079900741577
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,2,2,128,1,float16,float16,0,0.021503999829292297
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,2,2,128,1,float16,float16,0,0.029951998591423036
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,2,1,128,1,float16,float16,0,0.020377600193023683
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,2,1,128,1,float16,float16,0,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,2,2,128,1,float16,float16,0,0.019865599274635316
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,2,1,128,1,float16,float16,0,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,2,2,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,2,2,128,1,float16,float16,0,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,2,1,128,1,float16,float16,0,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,2,1,128,1,float16,float16,0,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,2,2,128,1,float16,float16,0,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,2,1,128,1,float16,float16,0,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,2,2,128,1,float16,float16,0,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,2,1,128,1,float16,float16,0,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,2,2,128,1,float16,float16,0,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,2,1,128,1,float16,float16,0,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16384,1,1,128,1,float16,float16,0,1.4111231803894042
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16384,1,1,128,1,float16,float16,0,0.5263872146606445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16384,1,1,128,1,float16,float16,0,0.8537088394165039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,12288,1,1,128,1,float16,float16,0,0.8934399604797363
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,12288,1,1,128,1,float16,float16,0,0.49566078186035156
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,12288,1,1,128,1,float16,float16,0,0.2639359951019287
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,10240,1,1,128,1,float16,float16,0,0.6657536029815674
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,10240,1,1,128,1,float16,float16,0,0.39976959228515624
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,10240,1,1,128,1,float16,float16,0,0.22287359237670898
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,8192,1,1,128,1,float16,float16,0,0.8099840164184571
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,8192,1,1,128,1,float16,float16,0,0.47277121543884276
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,8192,1,1,128,1,float16,float16,0,0.3184639930725098
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,8192,1,1,128,1,float16,float16,0,0.1778175950050354
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,6144,1,1,128,1,float16,float16,0,0.5038591861724854
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,6144,1,1,128,1,float16,float16,0,0.1655295968055725
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,6144,1,1,128,1,float16,float16,0,0.27648000717163085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,6144,1,1,128,1,float16,float16,0,0.10577919483184814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,4096,1,1,128,1,float16,float16,0,0.4848639965057373
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,4096,1,1,128,1,float16,float16,0,0.2767359972000122
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,4096,1,1,128,1,float16,float16,0,0.17786879539489747
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,4096,1,1,128,1,float16,float16,0,0.10721280574798583
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,4096,1,1,128,1,float16,float16,0,0.0667136013507843
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,3072,1,1,128,1,float16,float16,0,0.3023360013961792
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,3072,1,1,128,1,float16,float16,0,0.18114240169525148
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,3072,1,1,128,1,float16,float16,0,0.11284480094909669
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,3072,1,1,128,1,float16,float16,0,0.07045120000839233
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,3072,1,1,128,1,float16,float16,0,0.05068640112876892
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,2048,1,1,128,1,float16,float16,0,0.31549439430236814
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,2048,1,1,128,1,float16,float16,0,0.1745919942855835
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,2048,1,1,128,1,float16,float16,0,0.10818560123443603
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,2048,1,1,128,1,float16,float16,0,0.07116479873657226
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,2048,1,1,128,1,float16,float16,0,0.047513601183891294
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,2048,1,1,128,1,float16,float16,0,0.037836799025535585
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1536,1,1,128,1,float16,float16,0,0.1969151973724365
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1536,1,1,128,1,float16,float16,0,0.12072960138320923
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1536,1,1,128,1,float16,float16,0,0.08064000010490417
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1536,1,1,128,1,float16,float16,0,0.05109120011329651
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1536,1,1,128,1,float16,float16,0,0.03983359932899475
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1536,1,1,128,1,float16,float16,0,0.03358719944953918
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,1024,1,1,128,1,float16,float16,0,0.2307584047317505
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,1024,1,1,128,1,float16,float16,0,0.07234560251235962
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,1024,1,1,128,1,float16,float16,0,0.11852799654006958
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,1024,1,1,128,1,float16,float16,0,0.05263360142707825
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,1024,1,1,128,1,float16,float16,0,0.03988479971885681
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,1024,1,1,128,1,float16,float16,0,0.03353599905967712
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,1024,1,1,128,1,float16,float16,0,0.03184640109539032
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,512,1,1,128,1,float16,float16,0,0.1848304033279419
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,512,1,1,128,1,float16,float16,0,0.045363199710845944
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,512,1,1,128,1,float16,float16,0,0.09052159786224365
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,512,1,1,128,1,float16,float16,0,0.053350400924682614
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,512,1,1,128,1,float16,float16,0,0.03548159897327423
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,512,1,1,128,1,float16,float16,0,0.02913280129432678
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,512,1,1,128,1,float16,float16,0,0.027033600211143493
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,512,1,1,128,1,float16,float16,0,0.02662079930305481
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,256,1,1,128,1,float16,float16,0,0.060057598352432254
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,256,1,1,128,1,float16,float16,0,0.02959359884262085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,256,1,1,128,1,float16,float16,0,0.03671039938926697
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,256,1,1,128,1,float16,float16,0,0.026163199543952943
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,256,1,1,128,1,float16,float16,0,0.024934400618076325
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,256,1,1,128,1,float16,float16,0,0.024217599630355836
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,256,1,1,128,1,float16,float16,0,0.02467840015888214
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,256,1,1,128,1,float16,float16,0,0.022835199534893037
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,128,1,1,128,1,float16,float16,0,0.033843201398849485
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,128,1,1,128,1,float16,float16,0,0.026060798764228822
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,128,1,1,128,1,float16,float16,0,0.02247679978609085
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,128,1,1,128,1,float16,float16,0,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,128,1,1,128,1,float16,float16,0,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,128,1,1,128,1,float16,float16,0,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,128,1,1,128,1,float16,float16,0,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,128,1,1,128,1,float16,float16,0,0.018430399894714355
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,64,1,1,128,1,float16,float16,0,0.027136000990867614
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,64,1,1,128,1,float16,float16,0,0.02252800017595291
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,64,1,1,128,1,float16,float16,0,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,64,1,1,128,1,float16,float16,0,0.01966080069541931
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,64,1,1,128,1,float16,float16,0,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,64,1,1,128,1,float16,float16,0,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,64,1,1,128,1,float16,float16,0,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,64,1,1,128,1,float16,float16,0,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,32,1,1,128,1,float16,float16,0,0.02391040027141571
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,32,1,1,128,1,float16,float16,0,0.020787200331687926
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,32,1,1,128,1,float16,float16,0,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,32,1,1,128,1,float16,float16,0,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,32,1,1,128,1,float16,float16,0,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,32,1,1,128,1,float16,float16,0,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,32,1,1,128,1,float16,float16,0,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,128,16,1,1,128,1,float16,float16,0,0.022835199534893037
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,32,1,1,128,1,float16,float16,0,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,64,16,1,1,128,1,float16,float16,0,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,32,16,1,1,128,1,float16,float16,0,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,16,16,1,1,128,1,float16,float16,0,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,8,16,1,1,128,1,float16,float16,0,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,4,16,1,1,128,1,float16,float16,0,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,2,16,1,1,128,1,float16,float16,0,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,context_attention,flash_attention,1,16,1,1,128,1,float16,float16,0,0.017868800461292265
