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,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,1,0.202239990234375
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,3,0.20346879959106445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,31,0.20505599975585936
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,127,0.3899904012680054
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,7,0.20188159942626954
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,255,0.7585279941558838
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,15,0.20444159507751464
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,511,1.5073280334472656
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,63,0.21637120246887206
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,1023,3.045376014709473
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,3,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,64,128,1,float16,float16,2047,6.476493072509766
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,1,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,7,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,15,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,31,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,63,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,127,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,255,0.022784000635147093
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,511,0.03297280073165894
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,2047,0.06446080207824707
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,64,128,1,float16,float16,1023,0.04141919910907745
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,7,0.019097599387168884
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,3,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,15,0.019097599387168884
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,1,0.019145600497722626
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,31,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,63,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,127,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,255,0.023705600202083586
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,1,0.11146240234375
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,1023,0.061849600076675414
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,511,0.03973119854927063
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,64,128,1,float16,float16,2047,0.22451200485229492
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,7,0.11151360273361206
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,15,0.11187200546264649
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,31,0.11151360273361206
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,63,0.1144320011138916
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,127,0.20126719474792482
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,255,0.39388160705566405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,511,0.7782815933227539
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,3,0.11136000156402588
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,1023,1.5319040298461915
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,1,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,64,128,1,float16,float16,2047,3.134207916259766
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,3,0.025088000297546386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,7,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,15,0.025241601467132568
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,31,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,127,0.026419198513031004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,255,0.035174399614334106
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,511,0.06507520079612732
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,1023,0.24396800994873047
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,2047,0.42183680534362794
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,1,0.37938239574432375
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,3,0.3788800001144409
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,7,0.386297607421875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,15,0.384768009185791
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,31,0.3882496118545532
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,64,128,1,float16,float16,63,0.025856000185012818
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,63,0.4156928062438965
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,127,0.7648240089416504
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,255,1.530675220489502
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,511,3.0969776153564452
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,64,128,1,float16,float16,1,0.727291202545166
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,64,128,1,float16,float16,15,0.7321599960327149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,64,128,1,float16,float16,3,0.7312384128570557
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,64,128,1,float16,float16,31,0.7474175930023194
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,64,128,1,float16,float16,63,0.8479743957519531
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,64,128,1,float16,float16,1023,6.305843353271484
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,64,128,1,float16,float16,127,1.6651775360107421
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,64,128,1,float16,float16,1,1.4081024169921874
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,64,128,1,float16,float16,255,2.9928447723388674
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,64,128,1,float16,float16,3,1.4108672142028809
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,64,128,1,float16,float16,7,1.4118399620056152
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,64,128,1,float16,float16,15,1.4329343795776368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,64,128,1,float16,float16,7,0.729852819442749
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,64,128,1,float16,float16,31,1.48602237701416
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,64,128,1,float16,float16,63,1.643110466003418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,1,0.03804160058498383
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,3,0.037939199805259706
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,7,0.03824479877948761
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,15,0.03855200111865997
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,64,128,1,float16,float16,127,3.007948875427246
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,63,0.0390608012676239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,127,0.03993600010871887
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,255,0.0582144021987915
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,511,0.22574079036712646
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,1023,0.4147200107574463
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,2047,0.8093695640563965
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,64,128,1,float16,float16,31,0.03850080072879791
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,64,128,1,float16,float16,1,2.8145151138305664
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,64,128,1,float16,float16,3,2.8210176467895507
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,64,128,1,float16,float16,7,2.834432029724121
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,64,128,1,float16,float16,15,2.834379196166992
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,64,128,1,float16,float16,31,2.8916223526000975
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,64,128,1,float16,float16,63,3.2866817474365235
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,64,128,1,float16,float16,1,5.642291259765625
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,1,0.06348800063133239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,3,0.06353920102119445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,64,128,1,float16,float16,15,5.661849594116211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,64,128,1,float16,float16,7,5.655238342285156
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,64,128,1,float16,float16,31,5.8101249694824215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,7,0.06415359973907471
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,64,128,1,float16,float16,3,5.645619201660156
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,15,0.0649728000164032
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,31,0.06435840129852295
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,63,0.06543359756469727
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,255,0.22256639003753662
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,511,0.41021437644958497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,127,0.06558719873428345
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,1023,0.8729599952697754
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,1,0.15800319910049437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,3,0.1546720027923584
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,7,0.15846400260925292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,64,128,1,float16,float16,2047,1.6332799911499023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,31,0.15800319910049437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,15,0.16030720472335816
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,127,0.3278336048126221
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,63,0.1633792042732239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,255,0.5745151996612549
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,511,1.1590144157409668
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,1,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,1023,2.364364814758301
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,3,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,31,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,7,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,48,48,128,1,float16,float16,2047,4.9240577697753904
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,63,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,255,0.022067199647426605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,511,0.03404799997806549
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,1023,0.046028798818588255
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,48,48,128,1,float16,float16,2047,0.07162879705429077
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,1,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,7,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,15,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,127,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,31,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,3,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,255,0.022835199534893037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,511,0.043977600336074826
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,1023,0.06333439946174621
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,1,0.08535040020942689
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,3,0.08422399759292602
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,48,48,128,1,float16,float16,2047,0.14289920330047606
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,7,0.085452800989151
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,31,0.08616960048675537
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,15,0.08535040020942689
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,63,0.08780800104141236
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,127,0.09533439874649048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,255,0.2968064069747925
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,511,0.5901311874389649
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,1023,1.1594240188598632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,1,0.024934400618076325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,48,48,128,1,float16,float16,2047,2.3708160400390623
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,7,0.024934400618076325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,3,0.0247296005487442
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,15,0.02447360008955002
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,31,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,127,0.02590720057487488
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,63,0.025088000297546386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,255,0.034457600116729735
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,511,0.06261759996414185
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,1023,0.14551039934158325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,1,0.29178879261016843
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,7,0.29670400619506837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,3,0.29614078998565674
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,48,48,128,1,float16,float16,2047,0.3206144094467163
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,15,0.2960383892059326
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,31,0.30259199142456056
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,63,0.32071518898010254
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,127,0.5792255878448487
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,255,1.1303936004638673
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,48,48,128,1,float16,float16,1,0.5509071826934815
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,511,2.2466047286987303
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,48,48,128,1,float16,float16,3,0.5581823825836182
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,48,48,128,1,float16,float16,7,0.5603839874267578
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,48,48,128,1,float16,float16,15,0.5576191902160644
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,48,48,128,1,float16,float16,1023,4.584908676147461
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,48,48,128,1,float16,float16,63,0.6436863899230957
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,48,48,128,1,float16,float16,31,0.5673984050750732
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,48,48,128,1,float16,float16,127,1.1427791595458985
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,48,48,128,1,float16,float16,1,1.0696687698364258
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,48,48,128,1,float16,float16,255,2.243891143798828
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,48,48,128,1,float16,float16,7,1.0699775695800782
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,48,48,128,1,float16,float16,3,1.0769408226013184
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,48,48,128,1,float16,float16,63,1.234483242034912
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,48,48,128,1,float16,float16,31,1.110630416870117
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,48,48,128,1,float16,float16,15,1.0782719612121583
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,1,0.03184640109539032
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,3,0.031641599535942075
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,15,0.031948798894882204
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,48,48,128,1,float16,float16,127,2.3681007385253907
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,7,0.031590399146080014
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,31,0.032051199674606325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,255,0.04745280146598816
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,63,0.03261440098285675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,511,0.14428160190582276
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,127,0.033432000875473024
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,1023,0.3456000089645386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,48,48,128,1,float16,float16,2047,0.6112768173217773
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,48,48,128,1,float16,float16,1,2.0988927841186524
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,48,48,128,1,float16,float16,3,2.087936019897461
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,48,48,128,1,float16,float16,7,2.1253551483154296
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,48,48,128,1,float16,float16,31,2.1657087326049806
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,48,48,128,1,float16,float16,15,2.147020721435547
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,48,48,128,1,float16,float16,63,2.4567808151245116
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,48,48,128,1,float16,float16,1,4.236288070678711
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,48,48,128,1,float16,float16,3,4.269926452636719
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,48,48,128,1,float16,float16,7,4.258355331420899
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,1,0.05073919892311096
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,48,48,128,1,float16,float16,15,4.244887924194336
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,7,0.050995200872421265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,3,0.05181440114974976
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,48,48,128,1,float16,float16,31,4.39997444152832
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,15,0.05125120282173157
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,63,0.05288959741592407
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,31,0.051660799980163576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,127,0.05242879986763001
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,255,0.10101759433746338
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,511,0.3146752119064331
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,1,0.1371135950088501
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,1023,0.5987328052520752
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,48,48,128,1,float16,float16,2047,1.185433578491211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,3,0.13849600553512573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,15,0.13767679929733276
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,31,0.1371135950088501
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,63,0.13992960453033448
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,7,0.13751840591430664
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,127,0.2506239891052246
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,255,0.4824575901031494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,511,0.9607168197631836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,1023,1.9712511062622071
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,1,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,3,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,40,40,128,1,float16,float16,2047,4.0624126434326175
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,63,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,31,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,127,0.017862400412559508
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,255,0.022169600427150726
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,511,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,1023,0.036556801199913024
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,1,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,40,40,128,1,float16,float16,2047,0.04802559912204742
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,7,0.018016000092029572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,15,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,127,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,63,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,255,0.022732800245285033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,511,0.04423680007457733
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,2047,0.11607040166854858
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,1,0.07746559977531434
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,40,40,128,1,float16,float16,1023,0.06353920102119445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,7,0.0772607982158661
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,15,0.07802879810333252
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,3,0.07833600044250488
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,31,0.0775168001651764
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,63,0.07895039916038513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,127,0.07930880188941955
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,511,0.5000703811645508
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,1023,0.973516845703125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,255,0.2507200002670288
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,1,0.02431679964065552
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,40,40,128,1,float16,float16,2047,1.9719680786132812
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,3,0.024371199309825897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,7,0.02462719976902008
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,31,0.0247296005487442
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,15,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,63,0.025190401077270507
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,255,0.03374080061912536
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,511,0.06231039762496948
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,1023,0.11996159553527833
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,127,0.02529279887676239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,40,40,128,1,float16,float16,2047,0.3181056022644043
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,3,0.2490367889404297
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,7,0.250163197517395
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,1,0.2510848045349121
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,15,0.25712640285491944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,31,0.25287680625915526
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,63,0.2731519937515259
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,127,0.4848624229431152
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,255,0.9466879844665528
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,511,1.930905532836914
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,40,40,128,1,float16,float16,1,0.46940159797668457
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,40,40,128,1,float16,float16,3,0.47662081718444826
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,40,40,128,1,float16,float16,7,0.46940159797668457
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,40,40,128,1,float16,float16,15,0.4729856014251709
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,40,40,128,1,float16,float16,1023,3.8056961059570313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,40,40,128,1,float16,float16,63,0.5244927883148194
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,40,40,128,1,float16,float16,31,0.4759039878845215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,40,40,128,1,float16,float16,127,1.0364416122436524
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,40,40,128,1,float16,float16,255,1.9364351272583007
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,40,40,128,1,float16,float16,3,0.8983039855957031
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,40,40,128,1,float16,float16,1,0.911411190032959
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,40,40,128,1,float16,float16,7,0.9032192230224609
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,40,40,128,1,float16,float16,15,0.9109503746032714
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,40,40,128,1,float16,float16,31,0.931123161315918
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,40,40,128,1,float16,float16,127,1.8978303909301757
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,40,40,128,1,float16,float16,63,1.0278400421142577
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,1,0.030668801069259642
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,3,0.03168480098247528
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,7,0.031534400582313535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,15,0.03128319978713989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,31,0.0317903995513916
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,63,0.032051199674606325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,255,0.045465600490570066
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,511,0.09958400130271912
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,2047,0.5318655967712402
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,1023,0.2790911912918091
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,40,40,128,1,float16,float16,127,0.03261440098285675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,40,40,128,1,float16,float16,1,1.7462783813476563
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,40,40,128,1,float16,float16,7,1.765376091003418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,40,40,128,1,float16,float16,3,1.7556991577148438
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,40,40,128,1,float16,float16,15,1.7868799209594726
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,40,40,128,1,float16,float16,31,1.8304000854492188
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,40,40,128,1,float16,float16,63,2.0537343978881837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,40,40,128,1,float16,float16,1,3.5656192779541014
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,40,40,128,1,float16,float16,7,3.547443389892578
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,40,40,128,1,float16,float16,3,3.5665409088134767
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,40,40,128,1,float16,float16,31,3.634124755859375
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,40,40,128,1,float16,float16,15,3.5516929626464844
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,1,0.04510720074176788
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,3,0.04469760060310364
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,7,0.04510720074176788
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,15,0.045363199710845944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,31,0.0447488009929657
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,63,0.045772799849510194
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,511,0.2673664093017578
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,127,0.04664320051670075
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,255,0.07178239822387696
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,1023,0.5056000232696534
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,40,40,128,1,float16,float16,2047,1.002188777923584
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,1,0.1135103940963745
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,3,0.11064319610595703
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,15,0.11258879899978638
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,31,0.11386719942092896
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,63,0.1149440050125122
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,127,0.20131840705871581
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,7,0.11284480094909669
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,255,0.41850881576538085
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,511,0.7745535850524903
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,1023,1.5333375930786133
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,1,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,15,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,2047,3.136819267272949
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,3,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,31,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,32,128,1,float16,float16,4095,6.467378997802735
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,511,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,2047,0.04116480052471161
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,1023,0.032971200346946714
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,32,128,1,float16,float16,4095,0.062463998794555664
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,1,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,15,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,31,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,127,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,511,0.03281759917736053
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,255,0.02252800017595291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,1023,0.04085760116577149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,2047,0.06272000074386597
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,32,128,1,float16,float16,4095,0.2274303913116455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,1,0.06415359973907471
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,3,0.06486880183219909
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,7,0.06512640118598938
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,15,0.06435840129852295
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,63,0.06696959733963012
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,127,0.06507520079612732
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,255,0.20305919647216797
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,31,0.06502400040626526
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,511,0.4104191780090332
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,2047,1.6041984558105469
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,1023,0.7962624073028565
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,3,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,1,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,7,0.018838399648666383
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,32,128,1,float16,float16,4095,3.262771224975586
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,31,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,15,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,127,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,255,0.02396160066127777
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,63,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,511,0.03988479971885681
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,1023,0.061952000856399535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,2047,0.2264575958251953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,32,128,1,float16,float16,4095,0.4488704204559326
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,1,0.20229120254516603
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,7,0.2027008056640625
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,3,0.20244479179382324
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,31,0.20464639663696288
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,15,0.20592639446258545
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,63,0.2170880079269409
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,127,0.3895807981491089
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,255,0.760319995880127
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,511,1.5415295600891112
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,32,128,1,float16,float16,1,0.38021121025085447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,32,128,1,float16,float16,3,0.3826175928115845
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,1023,3.045734405517578
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,32,128,1,float16,float16,15,0.3853312015533447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,32,128,1,float16,float16,7,0.38506720066070554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,32,128,1,float16,float16,31,0.3972543954849243
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,32,128,1,float16,float16,2047,6.247372817993164
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,32,128,1,float16,float16,127,0.7660031795501709
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,32,128,1,float16,float16,63,0.4413951873779297
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,32,128,1,float16,float16,1,0.7304704189300537
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,32,128,1,float16,float16,255,1.536716842651367
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,32,128,1,float16,float16,7,0.7416319847106934
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,32,128,1,float16,float16,3,0.7321087837219238
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,32,128,1,float16,float16,511,2.983884811401367
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,32,128,1,float16,float16,15,0.7399424076080322
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,32,128,1,float16,float16,31,0.7533055782318115
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,3,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,32,128,1,float16,float16,127,1.5163392066955566
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,7,0.025190401077270507
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,1,0.025190401077270507
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,32,128,1,float16,float16,63,0.8300543785095215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,15,0.025036799907684325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,32,128,1,float16,float16,255,3.00134391784668
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,31,0.026214399933815004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,63,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,127,0.026521599292755126
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,255,0.03542400002479553
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,1023,0.22409920692443847
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,2047,0.4583424091339111
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,4095,0.7910912036895752
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,32,128,1,float16,float16,511,0.06455680131912231
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,32,128,1,float16,float16,1,1.4131711959838866
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,32,128,1,float16,float16,3,1.408563232421875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,32,128,1,float16,float16,7,1.4147583961486816
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,32,128,1,float16,float16,15,1.4593024253845215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,32,128,1,float16,float16,31,1.4654975891113282
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,32,128,1,float16,float16,63,1.660211181640625
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,32,128,1,float16,float16,127,3.011891174316406
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,32,128,1,float16,float16,1,2.833152008056641
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,32,128,1,float16,float16,3,2.8293119430541993
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,32,128,1,float16,float16,7,2.846259117126465
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,32,128,1,float16,float16,15,2.8580863952636717
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,32,128,1,float16,float16,31,2.918707275390625
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,1,0.03809280097484589
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,32,128,1,float16,float16,63,3.268864059448242
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,3,0.038553598523139956
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,7,0.03865599930286408
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,31,0.03850240111351013
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,15,0.03834559917449951
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,63,0.039526399970054624
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,255,0.05882880091667175
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,127,0.040089601278305055
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,511,0.23802878856658935
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,1023,0.4135424137115479
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,1,0.08657919764518737
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,3,0.08509439826011658
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,4095,1.5737855911254883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,32,128,1,float16,float16,2047,0.8122879981994628
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,7,0.08678399920463561
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,15,0.08514559864997864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,31,0.0877568006515503
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,63,0.08908799886703492
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,127,0.09876480102539062
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,255,0.2969088077545166
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,511,0.6428160190582275
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,2047,2.3609344482421877
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,1023,1.202995204925537
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,1,0.017254400253295898
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,3,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,24,24,128,1,float16,float16,4095,4.589260864257812
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,31,0.017508800327777862
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,255,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,1023,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,511,0.02447360008955002
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,127,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,2047,0.037836799025535585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,24,24,128,1,float16,float16,4095,0.05232639908790589
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,3,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,7,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,31,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,15,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,255,0.02222079932689667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,63,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,127,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,511,0.03548159897327423
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,1023,0.046540799736976626
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,2047,0.07178239822387696
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,24,24,128,1,float16,float16,4095,0.18196480274200438
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,1,0.051913601160049436
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,3,0.051759999990463254
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,7,0.05232639908790589
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,31,0.05196800231933594
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,15,0.052275198698043826
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,63,0.052019202709198
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,255,0.09067519903182983
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,127,0.053401601314544675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,511,0.314465594291687
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,1023,0.5989888191223145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,2047,1.1977727890014649
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,3,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,24,24,128,1,float16,float16,4095,2.3634944915771485
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,7,0.018219199776649476
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,1,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,63,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,31,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,127,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,255,0.024217599630355836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,2047,0.1420799970626831
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,511,0.04403199851512909
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,4095,0.3216383934020996
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,24,24,128,1,float16,float16,1023,0.06384639739990235
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,1,0.1566208004951477
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,3,0.15820800065994262
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,7,0.15969280004501343
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,15,0.15856640338897704
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,31,0.16116960048675538
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,63,0.1605631947517395
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,127,0.29598560333251955
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,511,1.1655167579650878
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,255,0.5761536121368408
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,24,24,128,1,float16,float16,1,0.2951632022857666
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,1023,2.3693824768066407
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,24,24,128,1,float16,float16,3,0.29460480213165285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,24,24,128,1,float16,float16,7,0.29511680603027346
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,24,24,128,1,float16,float16,15,0.3018239974975586
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,24,24,128,1,float16,float16,2047,4.8714241027832035
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,24,24,128,1,float16,float16,31,0.30013439655303953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,24,24,128,1,float16,float16,63,0.32578558921813966
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,24,24,128,1,float16,float16,127,0.5791232109069824
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,24,24,128,1,float16,float16,255,1.1727343559265138
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,24,24,128,1,float16,float16,3,0.5631472110748291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,24,24,128,1,float16,float16,1,0.5590015888214112
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,24,24,128,1,float16,float16,511,2.2516223907470705
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,24,24,128,1,float16,float16,7,0.5656576156616211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,24,24,128,1,float16,float16,15,0.5603328227996827
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,24,24,128,1,float16,float16,63,0.6375936031341553
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,24,24,128,1,float16,float16,31,0.5820928096771241
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,1,0.024883200228214265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,24,24,128,1,float16,float16,127,1.1511296272277831
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,7,0.02467840015888214
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,15,0.025139200687408447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,3,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,24,24,128,1,float16,float16,255,2.251878356933594
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,31,0.025088000297546386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,127,0.025651198625564576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,63,0.025190401077270507
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,511,0.062105602025985716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,255,0.03461120128631592
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,2047,0.3201535940170288
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,1023,0.14028799533843994
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,24,24,128,1,float16,float16,4095,0.6554624080657959
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,24,24,128,1,float16,float16,1,1.072332763671875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,24,24,128,1,float16,float16,3,1.0718208312988282
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,24,24,128,1,float16,float16,15,1.0901503562927246
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,24,24,128,1,float16,float16,63,1.2513792037963867
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,24,24,128,1,float16,float16,31,1.1321855545043946
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,24,24,128,1,float16,float16,7,1.0777600288391114
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,24,24,128,1,float16,float16,127,2.336511993408203
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,24,24,128,1,float16,float16,1,2.102016067504883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,24,24,128,1,float16,float16,3,2.1097984313964844
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,24,24,128,1,float16,float16,7,2.1217792510986326
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,1,0.031839999556541446
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,24,24,128,1,float16,float16,15,2.1647872924804688
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,24,24,128,1,float16,float16,31,2.223513603210449
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,3,0.031948798894882204
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,7,0.031539198756217954
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,24,24,128,1,float16,float16,63,2.485094451904297
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,15,0.032204800844192506
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,31,0.03189760148525238
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,63,0.03297280073165894
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,127,0.033583998680114746
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,511,0.14443520307540894
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,255,0.04715520143508911
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,1023,0.3181488037109375
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,4095,1.1791359901428222
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,24,24,128,1,float16,float16,2047,0.6683648109436036
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,1,0.06568959951400757
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,3,0.06528000235557556
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,7,0.06656000018119812
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,15,0.06517760157585144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,31,0.06686720252037048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,63,0.06758400201797485
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,127,0.06840320229530335
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,255,0.2234879970550537
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,511,0.41118721961975097
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,1023,0.7949312210083008
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,2047,1.607321548461914
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,3,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,4095,3.261747360229492
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,1,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,15,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,16,128,1,float16,float16,8191,6.123822402954102
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,7,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,127,0.01791680008172989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,255,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,511,0.023756800591945647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,1023,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,2047,0.032358399033546446
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,8191,0.061337602138519284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,16,128,1,float16,float16,4095,0.04116480052471161
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,7,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,15,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,31,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,63,0.017659200727939604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,127,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,1023,0.03333120048046112
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,511,0.02815999984741211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,255,0.022067199647426605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,2047,0.04111360013484955
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,8191,0.227891206741333
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,16,128,1,float16,float16,4095,0.06272000074386597
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,7,0.03937279880046844
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,1,0.03916000127792359
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,3,0.03870719969272614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,15,0.038553598523139956
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,63,0.039628800749778745
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,255,0.05990399718284607
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,31,0.039419201016426084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,127,0.04080640077590943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,511,0.22108159065246583
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,1023,0.41216001510620115
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,2047,0.7990272045135498
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,4095,1.5705599784851074
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,3,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,7,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,1,0.01781280040740967
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,16,128,1,float16,float16,8191,3.004313659667969
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,15,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,31,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,127,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,255,0.022732800245285033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,511,0.03287039995193482
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,2047,0.0645632028579712
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,4095,0.22732799053192138
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,1023,0.04089919924736023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,16,128,1,float16,float16,8191,0.41175041198730467
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,1,0.11228159666061402
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,3,0.1124351978302002
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,15,0.11530239582061767
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,7,0.11530239582061767
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,31,0.11514879465103149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,63,0.11724799871444702
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,127,0.20142080783843994
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,255,0.3890687942504883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,511,0.7756288051605225
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,1023,1.5345664024353027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,1,0.2061311960220337
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,3,0.20674560070037842
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,2047,3.130726432800293
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,15,0.20874240398406982
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,16,128,1,float16,float16,4095,6.088907241821289
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,7,0.20700159072875976
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,31,0.20904960632324218
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,63,0.2353663921356201
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,127,0.38835198879241944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,255,0.7616000175476074
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,511,1.6796159744262695
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,16,128,1,float16,float16,3,0.38558719158172605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,16,128,1,float16,float16,7,0.3931135892868042
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,16,128,1,float16,float16,1,0.39070720672607423
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,16,128,1,float16,float16,15,0.389683198928833
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,16,128,1,float16,float16,1023,3.3518081665039063
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,16,128,1,float16,float16,31,0.3950079917907715
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,16,128,1,float16,float16,63,0.43161602020263673
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,16,128,1,float16,float16,127,0.7617536067962647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,16,128,1,float16,float16,255,1.5341567993164062
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,1,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,3,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,7,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,15,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,31,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,16,128,1,float16,float16,511,2.9869056701660157
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,63,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,255,0.023756800591945647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,127,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,511,0.039987200498580934
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,1023,0.06415359973907471
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,2047,0.22574079036712646
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,4095,0.40944638252258303
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,16,128,1,float16,float16,8191,0.8675328254699707
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,16,128,1,float16,float16,1,0.7352831840515137
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,16,128,1,float16,float16,3,0.7322624206542969
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,16,128,1,float16,float16,7,0.7349199771881103
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,16,128,1,float16,float16,15,0.7368703842163086
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,16,128,1,float16,float16,63,0.840447998046875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,16,128,1,float16,float16,31,0.7647232055664063
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,16,128,1,float16,float16,127,1.5343104362487794
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,16,128,1,float16,float16,1,1.419926357269287
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,16,128,1,float16,float16,7,1.4235648155212401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,16,128,1,float16,float16,15,1.4517760276794434
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,16,128,1,float16,float16,3,1.4262271881103517
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,16,128,1,float16,float16,255,2.9884416580200197
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,16,128,1,float16,float16,31,1.4883824348449708
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,16,128,1,float16,float16,63,1.662771224975586
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,16,128,1,float16,float16,127,3.151411247253418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,3,0.025446400046348572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,1,0.025446400046348572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,7,0.02534399926662445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,15,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,127,0.02682879865169525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,63,0.025591999292373657
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,31,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,255,0.03573760092258453
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,511,0.06471679806709289
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,1023,0.2243583917617798
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,2047,0.42112002372741697
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,4095,0.789296007156372
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,3,0.053350400924682614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,16,128,1,float16,float16,8191,1.5319040298461915
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,1,0.053039997816085815
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,7,0.05294079780578613
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,15,0.05304319858551025
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,31,0.05283839702606201
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,63,0.053299200534820554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,127,0.05437440276145935
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,255,0.0932864010334015
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,511,0.3441663980484009
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,2047,1.1904512405395509
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,1023,0.6003136157989502
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,4095,2.3969791412353514
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,1,0.017254400253295898
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,3,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,7,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,15,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,12,12,128,1,float16,float16,8191,4.699187088012695
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,63,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,255,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,1023,0.02892799973487854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,511,0.024268800020217897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,2047,0.029337599873542786
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,4095,0.037990400195121767
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,12,12,128,1,float16,float16,8191,0.050329601764678954
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,3,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,15,0.01791680008172989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,7,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,31,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,63,0.017705599963665008
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,255,0.022169600427150726
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,511,0.025446400046348572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,4095,0.059443199634552
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,2047,0.037529599666595456
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,1023,0.029900801181793214
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,12,12,128,1,float16,float16,8191,0.17161920070648193
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,1,0.032204800844192506
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,3,0.032051199674606325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,7,0.03246079981327057
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,31,0.03261440098285675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,15,0.032051199674606325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,63,0.03312639892101288
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,255,0.047513601183891294
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,511,0.1400320053100586
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,127,0.03374080061912536
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,1023,0.319488000869751
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,2047,0.6113279819488525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,4095,1.1659775733947755
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,12,12,128,1,float16,float16,8191,2.379724884033203
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,31,0.017862400412559508
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,63,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,15,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,127,0.0187376007437706
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,255,0.022169600427150726
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,511,0.035473600029945374
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,1023,0.04730879962444305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,2047,0.07224159836769103
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,4095,0.18078720569610596
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,12,12,128,1,float16,float16,8191,0.4059648036956787
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,1,0.08714240193367004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,3,0.0866815984249115
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,7,0.08785920143127442
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,15,0.08770560026168824
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,31,0.08760319948196411
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,63,0.0896511971950531
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,255,0.2979248046875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,127,0.09937919974327088
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,511,0.5931007862091064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,1023,1.1623920440673827
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,2047,2.361599922180176
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,7,0.16097279787063598
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,3,0.16117759943008422
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,1,0.16209919452667237
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,15,0.16358400583267213
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,12,12,128,1,float16,float16,4095,4.574873733520508
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,31,0.16501760482788086
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,63,0.16752640008926392
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,127,0.3278847932815552
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,255,0.5787055969238282
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,511,1.1480575561523438
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,12,12,128,1,float16,float16,1,0.3035583972930908
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,12,12,128,1,float16,float16,3,0.30100479125976565
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,12,12,128,1,float16,float16,1023,2.380339241027832
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,12,12,128,1,float16,float16,15,0.3014656066894531
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,12,12,128,1,float16,float16,7,0.30208001136779783
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,12,12,128,1,float16,float16,31,0.30602240562438965
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,12,12,128,1,float16,float16,63,0.33464319705963136
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,12,12,128,1,float16,float16,127,0.5805056095123291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,12,12,128,1,float16,float16,255,1.1465727806091308
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,1,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,12,12,128,1,float16,float16,511,2.2576128005981446
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,3,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,15,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,31,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,7,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,63,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,127,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,255,0.02314240038394928
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,1023,0.06394879817962647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,511,0.044844800233840944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,2047,0.13946880102157594
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,4095,0.3247616052627563
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,12,12,128,1,float16,float16,8191,0.6014944076538086
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,12,12,128,1,float16,float16,1,0.5633535861968995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,12,12,128,1,float16,float16,3,0.560588788986206
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,12,12,128,1,float16,float16,7,0.5694464206695556
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,12,12,128,1,float16,float16,15,0.5638656139373779
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,12,12,128,1,float16,float16,31,0.5830080032348632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,12,12,128,1,float16,float16,63,0.6498223781585694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,12,12,128,1,float16,float16,127,1.154201602935791
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,12,12,128,1,float16,float16,3,1.0779647827148438
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,12,12,128,1,float16,float16,1,1.0810367584228515
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,12,12,128,1,float16,float16,7,1.086406421661377
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,12,12,128,1,float16,float16,255,2.3324655532836913
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,12,12,128,1,float16,float16,15,1.090559959411621
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,12,12,128,1,float16,float16,31,1.1482080459594726
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,12,12,128,1,float16,float16,63,1.2880895614624024
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,12,12,128,1,float16,float16,127,2.332569694519043
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,1,0.0247296005487442
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,7,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,3,0.024831999838352204
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,31,0.024985599517822265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,15,0.024883200228214265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,127,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,63,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,255,0.034560000896453856
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,511,0.0626688003540039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,1023,0.1410048007965088
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,2047,0.32179200649261475
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,3,0.0394239991903305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,1,0.03967199921607971
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,8191,1.1631104469299316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,12,12,128,1,float16,float16,4095,0.5986815929412842
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,7,0.0394239991903305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,15,0.03937279880046844
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,31,0.03947519958019256
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,63,0.03993600010871887
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,127,0.04085760116577149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,255,0.060211199522018435
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,511,0.2392064094543457
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,1023,0.4148687839508057
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,2047,0.8027647972106934
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,4095,1.5745535850524903
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,8191,3.1354864120483397
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,1,0.018268799781799315
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,15,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,8,128,1,float16,float16,16383,6.2593025207519535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,63,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,255,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,1023,0.027641600370407103
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,511,0.022732800245285033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,2047,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,4095,0.03251200020313263
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,8191,0.04101119935512543
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,8,128,1,float16,float16,16383,0.061952000856399535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,1,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,7,0.01775680035352707
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,3,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,15,0.01791200041770935
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,31,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,1023,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,511,0.02385919988155365
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,2047,0.03297280073165894
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,4095,0.0414112001657486
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,8191,0.061947202682495116
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,8,128,1,float16,float16,16383,0.22824480533599853
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,3,0.025446400046348572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,15,0.025804799795150758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,7,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,1,0.025139200687408447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,31,0.025753599405288697
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,63,0.026214399933815004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,127,0.026924800872802735
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,255,0.03562879860401154
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,511,0.06481919884681701
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,1023,0.22430720329284667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,4095,0.7905792236328125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,2047,0.4226560115814209
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,8191,1.5689215660095215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,1,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,8,128,1,float16,float16,16383,3.126323127746582
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,7,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,15,0.017867200076580048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,31,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,255,0.02237440049648285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,511,0.028364801406860353
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,1023,0.03287039995193482
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,2047,0.04101119935512543
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,4095,0.06328319907188415
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,8191,0.22819840908050537
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,8,128,1,float16,float16,16383,0.4119552135467529
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,1,0.0667136013507843
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,3,0.0671231985092163
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,7,0.06661120057106018
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,15,0.0681984007358551
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,63,0.06783999800682068
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,31,0.06855679750442505
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,127,0.06666240096092224
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,255,0.20505599975585936
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,511,0.41164798736572267
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,1023,0.8003071784973145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,2047,1.6369663238525392
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,1,0.11571199893951416
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,3,0.11432960033416747
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,4095,3.2222721099853517
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,7,0.11571199893951416
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,8,128,1,float16,float16,8191,6.139750289916992
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,15,0.11658240556716919
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,63,0.1186303973197937
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,31,0.11770880222320557
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,127,0.20234239101409912
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,255,0.3927551984786987
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,511,0.8630784034729004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,1023,1.5749119758605956
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,1,0.2095616102218628
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,3,0.20833280086517333
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,15,0.21206560134887695
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,8,128,1,float16,float16,2047,3.4335742950439454
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,7,0.21093919277191162
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,31,0.21125121116638185
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,63,0.22548480033874513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,127,0.43468799591064455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,255,0.7669760227203369
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,511,1.5682559967041017
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,3,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,7,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,8,128,1,float16,float16,1023,3.053107261657715
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,31,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,63,0.018171200156211854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,511,0.0330704003572464
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,1023,0.04095999896526337
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,127,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,255,0.02222079932689667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,2047,0.06307839751243591
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,4095,0.22804160118103028
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,8191,0.4146687984466553
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,8,128,1,float16,float16,16383,0.7862271785736084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,8,128,1,float16,float16,1,0.38727679252624514
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,8,128,1,float16,float16,3,0.387225604057312
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,8,128,1,float16,float16,7,0.38840320110321047
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,8,128,1,float16,float16,31,0.39726080894470217
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,8,128,1,float16,float16,15,0.3932159900665283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,8,128,1,float16,float16,63,0.43683838844299316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,8,128,1,float16,float16,127,0.7607295989990235
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,8,128,1,float16,float16,255,1.539020824432373
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,8,128,1,float16,float16,1,0.7311359882354737
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,8,128,1,float16,float16,3,0.7358975887298584
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,8,128,1,float16,float16,15,0.7421951770782471
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,8,128,1,float16,float16,511,3.1045631408691405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,8,128,1,float16,float16,31,0.7635456085205078
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,8,128,1,float16,float16,7,0.7370751857757568
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,8,128,1,float16,float16,63,0.8335359573364258
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,8,128,1,float16,float16,127,1.4968319892883302
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,1,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,3,0.018838399648666383
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,7,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,15,0.018833599984645844
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,63,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,8,128,1,float16,float16,255,3.0992895126342774
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,31,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,127,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,255,0.02375520020723343
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,511,0.039987200498580934
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,1023,0.06364160180091857
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,2047,0.23218400478363038
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,4095,0.45352959632873535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,8191,0.7853040218353271
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,8,128,1,float16,float16,16383,1.5916031837463378
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,1,0.026105600595474242
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,3,0.027084800601005554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,7,0.02611039876937866
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,15,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,31,0.026265600323677064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,63,0.026265600323677064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,255,0.03614720106124878
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,127,0.027084800601005554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,511,0.06604160070419311
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,1023,0.22446079254150392
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,2047,0.4216832160949707
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,4095,0.7964159965515136
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,8191,1.5590911865234376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,16383,3.1450624465942383
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,7,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,15,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,31,0.017351999878883362
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,4,128,1,float16,float16,32767,6.292121505737304
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,63,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,255,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,2047,0.027852800488471986
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,1023,0.026931199431419372
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,511,0.022784000635147093
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,4095,0.02821120023727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,8191,0.03343360126018524
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,16383,0.04203520119190216
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,4,128,1,float16,float16,32767,0.06246240139007568
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,1,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,3,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,15,0.017660799622535705
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,63,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,255,0.02227199971675873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,511,0.023347200453281404
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,1023,0.028415998816490172
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,2047,0.02959359884262085
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,4095,0.03363839983940124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,8191,0.04110879898071289
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,16383,0.06159359812736511
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,4,128,1,float16,float16,32767,0.22865920066833495
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,1,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,7,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,3,0.019247999787330626
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,15,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,127,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,63,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,31,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,255,0.02391040027141571
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,511,0.040038400888442995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,2047,0.22783999443054198
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,1023,0.06307520270347595
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,4095,0.4109312057495117
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,8191,0.7812607765197754
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,16383,1.5747072219848632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,3,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,7,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,4,128,1,float16,float16,32767,3.0185983657836912
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,15,0.01801760047674179
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,63,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,255,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,1023,0.02892639935016632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,2047,0.03353599905967712
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,4095,0.04188160002231598
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,511,0.02467840015888214
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,8191,0.06097919940948486
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,16383,0.22824320793151856
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,1,0.040038400888442995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,4,128,1,float16,float16,32767,0.41390080451965333
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,3,0.04039680063724518
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,7,0.03993600010871887
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,15,0.040550398826599124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,31,0.04039680063724518
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,63,0.0404992014169693
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,127,0.04121600091457367
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,255,0.06097919940948486
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,511,0.22292160987854004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,1023,0.4143104076385498
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,2047,0.8030688285827636
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,4095,1.5696895599365235
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,1,0.06768640279769897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,8191,3.1365631103515623
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,3,0.06783999800682068
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,7,0.06891520023345947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,15,0.06753280162811279
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,31,0.06814720034599304
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,4,128,1,float16,float16,16383,6.265958404541015
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,63,0.07004160284996033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,127,0.0689087986946106
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,255,0.2052095890045166
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,511,0.45096960067749026
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,1023,0.7987135887145996
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,2047,1.6297983169555663
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,1,0.12026879787445069
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,3,0.11960320472717285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,4,128,1,float16,float16,4095,3.23768310546875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,7,0.11965440511703491
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,15,0.1197983980178833
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,31,0.12082240581512452
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,63,0.12170239686965942
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,127,0.22574079036712646
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,255,0.3952640056610107
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,511,0.7824895858764649
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,1023,1.684377670288086
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,4,128,1,float16,float16,2047,3.248179244995117
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,3,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,31,0.01791519969701767
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,63,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,255,0.022067199647426605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,511,0.02913280129432678
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,1023,0.0336896002292633
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,4095,0.0626688003540039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,2047,0.04162560105323791
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,8191,0.22963199615478516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,16383,0.4140031814575195
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,4,128,1,float16,float16,32767,0.9226240158081055
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,1,0.208076810836792
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,3,0.20843520164489746
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,7,0.2116096019744873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,15,0.21079039573669434
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,31,0.21365759372711182
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,63,0.22937440872192383
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,127,0.41190400123596194
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,255,0.7724544048309326
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,511,1.53635835647583
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,4,128,1,float16,float16,1,0.39152638912200927
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,4,128,1,float16,float16,1023,3.163438415527344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,4,128,1,float16,float16,3,0.3857408046722412
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,4,128,1,float16,float16,15,0.39127039909362793
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,4,128,1,float16,float16,31,0.39301118850708006
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,4,128,1,float16,float16,7,0.38993918895721436
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,4,128,1,float16,float16,63,0.43781118392944335
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,4,128,1,float16,float16,127,0.8123392105102539
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,4,128,1,float16,float16,255,1.6648191452026366
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,3,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,15,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,1,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,4,128,1,float16,float16,511,2.995968055725098
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,31,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,63,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,127,0.018532800674438476
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,511,0.033792001008987424
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,255,0.022732800245285033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,1023,0.04142079949378967
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,2047,0.06507520079612732
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,8191,0.4147200107574463
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,4095,0.23147039413452147
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,16383,0.7865856170654297
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,1,0.019551999866962433
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,3,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,4,128,1,float16,float16,32767,1.5517696380615233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,7,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,15,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,31,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,63,0.01950239986181259
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,127,0.020019200444221497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,255,0.02452480047941208
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,511,0.04079999923706055
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,1023,0.06394879817962647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,2047,0.2214400053024292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,4095,0.40007681846618653
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,8191,0.7606272220611572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,16383,1.528831958770752
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,1,0.01648640036582947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,32767,3.059199905395508
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,3,0.016128000617027283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,7,0.016383999586105348
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,15,0.016128000617027283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,31,0.016281600296497344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,2,128,1,float16,float16,65535,5.755340957641602
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,63,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,127,0.01664000004529953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,511,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,255,0.020377600193023683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,1023,0.025543999671936036
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,2047,0.026009601354599
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,4095,0.02682879865169525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,8191,0.029183998703956604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,32767,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,65535,0.06225759983062744
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,2,128,1,float16,float16,16383,0.03333120048046112
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,1,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,3,0.01786559969186783
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,7,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,31,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,255,0.02222079932689667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,1023,0.027289599180221558
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,511,0.02314240038394928
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,127,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,2047,0.028467199206352232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,4095,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,8191,0.035071998834609985
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,16383,0.04341759979724884
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,65535,0.22108159065246583
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,2,128,1,float16,float16,32767,0.06359040141105651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,3,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,7,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,1,0.017961600422859193
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,31,0.018166400492191315
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,15,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,63,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,255,0.022886399924755097
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,127,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,511,0.03425279855728149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,1023,0.04223999977111816
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,4095,0.22405118942260743
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,2047,0.06589440107345582
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,8191,0.40366082191467284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,16383,0.765337610244751
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,32767,1.4846976280212403
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,3,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,2,128,1,float16,float16,65535,2.920089530944824
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,31,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,511,0.02380799949169159
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,255,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,1023,0.028364801406860353
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,2047,0.02892799973487854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,4095,0.03306719958782196
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,8191,0.04321280121803284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,16383,0.06389759778976441
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,32767,0.22282240390777588
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,2,128,1,float16,float16,65535,0.40181121826171873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,1,0.026368001103401185
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,3,0.02682879865169525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,15,0.026521599292755126
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,31,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,7,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,63,0.027187201380729675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,127,0.027801600098609925
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,511,0.06656000018119812
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,255,0.036556801199913024
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,1023,0.21980159282684325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,2047,0.41103358268737794
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,4095,0.7668735980987549
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,8191,1.5251456260681153
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,16383,3.0555648803710938
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,1,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,3,0.0417279988527298
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,7,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,15,0.04131839871406555
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,31,0.0417279988527298
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,2,128,1,float16,float16,32767,6.131865692138672
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,127,0.043007999658584595
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,63,0.0419840008020401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,255,0.061440002918243405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,511,0.22128639221191407
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,1023,0.4065279960632324
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,2047,0.7847936153411865
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,4095,1.5451135635375977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,3,0.07198560237884521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,1,0.07152640223503112
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,7,0.07121919989585876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,2,128,1,float16,float16,8191,3.181056022644043
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,15,0.07193599939346314
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,31,0.07188479900360108
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,63,0.07275519967079162
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,127,0.0732159972190857
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,255,0.20397439002990722
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,1023,0.7801343917846679
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,511,0.40560641288757326
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,2047,1.6581119537353515
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,2,128,1,float16,float16,4095,3.153919982910156
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,127,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,63,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,255,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,511,0.02462719976902008
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,1023,0.029286399483680725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,4095,0.04223999977111816
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,2047,0.03327839970588684
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,8191,0.0637440025806427
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,16383,0.22322399616241456
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,32767,0.40729598999023436
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,2,128,1,float16,float16,65535,0.9540608406066895
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,1,0.11878399848937989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,7,0.11970560550689698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,3,0.11991039514541627
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,15,0.12026879787445069
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,31,0.11975680589675904
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,63,0.1207808017730713
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,127,0.20336639881134033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,255,0.38348801136016847
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,511,0.7625728130340577
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,1023,1.5312383651733399
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,1,0.20884480476379394
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,3,0.20792319774627685
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,2,128,1,float16,float16,2047,3.0454784393310548
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,7,0.2100735902786255
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,31,0.21329920291900634
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,63,0.24002559185028077
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,15,0.2110896110534668
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,127,0.4253695964813232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,255,0.744652795791626
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,1,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,511,1.6254928588867188
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,2,128,1,float16,float16,1023,3.121459197998047
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,3,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,31,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,63,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,127,0.018172800540924072
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,2047,0.042393600940704344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,1023,0.034406399726867674
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,511,0.0289792001247406
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,8191,0.2242527961730957
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,4095,0.06359040141105651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,16383,0.40221920013427737
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,32767,0.8959487915039063
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,1,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,3,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,7,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,2,128,1,float16,float16,65535,1.7744895935058593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,15,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,31,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,63,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,127,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,255,0.022886399924755097
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,511,0.034918400645256045
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,1023,0.042956799268722534
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,2047,0.06574079990386963
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,4095,0.22405118942260743
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,8191,0.4099071979522705
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,16383,0.7763936042785644
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,32767,1.5392239570617676
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,1,0.015513600409030914
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,7,0.01586720049381256
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,3,0.015513600409030914
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,65535,2.9401599884033205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,15,0.015667200088500977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,31,0.015868799388408662
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,1,1,128,1,float16,float16,131071,5.820057678222656
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,63,0.015820799767971037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,255,0.019916799664497376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,127,0.015820799767971037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,511,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,1023,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,4095,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,2047,0.025446400046348572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,8191,0.027750399708747864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,16383,0.031590399146080014
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,65535,0.04787200093269348
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,32767,0.03578880131244659
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,1,1,128,1,float16,float16,131071,0.06763359904289246
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,1,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,3,0.01647839993238449
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,7,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,15,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,31,0.01648640036582947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,63,0.016383999586105348
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,255,0.020377600193023683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,1023,0.026368001103401185
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,511,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,2047,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,8191,0.029286399483680725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,4095,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,16383,0.036652800440788266
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,32767,0.04490239918231964
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,65535,0.06502400040626526
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,1,1,128,1,float16,float16,131071,0.22522881031036376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,1,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,15,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,3,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,31,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,7,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,63,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,127,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,511,0.03010239899158478
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,255,0.02252800017595291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,1023,0.034918400645256045
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,2047,0.04259839951992035
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,4095,0.06333439946174621
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,8191,0.22548480033874513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,16383,0.4113920211791992
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,32767,0.9039360046386719
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,65535,1.7532415390014648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,1,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,1,1,128,1,float16,float16,131071,3.590963363647461
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,7,0.0174575999379158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,31,0.017508800327777862
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,15,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,127,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,511,0.023552000522613525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,1023,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,2047,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,4095,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,8191,0.035071998834609985
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,16383,0.04613119959831238
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,32767,0.06594560146331788
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,65535,0.2263040065765381
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,1,0.019865599274635316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,1,1,128,1,float16,float16,131071,0.4107776165008545
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,3,0.019912000000476836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,7,0.019763199985027312
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,31,0.019916799664497376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,15,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,63,0.020479999482631683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,127,0.02088959962129593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,511,0.04177919924259186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,255,0.025036799907684325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,1023,0.06528000235557556
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,2047,0.22323200702667237
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,4095,0.4048895835876465
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,8191,0.7698944091796875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,16383,1.5445504188537598
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,1,0.02784479856491089
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,32767,3.071641540527344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,3,0.02821120023727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,7,0.027852800488471986
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,15,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,1,1,128,1,float16,float16,65535,5.834137725830078
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,31,0.02815999984741211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,63,0.028467199206352232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,127,0.029235199093818665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,255,0.03834879994392395
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,1023,0.22215359210968016
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,511,0.06876159906387329
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,2047,0.41333761215209963
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,4095,0.8611328125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,1,0.0444927990436554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,8191,1.537279987335205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,15,0.04464640021324158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,3,0.04469760060310364
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,7,0.04459519982337952
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,1,1,128,1,float16,float16,16383,2.9418415069580077
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,31,0.04428800046443939
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,63,0.045311999320983884
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,127,0.046028798818588255
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,511,0.2240000009536743
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,1023,0.41159682273864745
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,255,0.0652783989906311
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,2047,0.7870463848114013
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,4095,1.671116828918457
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,1,1,128,1,float16,float16,8191,3.050137519836426
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,3,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,1,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,15,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,7,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,63,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,127,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,1023,0.028667199611663818
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,255,0.02887679934501648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,511,0.024003200232982635
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,2047,0.030054399371147157
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,8191,0.04357120096683502
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,4095,0.03363839983940124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,16383,0.06702079772949218
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,32767,0.22620160579681398
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,131071,0.7712768077850342
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,1,1,128,1,float16,float16,65535,0.4178944110870361
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,1,0.0717311978340149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,3,0.0717311978340149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,7,0.07178239822387696
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,15,0.07162879705429077
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,31,0.07219200134277344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,63,0.07301120162010193
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,127,0.07306240200996399
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,255,0.208076810836792
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,511,0.40816640853881836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,1023,0.7863808155059815
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,2047,1.569331169128418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,3,0.1198591947555542
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,1,1,128,1,float16,float16,4095,3.1878143310546876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,1,0.11934720277786255
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,7,0.11991039514541627
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,15,0.11950080394744873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,31,0.12123680114746094
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,63,0.12124160528182984
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,127,0.2239487886428833
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,255,0.3911168098449707
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,511,0.8487423896789551
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,1023,1.5441408157348633
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,1,1,128,1,float16,float16,2047,3.0572015762329103
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,511,0.025139200687408447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,1023,0.029286399483680725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,2047,0.03399679958820343
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,4095,0.04229120016098022
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,8191,0.06389759778976441
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,16383,0.22906880378723143
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,32767,0.4103151798248291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,65535,0.9467904090881347
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,1,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,1,1,128,1,float16,float16,131071,1.902284812927246
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,3,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,7,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,15,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,63,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,31,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,127,0.019968000054359437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,255,0.02380799949169159
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,511,0.03911679983139038
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,2047,0.08038399815559387
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,1,128,1,float16,float16,1023,0.04741120040416717
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,1,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,3,0.020172800123691558
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,7,0.020172800123691558
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,15,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,31,0.019865599274635316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,63,0.020838400721549986
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,127,0.020377600193023683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,255,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,1023,0.06717439889907836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,511,0.04152320027351379
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,2,128,1,float16,float16,2047,0.22855679988861083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,1,0.026158401370048524
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,3,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,7,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,15,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,31,0.026111999154090883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,127,0.0274944007396698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,63,0.026464000344276428
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,255,0.03619840145111084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,511,0.06661120057106018
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,1023,0.2312704086303711
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,4,128,1,float16,float16,2047,0.4277167797088623
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,1,0.03911679983139038
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,7,0.03947519958019256
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,3,0.03983359932899475
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,15,0.039680001139640805
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,63,0.040089601278305055
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,31,0.03993600010871887
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,127,0.04080640077590943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,255,0.06036480069160462
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,511,0.22963199615478516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,1023,0.42239999771118164
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,1,0.016025599837303162
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,3,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,128,8,128,1,float16,float16,2047,0.8104448318481445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,7,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,15,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,31,0.01653759926557541
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,63,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,127,0.01664000004529953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,255,0.02016959935426712
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,1023,0.026419198513031004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,2047,0.026470398902893065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,1,128,1,float16,float16,511,0.02247679978609085
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,1,0.016281600296497344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,3,0.01664000004529953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,7,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,15,0.01648640036582947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,63,0.017148800194263458
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,31,0.016025599837303162
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,127,0.01669120043516159
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,255,0.020531199872493744
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,511,0.021657599508762358
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,1023,0.026212799549102783
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,1,0.017505599558353423
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,2,128,1,float16,float16,2047,0.026726400852203368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,7,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,3,0.017100800573825837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,15,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,63,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,31,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,255,0.022169600427150726
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,511,0.02263039946556091
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,1023,0.02687999904155731
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,4,128,1,float16,float16,2047,0.027596798539161683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,1,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,3,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,31,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,63,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,255,0.02149440050125122
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,511,0.022681599855422972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,1023,0.027238398790359497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,1,0.0171424001455307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,128,8,128,1,float16,float16,2047,0.028415998816490172
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,3,0.017100800573825837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,7,0.016998399794101716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,15,0.017049600183963776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,31,0.01674239933490753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,63,0.017049600183963776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,127,0.017151999473571777
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,511,0.02309119999408722
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,255,0.021452799439430237
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,1023,0.02687999904155731
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,1,128,1,float16,float16,2047,0.0274944007396698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,1,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,7,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,31,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,63,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,255,0.022169600427150726
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,127,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,511,0.023296000063419343
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,1023,0.02739199995994568
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,2,128,1,float16,float16,2047,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,1,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,3,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,63,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,127,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,255,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,511,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,1023,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,4,128,1,float16,float16,2047,0.02887679934501648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,3,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,7,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,15,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,31,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,63,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,127,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,255,0.022886399924755097
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,511,0.024217599630355836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,1023,0.028364801406860353
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,128,8,128,1,float16,float16,2047,0.033225598931312564
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,1,0.018430399894714355
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,3,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,7,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,15,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,63,0.018935999274253844
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,31,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,127,0.019097599387168884
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,255,0.022835199534893037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,511,0.033020800352096556
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,1023,0.03819519877433777
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,1,128,1,float16,float16,2047,0.04633600115776062
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,1,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,7,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,3,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,15,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,31,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,127,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,255,0.022886399924755097
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,511,0.03563520014286041
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,63,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,2047,0.06747840046882629
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,2,128,1,float16,float16,1023,0.04382719993591309
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,1,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,3,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,15,0.019139200448989868
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,7,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,31,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,63,0.019097599387168884
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,127,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,255,0.02446880042552948
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,511,0.040652799606323245
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,1023,0.06410239934921265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,4,128,1,float16,float16,2047,0.23173120021820068
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,3,0.025497600436210632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,1,0.02534399926662445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,7,0.025548800826072693
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,15,0.026470398902893065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,31,0.025856000185012818
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,63,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,255,0.03594239950180054
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,127,0.026726400852203368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,511,0.06533120274543762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,1023,0.22763519287109374
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,1,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,128,8,128,1,float16,float16,2047,0.42546401023864744
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,3,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,7,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,15,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,31,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,63,0.01955360025167465
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,127,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,255,0.02247679978609085
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,511,0.02457599937915802
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,1023,0.02826080024242401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,1,128,1,float16,float16,2047,0.029491201043128967
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,1,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,3,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,15,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,31,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,63,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,127,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,511,0.023756800591945647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,1023,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,2,128,1,float16,float16,2047,0.029286399483680725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,3,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,7,0.017859199643135072
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,15,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,31,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,63,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,511,0.024371199309825897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,1023,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,4,128,1,float16,float16,2047,0.03363839983940124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,1,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,7,0.017611199617385866
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,31,0.017862400412559508
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,15,0.017502400279045104
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,63,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,255,0.02237440049648285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,511,0.028672000765800475
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,1023,0.03338240087032318
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,128,8,128,1,float16,float16,2047,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,3,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,1,0.021499200165271758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,7,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,15,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,31,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,63,0.02222079932689667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,127,0.02242559939622879
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,511,0.04407840073108673
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,255,0.026624000072479247
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,1,128,1,float16,float16,1023,0.0692736029624939
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,1,0.027647998929023743
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,3,0.02754240036010742
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,7,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,15,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,63,0.028203201293945313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,31,0.027952000498771667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,255,0.037836799025535585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,127,0.02887519896030426
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,511,0.07019519805908203
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,2,128,1,float16,float16,1023,0.2323456048965454
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,1,0.040755200386047366
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,3,0.040191999077796935
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,7,0.04039680063724518
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,15,0.040038400888442995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,31,0.040703999996185306
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,63,0.04039680063724518
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,127,0.0417279988527298
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,255,0.06062080264091492
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,511,0.2376703977584839
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,4,128,1,float16,float16,1023,0.4298240184783936
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,1,0.06702079772949218
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,3,0.06625279784202576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,7,0.066048002243042
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,15,0.0675823986530304
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,63,0.06799359917640686
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,31,0.06732800006866455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,127,0.06840320229530335
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,255,0.21862399578094482
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,511,0.430079984664917
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,1,128,1,float16,float16,1,0.03101919889450073
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,128,8,128,1,float16,float16,1023,0.8180224418640136
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,1,128,1,float16,float16,3,0.03128319978713989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,1,128,1,float16,float16,7,0.03107360005378723
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,1,128,1,float16,float16,15,0.03128319978713989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,1,128,1,float16,float16,31,0.031692799925804135
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,1,128,1,float16,float16,63,0.031999999284744264
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,1,128,1,float16,float16,127,0.033118399977684024
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,2,128,1,float16,float16,1,0.042803201079368594
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,1,128,1,float16,float16,255,0.04152320027351379
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,2,128,1,float16,float16,3,0.04254719913005829
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,2,128,1,float16,float16,7,0.042956799268722534
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,2,128,1,float16,float16,15,0.043059200048446655
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,2,128,1,float16,float16,31,0.043110400438308716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,2,128,1,float16,float16,63,0.0434112012386322
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,2,128,1,float16,float16,127,0.045311999320983884
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,2,128,1,float16,float16,255,0.06312959790229797
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,4,128,1,float16,float16,1,0.06912000179290771
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,4,128,1,float16,float16,3,0.0681984007358551
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,4,128,1,float16,float16,7,0.06891520023345947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,4,128,1,float16,float16,31,0.06901760101318359
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,4,128,1,float16,float16,63,0.07029759883880615
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,4,128,1,float16,float16,15,0.06906880140304565
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,4,128,1,float16,float16,127,0.06999040246009827
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,8,128,1,float16,float16,1,0.11412479877471923
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,4,128,1,float16,float16,255,0.23249919414520265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,8,128,1,float16,float16,3,0.11653120517730713
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,8,128,1,float16,float16,7,0.11648000478744507
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,8,128,1,float16,float16,15,0.11704319715499878
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,8,128,1,float16,float16,31,0.11786240339279175
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,8,128,1,float16,float16,63,0.11842559576034546
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,8,128,1,float16,float16,127,0.22911999225616456
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,128,8,128,1,float16,float16,255,0.42127361297607424
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,1,128,1,float16,float16,1,0.048537600040435794
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,1,128,1,float16,float16,3,0.04874080121517181
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,1,128,1,float16,float16,7,0.05012480020523071
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,1,128,1,float16,float16,15,0.04930399954319
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,1,128,1,float16,float16,31,0.04940800070762634
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,1,128,1,float16,float16,63,0.04991999864578247
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,1,128,1,float16,float16,127,0.051148802042007446
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,2,128,1,float16,float16,1,0.07296000123023987
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,2,128,1,float16,float16,3,0.07342079877853394
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,2,128,1,float16,float16,7,0.0729088008403778
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,2,128,1,float16,float16,15,0.07398399710655212
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,2,128,1,float16,float16,63,0.07383040189743043
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,2,128,1,float16,float16,31,0.0737280011177063
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,4,128,1,float16,float16,1,0.11970560550689698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,2,128,1,float16,float16,127,0.09251840114593506
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,4,128,1,float16,float16,7,0.12042239904403687
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,4,128,1,float16,float16,3,0.12098560333251954
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,4,128,1,float16,float16,15,0.1207808017730713
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,4,128,1,float16,float16,31,0.11980799436569214
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,4,128,1,float16,float16,63,0.12467199563980103
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,4,128,1,float16,float16,127,0.25804800987243653
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,8,128,1,float16,float16,3,0.20976641178131103
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,8,128,1,float16,float16,1,0.20971519947052003
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,8,128,1,float16,float16,7,0.21263360977172852
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,8,128,1,float16,float16,15,0.210534405708313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,8,128,1,float16,float16,31,0.2135551929473877
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,8,128,1,float16,float16,63,0.25963521003723145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,128,8,128,1,float16,float16,127,0.4469759941101074
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,1,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,3,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,7,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,15,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,63,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,31,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,127,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,255,0.022732800245285033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,511,0.025088000297546386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,1023,0.029900801181793214
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,1,128,1,float16,float16,2047,0.031385600566864014
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,1,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,15,0.018171200156211854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,31,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,63,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,127,0.018171200156211854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,1023,0.029747200012207032
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,511,0.025446400046348572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,2,128,1,float16,float16,2047,0.034200000762939456
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,3,0.0174575999379158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,7,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,63,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,31,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,511,0.028620800375938414
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,1023,0.03353599905967712
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,4,128,1,float16,float16,2047,0.04264959990978241
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,1,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,7,0.017867200076580048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,15,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,31,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,63,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,255,0.022579200565814972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,511,0.03317759931087494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,1023,0.04136959910392761
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,1,128,1,float16,float16,1,0.07531520128250122
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,128,8,128,1,float16,float16,2047,0.06405119895935059
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,1,128,1,float16,float16,3,0.0762880027294159
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,1,128,1,float16,float16,31,0.07745760083198547
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,1,128,1,float16,float16,7,0.07633919715881347
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,1,128,1,float16,float16,63,0.09190400242805481
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,1,128,1,float16,float16,15,0.07731199860572815
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,2,128,1,float16,float16,1,0.12093440294265748
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,2,128,1,float16,float16,3,0.11996159553527833
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,2,128,1,float16,float16,15,0.12098560333251954
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,2,128,1,float16,float16,7,0.12124160528182984
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,2,128,1,float16,float16,31,0.12344319820404052
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,4,128,1,float16,float16,1,0.2110464096069336
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,2,128,1,float16,float16,63,0.2082304000854492
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,4,128,1,float16,float16,3,0.2125312089920044
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,4,128,1,float16,float16,7,0.21299200057983397
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,4,128,1,float16,float16,31,0.2209791898727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,4,128,1,float16,float16,15,0.21340160369873046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,4,128,1,float16,float16,63,0.3052544116973877
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,8,128,1,float16,float16,1,0.39014239311218263
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,8,128,1,float16,float16,7,0.39086079597473145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,8,128,1,float16,float16,3,0.3899904012680054
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,8,128,1,float16,float16,15,0.39439361095428466
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,8,128,1,float16,float16,31,0.40524802207946775
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,1,128,1,float16,float16,1,0.21760001182556152
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,128,8,128,1,float16,float16,63,0.495359992980957
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,1,128,1,float16,float16,3,0.21765120029449464
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,1,128,1,float16,float16,7,0.21616640090942382
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,1,128,1,float16,float16,15,0.22179839611053467
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,2,128,1,float16,float16,1,0.2525696039199829
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,1,128,1,float16,float16,31,0.2401279926300049
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,2,128,1,float16,float16,3,0.2530303955078125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,2,128,1,float16,float16,15,0.2623487949371338
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,2,128,1,float16,float16,7,0.25528318881988527
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,2,128,1,float16,float16,31,0.2721791982650757
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,4,128,1,float16,float16,1,0.41292800903320315
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,4,128,1,float16,float16,3,0.4131328105926514
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,4,128,1,float16,float16,15,0.4229119777679443
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,4,128,1,float16,float16,7,0.4193280220031738
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,4,128,1,float16,float16,31,0.42588157653808595
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,8,128,1,float16,float16,1,0.7687151908874512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,8,128,1,float16,float16,3,0.7684607982635498
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,8,128,1,float16,float16,7,0.7704576015472412
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,8,128,1,float16,float16,15,0.775219202041626
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,128,8,128,1,float16,float16,31,0.7789023876190185
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,1,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,3,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,7,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,15,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,31,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,63,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,127,0.018990400433540344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,255,0.023552000522613525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,511,0.027340799570083618
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,1023,0.031385600566864014
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,1,128,1,float16,float16,2047,0.03670240044593811
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,1,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,7,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,31,0.018174399435520173
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,63,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,255,0.02237440049648285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,511,0.030156800150871278
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,1023,0.034867200255393985
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,2,128,1,float16,float16,2047,0.043059200048446655
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,1,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,3,0.017662400007247926
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,7,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,15,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,31,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,63,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,127,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,255,0.02263039946556091
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,511,0.033843201398849485
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,1023,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,4,128,1,float16,float16,2047,0.06481919884681701
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,1,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,3,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,7,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,15,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,31,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,63,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,127,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,511,0.040038400888442995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,255,0.023603199422359465
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,1023,0.06353920102119445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,128,8,128,1,float16,float16,2047,0.22906880378723143
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,1,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,3,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,7,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,15,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,63,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,31,0.019097599387168884
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,255,0.023548799753189086
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,127,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,511,0.037478399276733396
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,1023,0.046079999208450316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,1,128,1,float16,float16,2047,0.06906560063362122
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,1,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,3,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,7,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,15,0.01966080069541931
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,31,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,63,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,127,0.02088959962129593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,255,0.025033599138259886
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,511,0.04177919924259186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,1023,0.0652783989906311
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,2,128,1,float16,float16,2047,0.22492001056671143
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,1,0.026060798764228822
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,3,0.026009601354599
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,7,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,15,0.026214399933815004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,63,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,31,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,127,0.027238398790359497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,255,0.03630079925060272
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,511,0.06640639901161194
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,1023,0.23193600177764892
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,4,128,1,float16,float16,2047,0.4269567966461182
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,1,0.0394239991903305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,3,0.03941600024700165
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,7,0.03906559944152832
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,15,0.039526399970054624
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,31,0.039628800749778745
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,63,0.040140798687934874
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,127,0.04075360000133514
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,511,0.22901759147644044
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,255,0.060108798742294314
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,1023,0.4196352005004883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,96,8,128,1,float16,float16,2047,0.8094207763671875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,7,0.016740800440311433
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,1,0.015974399447441102
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,3,0.015667200088500977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,15,0.015974399447441102
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,31,0.016383999586105348
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,63,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,127,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,255,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,1023,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,511,0.02175839990377426
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,1,128,1,float16,float16,2047,0.026214399933815004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,7,0.01643519997596741
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,3,0.01648640036582947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,1,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,15,0.01653759926557541
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,63,0.01648640036582947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,31,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,127,0.016128000617027283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,255,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,511,0.021857599914073943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,2047,0.026265600323677064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,2,128,1,float16,float16,1023,0.025651198625564576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,1,0.017203199863433837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,7,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,15,0.017148800194263458
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,31,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,63,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,511,0.022579200565814972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,255,0.021555200219154358
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,1023,0.02687999904155731
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,4,128,1,float16,float16,2047,0.027801600098609925
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,1,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,3,0.017151999473571777
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,7,0.017609600722789765
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,15,0.017654399573802947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,63,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,31,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,255,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,511,0.022579200565814972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,1023,0.027289599180221558
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,96,8,128,1,float16,float16,2047,0.0287200003862381
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,1,0.016896000504493712
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,3,0.01679359972476959
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,15,0.016689600050449373
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,7,0.016683200001716615
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,31,0.01664000004529953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,63,0.01679359972476959
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,127,0.016784000396728515
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,255,0.021555200219154358
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,511,0.022681599855422972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,1023,0.02682720124721527
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,1,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,1,128,1,float16,float16,2047,0.026822400093078614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,7,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,15,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,63,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,127,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,255,0.021657599508762358
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,511,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,1023,0.027238398790359497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,2,128,1,float16,float16,2047,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,1,0.017399999499320983
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,3,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,31,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,63,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,255,0.02170879989862442
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,511,0.02309119999408722
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,1023,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,4,128,1,float16,float16,2047,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,3,0.017560000717639922
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,7,0.017611199617385866
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,31,0.017558400332927705
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,63,0.017560000717639922
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,127,0.018681600689888
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,255,0.022067199647426605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,511,0.024831999838352204
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,1023,0.02821120023727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,96,8,128,1,float16,float16,2047,0.033792001008987424
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,1,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,3,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,7,0.018272000551223754
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,15,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,31,0.01857759952545166
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,63,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,127,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,255,0.022886399924755097
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,511,0.031488001346588135
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,1023,0.0364544004201889
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,1,128,1,float16,float16,2047,0.0444927990436554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,1,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,3,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,7,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,15,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,31,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,63,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,127,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,511,0.035123199224472046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,255,0.022579200565814972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,1023,0.043673598766326906
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,2,128,1,float16,float16,2047,0.06563839912414551
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,1,0.01863519996404648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,3,0.01873600035905838
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,7,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,31,0.018987199664115904
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,15,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,63,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,127,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,255,0.024371199309825897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,511,0.04044800102710724
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,1023,0.06415359973907471
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,4,128,1,float16,float16,2047,0.23137280941009522
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,3,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,1,0.02534399926662445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,7,0.025446400046348572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,15,0.025651198625564576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,31,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,63,0.02590720057487488
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,127,0.026675200462341307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,255,0.03563520014286041
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,1023,0.22722558975219725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,511,0.06516799926757813
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,1,0.018067200481891633
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,96,8,128,1,float16,float16,2047,0.42485761642456055
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,3,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,15,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,31,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,63,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,127,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,255,0.02232320010662079
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,511,0.023449599742889404
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,1023,0.028262400627136232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,1,128,1,float16,float16,2047,0.02892799973487854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,3,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,1,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,7,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,31,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,15,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,63,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,127,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,511,0.023705600202083586
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,1023,0.02821120023727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,2,128,1,float16,float16,2047,0.029235199093818665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,3,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,7,0.018016000092029572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,31,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,63,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,127,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,255,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,511,0.024115200340747833
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,1023,0.028364801406860353
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,4,128,1,float16,float16,2047,0.0331167995929718
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,3,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,127,0.01786080002784729
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,511,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,1023,0.03358719944953918
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,96,8,128,1,float16,float16,2047,0.04162560105323791
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,1,0.02104319930076599
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,3,0.02099040001630783
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,15,0.02114560008049011
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,7,0.02114560008049011
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,31,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,63,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,127,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,255,0.026419198513031004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,511,0.043724799156188966
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,1,128,1,float16,float16,1023,0.06681600213050842
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,1,0.027136000990867614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,3,0.026982399821281432
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,7,0.027187201380729675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,15,0.02739199995994568
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,31,0.0274944007396698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,63,0.027801600098609925
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,127,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,255,0.037427198886871335
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,1023,0.22963199615478516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,2,128,1,float16,float16,511,0.06998559832572937
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,1,0.04034239947795868
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,3,0.04111360013484955
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,7,0.04044800102710724
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,15,0.040243199467658995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,31,0.040550398826599124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,63,0.040345600247383116
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,127,0.04167680144309997
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,255,0.06108160018920898
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,511,0.23541760444641113
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,4,128,1,float16,float16,1023,0.426035213470459
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,1,0.06691840291023254
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,3,0.06656000018119812
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,7,0.06625120043754577
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,15,0.06630399823188782
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,31,0.06753280162811279
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,63,0.06850559711456299
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,127,0.06763520240783691
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,255,0.2144767999649048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,511,0.4261888027191162
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,96,8,128,1,float16,float16,1023,0.8126463890075684
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,1,128,1,float16,float16,1,0.030105599761009218
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,1,128,1,float16,float16,3,0.030054399371147157
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,1,128,1,float16,float16,7,0.030105599761009218
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,1,128,1,float16,float16,15,0.0302592009305954
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,1,128,1,float16,float16,31,0.03056640028953552
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,1,128,1,float16,float16,63,0.030771198868751525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,1,128,1,float16,float16,127,0.03189760148525238
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,1,128,1,float16,float16,255,0.04075360000133514
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,2,128,1,float16,float16,1,0.042340800166130066
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,2,128,1,float16,float16,3,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,2,128,1,float16,float16,15,0.04264959990978241
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,2,128,1,float16,float16,7,0.042393600940704344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,2,128,1,float16,float16,31,0.042393600940704344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,2,128,1,float16,float16,63,0.04264959990978241
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,2,128,1,float16,float16,127,0.04387840032577515
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,2,128,1,float16,float16,255,0.06369280219078063
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,4,128,1,float16,float16,1,0.06743040084838867
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,4,128,1,float16,float16,3,0.06773759722709656
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,4,128,1,float16,float16,15,0.06891520023345947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,4,128,1,float16,float16,7,0.06850559711456299
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,4,128,1,float16,float16,31,0.06896640062332153
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,4,128,1,float16,float16,63,0.06993920207023621
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,4,128,1,float16,float16,127,0.0694271981716156
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,4,128,1,float16,float16,255,0.2261967897415161
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,8,128,1,float16,float16,1,0.11438080072402954
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,8,128,1,float16,float16,3,0.11376639604568481
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,8,128,1,float16,float16,7,0.1156607985496521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,8,128,1,float16,float16,15,0.11694079637527466
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,8,128,1,float16,float16,31,0.11770880222320557
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,8,128,1,float16,float16,63,0.1185263991355896
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,8,128,1,float16,float16,127,0.22256639003753662
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,96,8,128,1,float16,float16,255,0.41436161994934084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,1,128,1,float16,float16,1,0.04715520143508911
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,1,128,1,float16,float16,3,0.04761599898338318
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,1,128,1,float16,float16,7,0.04797439873218536
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,1,128,1,float16,float16,15,0.04761599898338318
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,1,128,1,float16,float16,31,0.04741120040416717
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,1,128,1,float16,float16,63,0.048537600040435794
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,1,128,1,float16,float16,127,0.04940800070762634
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,2,128,1,float16,float16,1,0.07234560251235962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,2,128,1,float16,float16,3,0.07208799719810485
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,2,128,1,float16,float16,7,0.07239680290222168
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,2,128,1,float16,float16,15,0.07162879705429077
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,2,128,1,float16,float16,31,0.07260159850120544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,2,128,1,float16,float16,63,0.07336959838867188
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,2,128,1,float16,float16,127,0.07505919933319091
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,4,128,1,float16,float16,1,0.12026879787445069
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,4,128,1,float16,float16,3,0.12165119647979736
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,4,128,1,float16,float16,7,0.11893759965896607
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,4,128,1,float16,float16,15,0.12124160528182984
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,4,128,1,float16,float16,31,0.12006399631500245
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,4,128,1,float16,float16,63,0.12590080499649048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,8,128,1,float16,float16,1,0.20945920944213867
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,4,128,1,float16,float16,127,0.24519679546356202
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,8,128,1,float16,float16,3,0.20884480476379394
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,8,128,1,float16,float16,7,0.21140480041503906
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,8,128,1,float16,float16,15,0.21145598888397216
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,8,128,1,float16,float16,31,0.21462559700012207
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,8,128,1,float16,float16,63,0.24837119579315187
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,3,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,1,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,96,8,128,1,float16,float16,127,0.43330559730529783
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,7,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,15,0.017867200076580048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,31,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,63,0.018428799510002137
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,127,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,255,0.02242559939622879
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,2047,0.03051519989967346
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,1023,0.02903040051460266
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,1,128,1,float16,float16,511,0.02447360008955002
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,3,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,7,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,31,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,63,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,127,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,255,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,511,0.02452480047941208
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,1023,0.02913280129432678
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,2,128,1,float16,float16,2047,0.03404799997806549
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,3,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,7,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,15,0.017608000338077544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,31,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,127,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,255,0.02232320010662079
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,511,0.02943359911441803
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,2047,0.04177919924259186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,4,128,1,float16,float16,1023,0.03409920036792755
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,1,0.019144000113010408
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,7,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,31,0.018115200102329254
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,63,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,127,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,255,0.02237440049648285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,511,0.033228799700737
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,1023,0.04095839858055115
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,96,8,128,1,float16,float16,2047,0.06425600051879883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,1,128,1,float16,float16,1,0.07393280267715455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,1,128,1,float16,float16,3,0.07500799894332885
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,1,128,1,float16,float16,7,0.07546240091323853
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,1,128,1,float16,float16,15,0.07500320076942443
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,1,128,1,float16,float16,31,0.07531520128250122
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,1,128,1,float16,float16,63,0.07541760206222534
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,2,128,1,float16,float16,1,0.11975680589675904
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,2,128,1,float16,float16,3,0.1210368037223816
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,2,128,1,float16,float16,7,0.11960320472717285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,2,128,1,float16,float16,15,0.12067359685897827
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,2,128,1,float16,float16,31,0.11939359903335571
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,2,128,1,float16,float16,63,0.17838079929351808
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,4,128,1,float16,float16,1,0.21027839183807373
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,4,128,1,float16,float16,3,0.21186559200286864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,4,128,1,float16,float16,7,0.21227519512176513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,4,128,1,float16,float16,15,0.2164223909378052
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,4,128,1,float16,float16,31,0.21381120681762694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,4,128,1,float16,float16,63,0.2827775955200195
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,8,128,1,float16,float16,1,0.386406397819519
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,8,128,1,float16,float16,3,0.38830080032348635
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,8,128,1,float16,float16,7,0.38937599658966066
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,8,128,1,float16,float16,15,0.3941888093948364
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,8,128,1,float16,float16,31,0.40503840446472167
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,96,8,128,1,float16,float16,63,0.4777984142303467
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,1,128,1,float16,float16,1,0.1339903950691223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,1,128,1,float16,float16,3,0.13675520420074463
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,1,128,1,float16,float16,7,0.14254080057144164
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,1,128,1,float16,float16,15,0.15032320022583007
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,1,128,1,float16,float16,31,0.17940479516983032
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,2,128,1,float16,float16,1,0.225382399559021
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,2,128,1,float16,float16,3,0.22804479598999022
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,2,128,1,float16,float16,7,0.2307584047317505
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,2,128,1,float16,float16,15,0.23342080116271974
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,2,128,1,float16,float16,31,0.24294400215148926
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,4,128,1,float16,float16,1,0.4039680004119873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,4,128,1,float16,float16,3,0.4051968097686768
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,4,128,1,float16,float16,7,0.40222558975219724
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,4,128,1,float16,float16,15,0.4127744197845459
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,4,128,1,float16,float16,31,0.4159999847412109
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,8,128,1,float16,float16,1,0.7674367904663086
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,8,128,1,float16,float16,3,0.7651328086853028
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,8,128,1,float16,float16,7,0.7667712211608887
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,8,128,1,float16,float16,15,0.7680511951446534
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,96,8,128,1,float16,float16,31,0.7723008155822754
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,1,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,3,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,7,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,15,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,31,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,127,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,255,0.02309119999408722
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,511,0.026368001103401185
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,1023,0.03031040132045746
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,1,128,1,float16,float16,2047,0.03543039858341217
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,15,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,31,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,255,0.021963199973106383
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,511,0.029542401432991028
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,1023,0.034201601147651674
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,2,128,1,float16,float16,2047,0.04259839951992035
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,1,0.018067200481891633
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,3,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,7,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,31,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,63,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,127,0.01812320053577423
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,255,0.02242559939622879
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,511,0.03399679958820343
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,1023,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,4,128,1,float16,float16,2047,0.06553599834442139
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,1,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,3,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,7,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,31,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,15,0.018785600364208222
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,63,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,127,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,255,0.023705600202083586
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,2047,0.22814719676971434
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,511,0.039628800749778745
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,96,8,128,1,float16,float16,1023,0.06323199868202209
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,1,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,3,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,7,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,15,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,31,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,63,0.019551999866962433
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,127,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,255,0.023039999604225158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,511,0.03624959886074066
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,1023,0.04413439929485321
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,2047,0.06855679750442505
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,1,128,1,float16,float16,4095,0.2337791919708252
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,1,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,3,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,7,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,15,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,31,0.019763199985027312
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,63,0.01966080069541931
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,127,0.019968000054359437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,255,0.024268800020217897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,1023,0.06492159962654113
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,511,0.04085760116577149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,2047,0.22425599098205568
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,1,0.026111999154090883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,2,128,1,float16,float16,4095,0.40320000648498533
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,3,0.026060798764228822
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,7,0.025856000185012818
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,15,0.025856000185012818
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,31,0.026111999154090883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,63,0.026313599944114686
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,255,0.03681280016899109
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,127,0.027136000990867614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,511,0.0662015974521637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,1023,0.2298367977142334
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,2047,0.42429437637329104
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,4,128,1,float16,float16,4095,0.7990784168243408
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,1,0.0394239991903305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,3,0.03906559944152832
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,15,0.03962239921092987
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,31,0.03978239893913269
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,7,0.0392192006111145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,63,0.03993600010871887
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,255,0.060057598352432254
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,127,0.04059999883174896
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,511,0.22676479816436768
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,1023,0.4184576034545898
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,2047,0.8070143699645996
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,1,0.01592320054769516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,3,0.015769599378108977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,64,8,128,1,float16,float16,4095,1.5493120193481444
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,7,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,15,0.015462400019168853
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,31,0.016076800227165223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,63,0.015718400478363037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,127,0.016327999532222748
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,255,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,511,0.021086399257183076
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,1023,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,4095,0.026265600323677064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,1,128,1,float16,float16,2047,0.025497600436210632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,1,0.016889600455760954
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,7,0.016276800632476808
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,3,0.016076800227165223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,15,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,31,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,63,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,127,0.016281600296497344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,255,0.020838400721549986
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,511,0.021452799439430237
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,2047,0.026668798923492432
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,4095,0.02727999985218048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,2,128,1,float16,float16,1023,0.025497600436210632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,1,0.01780959963798523
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,3,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,7,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,15,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,31,0.01765599995851517
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,63,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,255,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,127,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,511,0.02237440049648285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,1023,0.026470398902893065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,2047,0.027596798539161683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,4,128,1,float16,float16,4095,0.028467199206352232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,3,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,7,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,15,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,31,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,63,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,127,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,255,0.021555200219154358
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,511,0.02252800017595291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,1023,0.02754560112953186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,2047,0.028262400627136232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,64,8,128,1,float16,float16,4095,0.03292160034179688
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,1,0.016327999532222748
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,3,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,7,0.016383999586105348
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,31,0.01674239933490753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,15,0.01684480011463165
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,63,0.01679359972476959
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,127,0.017100800573825837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,255,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,511,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,1023,0.02754560112953186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,2047,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,1,128,1,float16,float16,4095,0.027647998929023743
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,1,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,3,0.017764799296855927
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,7,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,15,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,63,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,31,0.018119999766349794
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,255,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,511,0.02263039946556091
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,1023,0.027187201380729675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,2047,0.02815999984741211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,2,128,1,float16,float16,4095,0.029183998703956604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,1,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,15,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,7,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,31,0.017403200268745422
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,63,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,127,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,255,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,2047,0.028723201155662535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,511,0.022886399924755097
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,1023,0.027596798539161683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,4,128,1,float16,float16,4095,0.03409920036792755
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,1,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,3,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,7,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,127,0.01791680008172989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,255,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,511,0.023603199422359465
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,1023,0.02821120023727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,2047,0.033024001121521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,1,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,64,8,128,1,float16,float16,4095,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,3,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,7,0.018068799376487733
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,15,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,63,0.018427200615406036
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,31,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,127,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,255,0.022579200565814972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,1023,0.03527680039405823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,2047,0.043110400438308716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,4095,0.06476799845695495
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,1,128,1,float16,float16,511,0.03031040132045746
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,1,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,3,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,7,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,15,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,31,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,63,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,255,0.02292799949645996
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,511,0.034201601147651674
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,1023,0.04203520119190216
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,127,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,2047,0.06492159962654113
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,2,128,1,float16,float16,4095,0.22860319614410402
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,1,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,3,0.01878879964351654
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,7,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,15,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,31,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,127,0.019763199985027312
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,63,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,255,0.02401279956102371
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,511,0.04039680063724518
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,1023,0.06430720090866089
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,2047,0.22963199615478516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,4,128,1,float16,float16,4095,0.415231990814209
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,1,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,3,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,7,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,15,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,31,0.025446400046348572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,63,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,127,0.026675200462341307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,255,0.03553279936313629
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,511,0.06599680185317994
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,1023,0.2271712064743042
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,2047,0.4240384101867676
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,1,0.017969599366188048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,64,8,128,1,float16,float16,4095,0.7932415962219238
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,7,0.017763200402259826
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,3,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,15,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,63,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,127,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,255,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,511,0.023347200453281404
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,1023,0.027596798539161683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,4095,0.02969599962234497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,1,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,1,128,1,float16,float16,2047,0.028415998816490172
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,15,0.017763200402259826
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,31,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,63,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,511,0.02263039946556091
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,2047,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,4095,0.03348479866981506
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,2,128,1,float16,float16,1023,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,3,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,7,0.017508800327777862
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,31,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,63,0.018529599905014037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,127,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,511,0.024166400730609893
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,255,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,1023,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,2047,0.03307519853115082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,4,128,1,float16,float16,4095,0.04142079949378967
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,1,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,3,0.01775680035352707
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,7,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,15,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,63,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,127,0.017918400466442108
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,255,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,511,0.02821120023727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,1023,0.03281919956207276
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,2047,0.04116480052471161
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,64,8,128,1,float16,float16,4095,0.06174719929695129
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,1,0.02088959962129593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,7,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,15,0.020735999941825865
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,31,0.020582400262355804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,63,0.020787200331687926
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,3,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,127,0.021244800090789794
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,255,0.025497600436210632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,511,0.04264959990978241
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,1023,0.06721600294113159
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,1,0.026982399821281432
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,1,128,1,float16,float16,2047,0.22850561141967773
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,3,0.026774400472640993
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,15,0.026982399821281432
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,7,0.02677760124206543
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,31,0.027033600211143493
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,63,0.02744320034980774
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,127,0.02810400128364563
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,255,0.03700799942016601
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,511,0.06768640279769897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,1023,0.22522881031036376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,2,128,1,float16,float16,2047,0.41794562339782715
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,3,0.03978239893913269
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,1,0.03973119854927063
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,7,0.04033919870853424
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,15,0.040294399857521056
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,31,0.040336000919342044
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,63,0.04054720103740692
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,127,0.04152320027351379
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,255,0.06087679862976074
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,511,0.2309119939804077
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,1,0.06717439889907836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,2047,0.8102399826049804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,4,128,1,float16,float16,1023,0.42245121002197267
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,3,0.06666240096092224
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,7,0.06686720252037048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,15,0.06773120164871216
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,31,0.06748160123825073
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,63,0.06758400201797485
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,255,0.21176319122314452
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,127,0.06824960112571717
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,1,128,1,float16,float16,1,0.02964479923248291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,1023,0.8075743675231933
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,1,128,1,float16,float16,3,0.028974398970603943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,511,0.42152957916259765
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,64,8,128,1,float16,float16,2047,1.6419328689575194
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,1,128,1,float16,float16,7,0.030105599761009218
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,1,128,1,float16,float16,15,0.029542401432991028
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,1,128,1,float16,float16,31,0.02937920093536377
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,1,128,1,float16,float16,63,0.029798400402069092
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,1,128,1,float16,float16,127,0.03041279911994934
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,1,128,1,float16,float16,511,0.07244799733161926
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,2,128,1,float16,float16,1,0.04172160029411316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,1,128,1,float16,float16,255,0.03891200125217438
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,2,128,1,float16,float16,3,0.04121600091457367
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,2,128,1,float16,float16,7,0.0417279988527298
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,2,128,1,float16,float16,15,0.04188160002231598
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,2,128,1,float16,float16,31,0.04213759899139404
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,2,128,1,float16,float16,63,0.04239040017127991
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,2,128,1,float16,float16,127,0.043161600828170776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,2,128,1,float16,float16,255,0.06272000074386597
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,4,128,1,float16,float16,3,0.06799359917640686
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,2,128,1,float16,float16,511,0.23398239612579347
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,4,128,1,float16,float16,7,0.0686079978942871
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,4,128,1,float16,float16,1,0.06743040084838867
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,4,128,1,float16,float16,15,0.06896640062332153
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,4,128,1,float16,float16,31,0.06876159906387329
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,4,128,1,float16,float16,63,0.07004160284996033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,4,128,1,float16,float16,127,0.06999040246009827
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,4,128,1,float16,float16,255,0.22077438831329346
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,4,128,1,float16,float16,511,0.43048958778381347
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,8,128,1,float16,float16,7,0.11586560010910034
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,8,128,1,float16,float16,1,0.11407040357589722
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,8,128,1,float16,float16,3,0.11448320150375366
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,8,128,1,float16,float16,15,0.11612160205841064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,8,128,1,float16,float16,31,0.11786240339279175
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,8,128,1,float16,float16,63,0.11827199459075928
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,8,128,1,float16,float16,127,0.21590399742126465
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,8,128,1,float16,float16,255,0.40673279762268066
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,64,8,128,1,float16,float16,511,0.8021424293518067
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,1,128,1,float16,float16,1,0.045516800880432126
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,1,128,1,float16,float16,7,0.046079999208450316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,1,128,1,float16,float16,3,0.045772799849510194
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,1,128,1,float16,float16,15,0.045926401019096376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,1,128,1,float16,float16,31,0.046438398957252505
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,1,128,1,float16,float16,63,0.04684799909591675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,1,128,1,float16,float16,127,0.04792320132255554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,1,128,1,float16,float16,255,0.06999040246009827
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,2,128,1,float16,float16,1,0.0719871997833252
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,2,128,1,float16,float16,3,0.07172799706459046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,2,128,1,float16,float16,15,0.0726527988910675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,2,128,1,float16,float16,31,0.07264959812164307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,2,128,1,float16,float16,63,0.07331839799880982
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,2,128,1,float16,float16,7,0.07137280106544494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,2,128,1,float16,float16,127,0.07311360239982605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,2,128,1,float16,float16,255,0.23126399517059326
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,4,128,1,float16,float16,1,0.118886399269104
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,4,128,1,float16,float16,7,0.12088320255279542
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,4,128,1,float16,float16,3,0.11842559576034546
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,4,128,1,float16,float16,15,0.12088320255279542
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,4,128,1,float16,float16,63,0.1210368037223816
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,4,128,1,float16,float16,127,0.23173120021820068
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,4,128,1,float16,float16,31,0.11980799436569214
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,4,128,1,float16,float16,255,0.42485761642456055
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,8,128,1,float16,float16,1,0.20981760025024415
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,8,128,1,float16,float16,3,0.21083199977874756
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,8,128,1,float16,float16,15,0.21099519729614258
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,8,128,1,float16,float16,7,0.210534405708313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,8,128,1,float16,float16,63,0.23889920711517335
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,8,128,1,float16,float16,31,0.21135358810424804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,8,128,1,float16,float16,255,0.7990784168243408
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,64,8,128,1,float16,float16,127,0.41722879409790037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,1,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,15,0.017712000012397765
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,31,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,63,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,127,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,255,0.022169600427150726
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,511,0.023552000522613525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,1023,0.028620800375938414
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,2047,0.029542401432991028
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,1,128,1,float16,float16,4095,0.03415040075778961
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,1,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,7,0.01781439930200577
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,3,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,31,0.01791680008172989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,127,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,255,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,511,0.024422399699687958
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,1023,0.02887679934501648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,2047,0.03353599905967712
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,2,128,1,float16,float16,4095,0.0419840008020401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,1,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,7,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,63,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,31,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,255,0.02263039946556091
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,511,0.028262400627136232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,1023,0.03358719944953918
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,2047,0.04111360013484955
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,4,128,1,float16,float16,4095,0.0626688003540039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,1,0.01791200041770935
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,15,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,127,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,255,0.02222079932689667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,511,0.03307519853115082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,2047,0.06394559741020203
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,1023,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,1,128,1,float16,float16,1,0.07418559789657593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,1,128,1,float16,float16,3,0.07357439994812012
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,64,8,128,1,float16,float16,4095,0.22952959537506104
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,1,128,1,float16,float16,15,0.07418879866600037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,1,128,1,float16,float16,7,0.07459840178489685
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,1,128,1,float16,float16,31,0.07418879866600037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,1,128,1,float16,float16,63,0.07444480061531067
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,2,128,1,float16,float16,1,0.12001279592514039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,1,128,1,float16,float16,127,0.1013759970664978
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,2,128,1,float16,float16,3,0.12037119865417481
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,2,128,1,float16,float16,31,0.1212928056716919
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,2,128,1,float16,float16,15,0.12170239686965942
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,2,128,1,float16,float16,63,0.1297919988632202
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,2,128,1,float16,float16,7,0.12077759504318238
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,2,128,1,float16,float16,127,0.2530816078186035
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,4,128,1,float16,float16,1,0.2100224018096924
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,4,128,1,float16,float16,7,0.21089279651641846
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,4,128,1,float16,float16,15,0.2120192050933838
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,4,128,1,float16,float16,31,0.21662719249725343
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,4,128,1,float16,float16,63,0.26029601097106936
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,4,128,1,float16,float16,3,0.20930559635162355
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,4,128,1,float16,float16,127,0.44170241355895995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,8,128,1,float16,float16,1,0.38743040561676023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,8,128,1,float16,float16,3,0.38830080032348635
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,8,128,1,float16,float16,7,0.3910655975341797
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,8,128,1,float16,float16,15,0.3932159900665283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,8,128,1,float16,float16,31,0.3988991975784302
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,8,128,1,float16,float16,127,0.8109984397888184
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,64,8,128,1,float16,float16,63,0.4568064212799072
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,1,128,1,float16,float16,1,0.12226560115814208
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,1,128,1,float16,float16,3,0.12211040258407593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,1,128,1,float16,float16,7,0.12180479764938354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,1,128,1,float16,float16,15,0.12159839868545533
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,1,128,1,float16,float16,31,0.12344319820404052
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,1,128,1,float16,float16,63,0.21031999588012695
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,2,128,1,float16,float16,1,0.2130431890487671
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,2,128,1,float16,float16,7,0.21370880603790282
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,2,128,1,float16,float16,3,0.21140480041503906
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,2,128,1,float16,float16,15,0.2125823974609375
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,2,128,1,float16,float16,31,0.2191823959350586
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,2,128,1,float16,float16,63,0.3047935962677002
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,4,128,1,float16,float16,1,0.38988800048828126
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,4,128,1,float16,float16,3,0.39290881156921387
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,4,128,1,float16,float16,7,0.39239680767059326
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,4,128,1,float16,float16,15,0.39613440036773684
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,8,128,1,float16,float16,1,0.7360511779785156
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,4,128,1,float16,float16,63,0.5012479782104492
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,8,128,1,float16,float16,3,0.7371263980865479
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,4,128,1,float16,float16,31,0.4059648036956787
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,8,128,1,float16,float16,7,0.7442431926727295
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,8,128,1,float16,float16,15,0.7659008026123046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,8,128,1,float16,float16,31,0.7685632228851318
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,64,8,128,1,float16,float16,63,0.8913408279418945
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,1,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,7,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,3,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,15,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,31,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,63,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,127,0.018016000092029572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,255,0.02227199971675873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,511,0.02529279887676239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,1023,0.02964479923248291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,2047,0.034406399726867674
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,1,0.018892799317836762
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,3,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,1,128,1,float16,float16,4095,0.042905598878860474
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,31,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,255,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,127,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,511,0.02892799973487854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,4095,0.06287360191345215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,2047,0.0417279988527298
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,2,128,1,float16,float16,1023,0.03394559919834137
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,1,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,3,0.018016000092029572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,31,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,255,0.02242559939622879
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,511,0.03358719944953918
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,127,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,1023,0.04142079949378967
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,2047,0.06389759778976441
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,4,128,1,float16,float16,4095,0.23152639865875244
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,3,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,1,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,7,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,15,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,31,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,63,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,127,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,255,0.023603199422359465
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,1023,0.06292480230331421
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,2047,0.22855679988861083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,511,0.03983359932899475
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,64,8,128,1,float16,float16,4095,0.4164095878601074
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,1,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,3,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,7,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,15,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,31,0.01863519996404648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,127,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,255,0.022886399924755097
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,511,0.034867200255393985
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,63,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,2047,0.06502400040626526
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,1023,0.042803201079368594
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,4095,0.22809600830078125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,1,128,1,float16,float16,8191,0.4108287811279297
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,3,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,1,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,7,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,15,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,31,0.01960960030555725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,63,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,255,0.024422399699687958
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,1023,0.06594560146331788
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,127,0.019865599274635316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,511,0.04080640077590943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,2047,0.22318079471588134
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,4095,0.4016128063201904
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,2,128,1,float16,float16,8191,0.7615968227386475
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,3,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,1,0.026368001103401185
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,31,0.026153600215911864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,15,0.02590720057487488
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,63,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,7,0.026265600323677064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,127,0.027238398790359497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,511,0.06563839912414551
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,255,0.03614720106124878
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,1023,0.22625279426574707
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,2047,0.4262400150299072
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,4095,0.7976960182189942
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,4,128,1,float16,float16,8191,1.5645695686340333
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,1,0.03932160139083862
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,3,0.03906559944152832
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,7,0.039419201016426084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,15,0.0394239991903305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,31,0.03962399959564209
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,63,0.03977920114994049
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,127,0.04111360013484955
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,255,0.0606719970703125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,511,0.22369279861450195
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,2047,0.8061440467834473
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,1023,0.41635842323303224
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,1,0.015820799767971037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,4095,1.5781375885009765
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,3,0.015558399260044098
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,7,0.01592320054769516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,32,8,128,1,float16,float16,8191,3.124830436706543
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,31,0.015872000157833098
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,15,0.015564799308776855
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,127,0.016126400232315062
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,63,0.015718400478363037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,1023,0.025139200687408447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,255,0.019864000380039215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,511,0.02073120027780533
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,2047,0.025088000297546386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,4095,0.02600640058517456
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,1,0.016281600296497344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,1,128,1,float16,float16,8191,0.02739199995994568
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,31,0.01643519997596741
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,7,0.016025599837303162
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,3,0.016025599837303162
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,15,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,63,0.015974399447441102
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,127,0.01664000004529953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,255,0.020582400262355804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,1023,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,4095,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,8191,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,2047,0.02590720057487488
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,2,128,1,float16,float16,511,0.021296000480651854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,1,0.017100800573825837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,3,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,7,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,15,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,31,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,63,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,127,0.017609600722789765
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,255,0.021657599508762358
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,511,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,1023,0.026470398902893065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,2047,0.027639999985694885
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,4095,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,4,128,1,float16,float16,8191,0.033228799700737
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,1,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,3,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,15,0.01709119975566864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,31,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,63,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,7,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,127,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,255,0.0216511994600296
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,511,0.02237440049648285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,2047,0.027852800488471986
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,4095,0.03281919956207276
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,1023,0.027136000990867614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,1,0.01658879965543747
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,32,8,128,1,float16,float16,8191,0.04136959910392761
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,3,0.01648640036582947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,7,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,15,0.016383999586105348
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,31,0.01643519997596741
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,127,0.01669120043516159
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,63,0.016220800578594208
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,255,0.020428800582885744
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,1023,0.02585119903087616
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,511,0.021503999829292297
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,2047,0.02590720057487488
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,4095,0.026926401257514953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,1,128,1,float16,float16,8191,0.0289792001247406
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,3,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,1,0.017100800573825837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,7,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,63,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,31,0.017507199943065644
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,511,0.022780799865722658
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,1023,0.026675200462341307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,4095,0.029183998703956604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,2047,0.027902400493621825
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,1,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,3,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,7,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,2,128,1,float16,float16,8191,0.0336896002292633
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,31,0.017254400253295898
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,63,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,127,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,255,0.021803200244903564
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,1023,0.027590399980545043
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,2047,0.029388800263404846
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,511,0.022579200565814972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,8191,0.04111360013484955
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,4,128,1,float16,float16,4095,0.03281919956207276
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,1,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,3,0.017660799622535705
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,7,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,63,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,127,0.01780959963798523
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,511,0.023756800591945647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,255,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,2047,0.033024001121521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,4095,0.04208639860153198
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,1023,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,1,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,32,8,128,1,float16,float16,8191,0.061337602138519284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,3,0.01786399930715561
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,7,0.01812320053577423
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,15,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,31,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,127,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,63,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,255,0.02222079932689667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,2047,0.04249599874019623
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,1023,0.03409920036792755
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,511,0.029440000653266907
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,8191,0.23408639430999756
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,1,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,1,128,1,float16,float16,4095,0.06302719712257385
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,7,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,3,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,31,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,15,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,127,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,255,0.022784000635147093
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,511,0.033792001008987424
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,1023,0.0418287992477417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,2047,0.063128000497818
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,8191,0.40412158966064454
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,2,128,1,float16,float16,4095,0.2308095932006836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,3,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,7,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,1,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,31,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,63,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,15,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,255,0.02452480047941208
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,127,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,511,0.040243199467658995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,1023,0.06425600051879883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,2047,0.22855679988861083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,4095,0.41262078285217285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,3,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,7,0.02534399926662445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,1,0.029388800263404846
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,4,128,1,float16,float16,8191,0.7830016136169433
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,15,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,31,0.026265600323677064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,255,0.0359935998916626
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,127,0.02677760124206543
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,63,0.026009601354599
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,511,0.06492159962654113
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,1023,0.2243583917617798
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,4095,0.7920639991760254
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,2047,0.4217343807220459
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,32,8,128,1,float16,float16,8191,1.5664128303527831
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,1,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,3,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,15,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,31,0.017347200214862822
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,63,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,1023,0.027187201380729675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,511,0.022732800245285033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,2047,0.027852800488471986
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,8191,0.03363839983940124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,1,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,1,128,1,float16,float16,4095,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,3,0.017815999686717987
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,7,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,31,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,511,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,1023,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,4095,0.03343360126018524
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,8191,0.04152320027351379
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,2,128,1,float16,float16,2047,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,7,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,31,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,63,0.01786559969186783
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,511,0.02385919988155365
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,255,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,2047,0.033024001121521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,1023,0.027796798944473268
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,4095,0.04157440066337585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,1,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,4,128,1,float16,float16,8191,0.06137920022010803
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,3,0.017558400332927705
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,7,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,31,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,63,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,15,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,127,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,255,0.02222079932689667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,1023,0.03287039995193482
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,2047,0.04111360013484955
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,511,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,4095,0.06302719712257385
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,32,8,128,1,float16,float16,8191,0.22773759365081786
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,3,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,1,0.019812799990177155
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,7,0.020326399803161622
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,15,0.019865599274635316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,63,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,31,0.019916799664497376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,255,0.025088000297546386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,511,0.0419840008020401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,127,0.020633600652217865
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,2047,0.22497279644012452
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,1023,0.06635519862174988
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,1,0.02687999904155731
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,1,128,1,float16,float16,4095,0.40944638252258303
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,3,0.026419198513031004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,7,0.026624000072479247
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,15,0.026521599292755126
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,31,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,63,0.027187201380729675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,127,0.027955201268196107
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,255,0.03686400055885315
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,511,0.06635519862174988
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,1023,0.22307839393615722
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,2047,0.41446399688720703
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,2,128,1,float16,float16,4095,0.7714288234710693
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,3,0.04023520052433014
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,1,0.04028800129890442
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,15,0.040089601278305055
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,7,0.040140798687934874
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,31,0.03973119854927063
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,63,0.04177919924259186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,127,0.04152320027351379
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,255,0.06097919940948486
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,511,0.22732799053192138
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,1023,0.4188672065734863
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,2047,0.8104448318481445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,1,0.06819519996643067
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,7,0.06707199811935424
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,3,0.06625279784202576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,4,128,1,float16,float16,4095,1.572812843322754
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,15,0.06783999800682068
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,31,0.06773279905319214
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,127,0.06768640279769897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,63,0.06773759722709656
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,255,0.2077183961868286
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,511,0.41656317710876467
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,1023,0.8036352157592773
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,1,0.02815999984741211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,2047,1.617203140258789
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,32,8,128,1,float16,float16,4095,3.1236095428466797
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,3,0.028518399596214293
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,15,0.02815999984741211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,63,0.028774398565292358
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,31,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,7,0.028262400627136232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,127,0.029440000653266907
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,511,0.07301120162010193
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,1023,0.23004159927368165
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,1,128,1,float16,float16,255,0.03834879994392395
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,7,0.04167680144309997
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,1,0.04167680144309997
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,3,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,15,0.04162560105323791
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,63,0.042444801330566405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,31,0.0417279988527298
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,127,0.04321120083332062
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,255,0.061745601892471316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,511,0.22783999443054198
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,2,128,1,float16,float16,1023,0.4123136043548584
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,1,0.06748160123825073
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,3,0.0679423987865448
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,15,0.06876159906387329
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,31,0.06963199973106385
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,7,0.06855520009994506
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,127,0.06896640062332153
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,63,0.06901760101318359
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,255,0.21273601055145264
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,511,0.4227583885192871
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,7,0.11612160205841064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,1,0.11576319932937622
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,3,0.1159168004989624
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,4,128,1,float16,float16,1023,0.810086441040039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,15,0.11612160205841064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,63,0.1171455979347229
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,31,0.1156607985496521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,127,0.20730879306793212
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,255,0.3991039991378784
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,511,0.7904767990112305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,32,8,128,1,float16,float16,1023,1.5847935676574707
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,1,128,1,float16,float16,3,0.04505600035190582
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,1,128,1,float16,float16,1,0.04485119879245758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,1,128,1,float16,float16,7,0.045256000757217404
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,1,128,1,float16,float16,15,0.04490239918231964
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,1,128,1,float16,float16,31,0.04480000138282776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,1,128,1,float16,float16,63,0.045516800880432126
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,1,128,1,float16,float16,127,0.04628480076789856
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,1,128,1,float16,float16,511,0.23879680633544922
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,1,128,1,float16,float16,255,0.0662015974521637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,2,128,1,float16,float16,1,0.0719327986240387
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,2,128,1,float16,float16,3,0.07152640223503112
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,2,128,1,float16,float16,31,0.07270399928092956
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,2,128,1,float16,float16,7,0.07132160067558288
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,2,128,1,float16,float16,15,0.07132160067558288
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,2,128,1,float16,float16,63,0.073471999168396
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,2,128,1,float16,float16,127,0.07274720072746277
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,2,128,1,float16,float16,255,0.21872639656066895
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,4,128,1,float16,float16,1,0.11832319498062134
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,4,128,1,float16,float16,3,0.11857919692993164
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,2,128,1,float16,float16,511,0.42931199073791504
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,4,128,1,float16,float16,7,0.11791360378265381
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,4,128,1,float16,float16,63,0.12098560333251954
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,4,128,1,float16,float16,15,0.1186303973197937
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,4,128,1,float16,float16,31,0.12174880504608154
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,4,128,1,float16,float16,127,0.21775360107421876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,4,128,1,float16,float16,511,0.8048576354980469
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,8,128,1,float16,float16,3,0.208076810836792
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,4,128,1,float16,float16,255,0.4108799934387207
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,8,128,1,float16,float16,1,0.2096127986907959
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,8,128,1,float16,float16,7,0.2091520071029663
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,8,128,1,float16,float16,31,0.21370079517364501
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,8,128,1,float16,float16,15,0.21299200057983397
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,8,128,1,float16,float16,63,0.2348543882369995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,8,128,1,float16,float16,127,0.41175041198730467
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,8,128,1,float16,float16,255,0.7820288181304932
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,7,0.017815999686717987
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,32,8,128,1,float16,float16,511,1.5763968467712401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,15,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,3,0.017662400007247926
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,255,0.022064000368118286
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,511,0.02314240038394928
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,127,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,1023,0.02739199995994568
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,2047,0.0289792001247406
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,4095,0.03317759931087494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,1,128,1,float16,float16,8191,0.0419840008020401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,1,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,31,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,63,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,127,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,511,0.024268800020217897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,255,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,1023,0.02908160090446472
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,2047,0.0333296000957489
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,4095,0.042803201079368594
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,2,128,1,float16,float16,8191,0.061388802528381345
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,7,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,3,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,63,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,15,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,255,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,511,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,1023,0.033024001121521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,4095,0.06297600269317627
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,1,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,2047,0.04085760116577149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,3,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,4,128,1,float16,float16,8191,0.22917120456695556
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,31,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,63,0.018174399435520173
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,127,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,511,0.032767999172210696
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,255,0.02232320010662079
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,2047,0.06343680024147033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,4095,0.2283008098602295
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,1023,0.040755200386047366
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,32,8,128,1,float16,float16,8191,0.4145648002624512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,1,128,1,float16,float16,1,0.07223520278930665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,1,128,1,float16,float16,3,0.07280640006065368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,1,128,1,float16,float16,7,0.07239680290222168
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,1,128,1,float16,float16,31,0.07301120162010193
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,1,128,1,float16,float16,15,0.07280640006065368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,1,128,1,float16,float16,63,0.0729088008403778
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,1,128,1,float16,float16,127,0.0726527988910675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,1,128,1,float16,float16,255,0.2337791919708252
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,2,128,1,float16,float16,1,0.11914240121841431
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,2,128,1,float16,float16,3,0.12062720060348511
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,2,128,1,float16,float16,7,0.1198591947555542
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,2,128,1,float16,float16,15,0.12149759531021118
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,2,128,1,float16,float16,63,0.12180160284042359
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,2,128,1,float16,float16,31,0.12113920450210572
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,2,128,1,float16,float16,255,0.41446399688720703
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,2,128,1,float16,float16,127,0.22758400440216064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,4,128,1,float16,float16,1,0.20904960632324218
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,4,128,1,float16,float16,3,0.20889599323272706
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,4,128,1,float16,float16,7,0.21012480258941652
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,4,128,1,float16,float16,15,0.2134943962097168
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,4,128,1,float16,float16,31,0.2116096019744873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,4,128,1,float16,float16,127,0.4146687984466553
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,4,128,1,float16,float16,63,0.24038400650024414
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,8,128,1,float16,float16,3,0.3858896017074585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,8,128,1,float16,float16,1,0.3880431890487671
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,4,128,1,float16,float16,255,0.7990272045135498
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,8,128,1,float16,float16,7,0.3916287899017334
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,8,128,1,float16,float16,15,0.39393279552459715
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,8,128,1,float16,float16,31,0.39531519412994387
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,8,128,1,float16,float16,63,0.44431362152099607
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,1,128,1,float16,float16,1,0.12154879570007324
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,8,128,1,float16,float16,127,0.7827936172485351
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,32,8,128,1,float16,float16,255,1.5444992065429688
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,1,128,1,float16,float16,3,0.12093440294265748
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,1,128,1,float16,float16,7,0.12042239904403687
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,1,128,1,float16,float16,15,0.12288000583648681
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,1,128,1,float16,float16,31,0.12170239686965942
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,1,128,1,float16,float16,63,0.12390400171279907
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,2,128,1,float16,float16,1,0.20879359245300294
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,2,128,1,float16,float16,3,0.2116607904434204
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,1,128,1,float16,float16,127,0.26296319961547854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,2,128,1,float16,float16,15,0.21155838966369628
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,2,128,1,float16,float16,7,0.20930559635162355
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,2,128,1,float16,float16,31,0.21519360542297364
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,2,128,1,float16,float16,63,0.2617343902587891
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,2,128,1,float16,float16,127,0.4339712142944336
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,4,128,1,float16,float16,1,0.38702080249786375
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,4,128,1,float16,float16,3,0.3890176057815552
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,4,128,1,float16,float16,15,0.39290881156921387
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,4,128,1,float16,float16,7,0.389683198928833
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,4,128,1,float16,float16,63,0.4659167766571045
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,4,128,1,float16,float16,31,0.3964416027069092
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,4,128,1,float16,float16,127,0.8069120407104492
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,8,128,1,float16,float16,1,0.7357952117919921
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,8,128,1,float16,float16,3,0.7345119953155518
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,8,128,1,float16,float16,7,0.733900785446167
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,8,128,1,float16,float16,15,0.7423488140106201
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,8,128,1,float16,float16,31,0.7621056079864502
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,1,0.01780959963798523
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,8,128,1,float16,float16,63,0.8559616088867188
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,3,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,32,8,128,1,float16,float16,127,1.571225643157959
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,7,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,15,0.01781439930200577
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,31,0.017611199617385866
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,255,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,1023,0.028518399596214293
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,2047,0.03363839983940124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,4095,0.04223519861698151
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,511,0.024422399699687958
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,1,0.024063999950885772
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,1,128,1,float16,float16,8191,0.06164479851722717
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,3,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,31,0.01796319931745529
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,63,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,15,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,511,0.029337599873542786
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,255,0.023193599283695222
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,1023,0.033583998680114746
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,2047,0.0419840008020401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,4095,0.06277120113372803
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,2,128,1,float16,float16,8191,0.22338559627532958
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,1,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,7,0.01791519969701767
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,31,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,127,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,255,0.022571200132369997
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,511,0.03287039995193482
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,2047,0.06364160180091857
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,1023,0.04080640077590943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,4095,0.2318336009979248
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,4,128,1,float16,float16,8191,0.4146687984466553
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,3,0.0186271995306015
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,1,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,7,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,15,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,31,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,63,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,127,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,255,0.02396160066127777
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,511,0.03978239893913269
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,1023,0.061849600076675414
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,2047,0.22814719676971434
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,4095,0.41006078720092776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,32,8,128,1,float16,float16,8191,0.7802879810333252
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,1,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,7,0.018681600689888
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,15,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,3,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,31,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,63,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,127,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,255,0.023193599283695222
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,1023,0.042393600940704344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,511,0.03415040075778961
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,2047,0.06487039923667907
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,4095,0.22574079036712646
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,8191,0.40801281929016114
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,1,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,1,128,1,float16,float16,16383,0.7792128086090088
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,3,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,7,0.019200000166893005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,15,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,31,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,63,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,127,0.019916799664497376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,511,0.04136959910392761
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,255,0.024422399699687958
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,1023,0.0634335994720459
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,2047,0.22215359210968016
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,4095,0.4003759860992432
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,8191,0.7599103927612305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,1,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,7,0.025753599405288697
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,2,128,1,float16,float16,16383,1.5284223556518555
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,15,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,3,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,31,0.026214399933815004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,63,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,127,0.027136000990867614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,255,0.0359935998916626
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,511,0.06630399823188782
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,1023,0.24381439685821532
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,2047,0.4236800193786621
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,4095,0.7944191932678223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,8191,1.5842816352844238
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,4,128,1,float16,float16,16383,3.146444892883301
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,1,0.039680001139640805
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,3,0.03931199908256531
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,7,0.0394239991903305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,15,0.03911679983139038
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,31,0.03962399959564209
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,63,0.040188801288604734
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,255,0.060313600301742556
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,511,0.2234879970550537
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,127,0.040652799606323245
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,1023,0.41420798301696776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,2047,0.8039327621459961
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,4095,1.5469568252563477
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,1,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,8191,3.128780746459961
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,7,0.015769599378108977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,15,0.015667200088500977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,3,0.015564799308776855
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,16,8,128,1,float16,float16,16383,6.271897506713867
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,31,0.015360000729560851
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,63,0.015718400478363037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,127,0.016070400178432465
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,255,0.019812799990177155
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,511,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,1023,0.0247296005487442
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,2047,0.025190401077270507
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,4095,0.026009601354599
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,8191,0.026822400093078614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,1,128,1,float16,float16,16383,0.030003198981285097
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,1,0.016947199404239655
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,7,0.01684480011463165
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,3,0.016128000617027283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,15,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,31,0.01648640036582947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,127,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,63,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,255,0.020377600193023683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,511,0.02114560008049011
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,2047,0.025702399015426636
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,1023,0.025036799907684325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,4095,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,8191,0.028518399596214293
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,1,0.016998399794101716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,2,128,1,float16,float16,16383,0.03338240087032318
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,3,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,15,0.01770720034837723
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,127,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,63,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,255,0.021503999829292297
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,2047,0.027596798539161683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,1023,0.026214399933815004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,511,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,4095,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,8191,0.03317759931087494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,4,128,1,float16,float16,16383,0.04203520119190216
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,7,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,3,0.017151999473571777
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,15,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,63,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,31,0.017203199863433837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,127,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,255,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,511,0.02252800017595291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,2047,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,1023,0.026726400852203368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,4095,0.03261440098285675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,8191,0.04116480052471161
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,16,8,128,1,float16,float16,16383,0.06169599890708923
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,3,0.01648000031709671
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,1,0.016128000617027283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,7,0.016281600296497344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,15,0.01643519997596741
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,31,0.016072000563144683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,63,0.016281600296497344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,127,0.01674239933490753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,511,0.02160640060901642
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,1023,0.02534399926662445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,255,0.020531199872493744
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,2047,0.025804799795150758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,4095,0.027750399708747864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,8191,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,1,128,1,float16,float16,16383,0.03409920036792755
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,1,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,3,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,7,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,15,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,63,0.01786559969186783
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,127,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,255,0.021904000639915468
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,511,0.022422400116920472
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,1023,0.02682879865169525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,2047,0.027340799570083618
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,16383,0.04223999977111816
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,4095,0.02903040051460266
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,2,128,1,float16,float16,8191,0.0336896002292633
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,1,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,7,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,31,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,127,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,63,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,511,0.022780799865722658
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,255,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,1023,0.027076798677444457
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,2047,0.028672000765800475
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,4095,0.03297280073165894
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,8191,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,4,128,1,float16,float16,16383,0.06231039762496948
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,1,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,3,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,15,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,31,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,63,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,127,0.01780959963798523
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,1023,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,511,0.023756800591945647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,2047,0.03261440098285675
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,8191,0.06149119734764099
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,16383,0.22850561141967773
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,16,8,128,1,float16,float16,4095,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,3,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,7,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,31,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,127,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,63,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,255,0.02237440049648285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,511,0.029798400402069092
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,1023,0.033843201398849485
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,2047,0.04264959990978241
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,4095,0.06236159801483154
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,8191,0.2238384008407593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,1,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,1,128,1,float16,float16,16383,0.4082176208496094
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,3,0.01812160015106201
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,7,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,15,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,31,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,127,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,511,0.03348000049591064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,255,0.024371199309825897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,2047,0.06394879817962647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,1023,0.04167680144309997
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,4095,0.22277119159698486
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,8191,0.40417118072509767
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,2,128,1,float16,float16,16383,0.7667200088500976
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,3,0.018988800048828126
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,1,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,7,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,15,0.018935999274253844
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,63,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,127,0.019865599274635316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,31,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,511,0.04039680063724518
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,255,0.023756800591945647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,1023,0.06399999856948853
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,2047,0.22819840908050537
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,4095,0.4123119831085205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,8191,0.77915358543396
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,1,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,7,0.025548800826072693
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,3,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,4,128,1,float16,float16,16383,1.5459327697753906
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,15,0.02539519965648651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,31,0.025548800826072693
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,63,0.026163199543952943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,127,0.026926401257514953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,255,0.03563520014286041
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,511,0.06492159962654113
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,1023,0.2242527961730957
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,2047,0.4591616153717041
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,4095,0.7925727844238282
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,8191,1.5649279594421386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,16,8,128,1,float16,float16,16383,3.1258575439453127
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,7,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,3,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,63,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,255,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,511,0.023601600527763368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,2047,0.028563201427459717
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,1023,0.02682879865169525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,4095,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,8191,0.03317759931087494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,1,128,1,float16,float16,16383,0.04264959990978241
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,1,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,7,0.017815999686717987
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,15,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,31,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,63,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,127,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,1023,0.027750399708747864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,511,0.02247679978609085
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,2047,0.02903040051460266
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,4095,0.03317759931087494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,8191,0.04121600091457367
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,2,128,1,float16,float16,16383,0.0611840009689331
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,1,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,7,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,3,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,31,0.01780959963798523
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,63,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,127,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,511,0.023603199422359465
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,1023,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,4095,0.04147199988365173
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,2047,0.03297280073165894
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,8191,0.061440002918243405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,4,128,1,float16,float16,16383,0.23152639865875244
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,1,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,3,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,7,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,15,0.017764799296855927
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,63,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,31,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,127,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,511,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,255,0.022067199647426605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,1023,0.03287039995193482
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,2047,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,4095,0.06322879791259765
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,8191,0.22783999443054198
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,16,8,128,1,float16,float16,16383,0.41308159828186036
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,1,0.02017119973897934
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,3,0.019814400374889372
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,7,0.019916799664497376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,15,0.019913600385189058
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,31,0.020172800123691558
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,63,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,127,0.02099040001630783
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,255,0.025139200687408447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,511,0.0419840008020401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,1023,0.06492000222206115
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,2047,0.22302720546722413
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,4095,0.4074495792388916
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,1,128,1,float16,float16,8191,0.7705088138580323
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,1,0.02656959891319275
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,3,0.026617598533630372
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,7,0.026521599292755126
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,15,0.02677760124206543
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,31,0.026624000072479247
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,63,0.026982399821281432
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,127,0.02805759906768799
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,511,0.06686720252037048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,1023,0.22133760452270507
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,255,0.03665919899940491
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,2047,0.4128767967224121
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,4095,0.7674880027770996
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,2,128,1,float16,float16,8191,1.5321599960327148
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,3,0.03983359932899475
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,1,0.04157440066337585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,7,0.03993119895458221
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,15,0.040345600247383116
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,63,0.040703999996185306
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,31,0.03983359932899475
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,127,0.04126720130443573
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,255,0.06077439785003662
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,1023,0.4168191909790039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,511,0.24238080978393556
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,2047,0.8027647972106934
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,4095,1.5443967819213866
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,3,0.06558719873428345
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,1,0.06599680185317994
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,4,128,1,float16,float16,8191,3.1714303970336912
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,7,0.0671231985092163
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,15,0.0673792004585266
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,31,0.0673792004585266
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,63,0.06804479956626892
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,127,0.06824960112571717
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,255,0.20546560287475585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,511,0.41349120140075685
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,1023,0.8745471954345703
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,2047,1.6441343307495118
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,1,0.028364801406860353
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,4095,3.1234031677246095
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,7,0.027955201268196107
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,3,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,31,0.02821120023727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,15,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,16,8,128,1,float16,float16,8191,6.521497344970703
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,63,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,127,0.02969599962234497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,255,0.03840000033378601
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,511,0.06947839856147767
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,1023,0.225433611869812
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,1,128,1,float16,float16,2047,0.41635842323303224
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,3,0.04183039963245392
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,1,0.04121600091457367
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,7,0.04157440066337585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,15,0.04162560105323791
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,31,0.041771200299263
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,63,0.0419840008020401
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,127,0.04270080029964447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,255,0.06256639957427979
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,511,0.22318079471588134
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,1023,0.4093952178955078
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,1,0.06830080151557923
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,2,128,1,float16,float16,2047,0.78504958152771
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,3,0.06768159866333008
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,7,0.06824960112571717
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,15,0.06773759722709656
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,31,0.06886399984359741
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,127,0.06912000179290771
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,63,0.06917120218276977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,255,0.20884480476379394
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,1023,0.8044032096862793
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,511,0.42690558433532716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,1,0.11586560010910034
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,4,128,1,float16,float16,2047,1.6151552200317383
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,3,0.11509759426116943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,7,0.11539679765701294
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,15,0.1174496054649353
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,31,0.11755520105361938
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,127,0.22548480033874513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,63,0.11776000261306763
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,511,0.7844351768493653
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,255,0.3940864086151123
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,1023,1.5434752464294434
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,3,0.0447488009929657
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,1,0.044894400238990786
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,16,8,128,1,float16,float16,2047,3.2233470916748046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,7,0.044593599438667295
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,15,0.04433920085430145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,63,0.045465600490570066
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,31,0.045414400100708005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,255,0.06543359756469727
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,127,0.046079999208450316
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,511,0.23152639865875244
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,1,128,1,float16,float16,1023,0.45168638229370117
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,1,0.07127040028572082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,7,0.07234560251235962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,15,0.07188479900360108
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,3,0.0709119975566864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,31,0.07244799733161926
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,63,0.07219200134277344
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,127,0.07229120135307313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,255,0.2283008098602295
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,511,0.41502718925476073
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,2,128,1,float16,float16,1023,0.7878143787384033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,7,0.11944960355758667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,3,0.11883519887924195
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,1,0.11786240339279175
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,15,0.11883200407028198
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,31,0.12154239416122437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,63,0.121343994140625
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,127,0.21063680648803712
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,255,0.4137983798980713
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,511,0.7924223899841308
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,3,0.2070528030395508
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,4,128,1,float16,float16,1023,1.5541248321533203
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,7,0.2110975980758667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,1,0.20817921161651612
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,15,0.2100224018096924
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,31,0.21345279216766358
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,63,0.23403520584106446
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,127,0.3952640056610107
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,255,0.7725567817687988
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,1,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,511,1.529702377319336
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,3,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,16,8,128,1,float16,float16,1023,3.0577152252197264
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,7,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,15,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,127,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,63,0.018993599712848662
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,255,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,511,0.022579200565814972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,2047,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,4095,0.03358719944953918
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,1023,0.02815999984741211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,8191,0.0417279988527298
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,1,128,1,float16,float16,16383,0.062105602025985716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,3,0.017815999686717987
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,1,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,7,0.0173567995429039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,31,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,63,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,127,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,255,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,511,0.02385919988155365
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,1023,0.027852800488471986
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,2047,0.03256320059299469
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,8191,0.061952000856399535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,4095,0.04106239974498749
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,2,128,1,float16,float16,16383,0.22210559844970704
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,7,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,15,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,127,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,63,0.017654399573802947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,255,0.022067199647426605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,1023,0.03297280073165894
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,511,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,2047,0.040908798575401306
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,4095,0.061900800466537474
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,8191,0.2283008098602295
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,1,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,4,128,1,float16,float16,16383,0.4145103931427002
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,3,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,7,0.01806560009717941
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,63,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,127,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,511,0.03281919956207276
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,255,0.02222079932689667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,1023,0.040703999996185306
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,2047,0.06353920102119445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,4095,0.22824959754943847
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,8191,0.41251840591430666
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,16,8,128,1,float16,float16,16383,0.7861248016357422
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,1,128,1,float16,float16,1,0.07208960056304932
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,1,128,1,float16,float16,3,0.07193599939346314
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,1,128,1,float16,float16,7,0.07208960056304932
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,1,128,1,float16,float16,15,0.07178239822387696
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,1,128,1,float16,float16,31,0.07280640006065368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,1,128,1,float16,float16,63,0.07296000123023987
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,1,128,1,float16,float16,127,0.07249919772148132
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,1,128,1,float16,float16,255,0.2209791898727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,2,128,1,float16,float16,1,0.11919360160827637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,2,128,1,float16,float16,3,0.12001279592514039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,1,128,1,float16,float16,511,0.4290048122406006
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,2,128,1,float16,float16,7,0.11980799436569214
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,2,128,1,float16,float16,15,0.12006399631500245
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,2,128,1,float16,float16,31,0.11970560550689698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,2,128,1,float16,float16,63,0.1211359977722168
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,2,128,1,float16,float16,127,0.21452798843383789
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,2,128,1,float16,float16,255,0.4008448123931885
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,4,128,1,float16,float16,1,0.20884480476379394
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,4,128,1,float16,float16,3,0.20997118949890137
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,4,128,1,float16,float16,7,0.21068799495697021
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,2,128,1,float16,float16,511,0.783616018295288
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,4,128,1,float16,float16,15,0.21094400882720948
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,4,128,1,float16,float16,31,0.21324799060821534
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,4,128,1,float16,float16,63,0.23249919414520265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,4,128,1,float16,float16,127,0.40176639556884763
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,4,128,1,float16,float16,255,0.7829504013061523
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,8,128,1,float16,float16,1,0.38712320327758787
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,4,128,1,float16,float16,511,1.562777614593506
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,8,128,1,float16,float16,3,0.3870719909667969
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,8,128,1,float16,float16,7,0.38865759372711184
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,8,128,1,float16,float16,15,0.3935744047164917
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,8,128,1,float16,float16,31,0.39705600738525393
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,8,128,1,float16,float16,63,0.4377600193023682
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,8,128,1,float16,float16,127,0.7707647800445556
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,8,128,1,float16,float16,255,1.5553536415100098
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,1,128,1,float16,float16,1,0.12015999555587768
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,1,128,1,float16,float16,3,0.11806720495223999
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,1,128,1,float16,float16,7,0.11909120082855225
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,16,8,128,1,float16,float16,511,3.016499137878418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,1,128,1,float16,float16,15,0.12083200216293336
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,1,128,1,float16,float16,31,0.12252160310745239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,1,128,1,float16,float16,63,0.1212928056716919
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,1,128,1,float16,float16,127,0.2501120090484619
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,1,128,1,float16,float16,255,0.4192207813262939
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,2,128,1,float16,float16,1,0.20817921161651612
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,2,128,1,float16,float16,3,0.2105855941772461
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,2,128,1,float16,float16,15,0.21099519729614258
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,2,128,1,float16,float16,7,0.20971519947052003
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,2,128,1,float16,float16,31,0.2145792007446289
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,2,128,1,float16,float16,63,0.2461103916168213
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,2,128,1,float16,float16,127,0.4096992015838623
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,2,128,1,float16,float16,255,0.7787007808685302
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,4,128,1,float16,float16,1,0.38881280422210696
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,4,128,1,float16,float16,7,0.3908512115478516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,4,128,1,float16,float16,3,0.38650879859924314
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,4,128,1,float16,float16,31,0.3960319995880127
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,4,128,1,float16,float16,15,0.38978559970855714
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,4,128,1,float16,float16,63,0.4493824005126953
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,4,128,1,float16,float16,127,0.7828479766845703
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,8,128,1,float16,float16,1,0.7348671913146972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,8,128,1,float16,float16,7,0.735487985610962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,4,128,1,float16,float16,255,1.5741439819335938
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,8,128,1,float16,float16,3,0.7376895904541015
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,8,128,1,float16,float16,15,0.7361023902893067
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,8,128,1,float16,float16,31,0.7581183910369873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,8,128,1,float16,float16,63,0.851097583770752
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,8,128,1,float16,float16,127,1.5368703842163085
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,7,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,3,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,1,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,16,8,128,1,float16,float16,255,3.1308271408081056
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,63,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,127,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,1023,0.028415998816490172
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,255,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,511,0.024063999950885772
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,2047,0.033276799321174624
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,4095,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,8191,0.061849600076675414
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,1,128,1,float16,float16,16383,0.22323200702667237
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,1,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,7,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,63,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,255,0.022067199647426605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,511,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,1023,0.033228799700737
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,2047,0.040755200386047366
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,4095,0.061027199029922485
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,8191,0.2224128007888794
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,2,128,1,float16,float16,16383,0.4023280143737793
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,7,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,3,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,15,0.019916799664497376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,31,0.017713600397109987
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,127,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,255,0.02252800017595291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,511,0.033228799700737
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,1023,0.04095360040664673
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,2047,0.06399999856948853
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,4095,0.23070719242095947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,8191,0.41322717666625974
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,4,128,1,float16,float16,16383,0.7869952201843262
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,1,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,3,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,7,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,15,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,63,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,31,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,127,0.019148799777030944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,255,0.023552000522613525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,1023,0.06312959790229797
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,511,0.039987200498580934
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,2047,0.22717440128326416
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,4095,0.4142591953277588
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,8191,0.7835648059844971
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,16,8,128,1,float16,float16,16383,1.5771648406982421
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,1,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,7,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,3,0.018739199638366698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,15,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,31,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,63,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,127,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,511,0.03404799997806549
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,1023,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,255,0.023039999604225158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,2047,0.06348800063133239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,4095,0.22420480251312255
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,8191,0.4077568054199219
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,16383,0.7731200218200683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,1,0.019251200556755065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,1,128,1,float16,float16,32767,1.4946816444396973
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,7,0.018943999707698823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,15,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,3,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,31,0.01935359984636307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,63,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,127,0.020270399749279022
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,255,0.024422399699687958
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,1023,0.06343680024147033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,511,0.040345600247383116
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,2047,0.22128639221191407
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,4095,0.4015615940093994
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,8191,0.7581696033477783
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,16383,1.5313407897949218
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,2,128,1,float16,float16,32767,3.0549503326416017
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,1,0.026060798764228822
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,3,0.02595199942588806
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,7,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,15,0.025804799795150758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,63,0.026419198513031004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,31,0.02590720057487488
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,127,0.027033600211143493
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,255,0.03624959886074066
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,511,0.06528000235557556
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,1023,0.22748160362243652
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,2047,0.42193918228149413
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,4095,0.7964159965515136
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,8191,1.5415807723999024
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,1,0.015718400478363037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,16383,3.1740415573120115
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,3,0.015718400478363037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,7,0.015609599649906158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,15,0.015820799767971037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,31,0.015564799308776855
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,8,4,128,1,float16,float16,32767,6.282700729370117
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,63,0.015769599378108977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,127,0.016128000617027283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,511,0.020582400262355804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,255,0.02099200040102005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,2047,0.024934400618076325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,1023,0.024412800371646882
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,4095,0.02534399926662445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,8191,0.026675200462341307
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,32767,0.03573760092258453
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,1,128,1,float16,float16,16383,0.029798400402069092
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,3,0.016383999586105348
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,1,0.016076800227165223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,15,0.016432000696659087
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,7,0.016068799793720244
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,31,0.016128000617027283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,63,0.017100800573825837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,127,0.017254400253295898
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,511,0.021807999908924104
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,255,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,1023,0.0247296005487442
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,2047,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,4095,0.026257601380348206
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,8191,0.02810879945755005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,32767,0.043161600828170776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,2,128,1,float16,float16,16383,0.03266560137271881
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,1,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,15,0.017553600668907165
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,7,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,3,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,31,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,63,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,255,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,127,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,511,0.02232320010662079
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,2047,0.02743679881095886
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,1023,0.026111999154090883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,8191,0.033024001121521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,4095,0.02821120023727417
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,16383,0.04183039963245392
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,8,4,128,1,float16,float16,32767,0.06287360191345215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,3,0.01643519997596741
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,1,0.01592320054769516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,7,0.016383999586105348
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,15,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,31,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,63,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,127,0.016383999586105348
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,255,0.020684799551963805
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,511,0.021299199759960176
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,1023,0.024926400184631346
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,2047,0.02590720057487488
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,4095,0.026571199297904968
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,8191,0.028467199206352232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,16383,0.03358719944953918
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,1,128,1,float16,float16,32767,0.04249599874019623
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,3,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,7,0.017151999473571777
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,63,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,511,0.02314240038394928
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,1023,0.026726400852203368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,2047,0.02815839946269989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,4095,0.028513601422309874
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,8191,0.03338240087032318
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,16383,0.04198080003261566
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,1,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,2,128,1,float16,float16,32767,0.062412798404693604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,3,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,31,0.017449599504470826
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,63,0.017609600722789765
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,127,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,1023,0.027596798539161683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,2047,0.028262400627136232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,511,0.022732800245285033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,4095,0.033024001121521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,8191,0.04080640077590943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,16383,0.06169599890708923
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,8,4,128,1,float16,float16,32767,0.23787519931793213
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,3,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,7,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,31,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,15,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,63,0.01796319931745529
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,127,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,255,0.022169600427150726
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,511,0.02892799973487854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,1023,0.03353599905967712
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,2047,0.04157440066337585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,4095,0.06410239934921265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,8191,0.22266879081726074
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,16383,0.4068863868713379
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,1,128,1,float16,float16,32767,0.9001472473144532
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,1,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,3,0.018273599445819855
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,7,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,15,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,31,0.018785600364208222
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,127,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,255,0.02252800017595291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,511,0.033228799700737
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,1023,0.04213759899139404
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,2047,0.06389759778976441
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,4095,0.22328319549560546
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,8191,0.4013055801391602
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,16383,0.7651328086853028
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,7,0.019446399807929993
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,3,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,1,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,15,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,2,128,1,float16,float16,32767,1.5276032447814942
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,31,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,63,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,511,0.040089601278305055
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,127,0.01945600062608719
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,1023,0.0647599995136261
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,255,0.02391040027141571
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,2047,0.22804479598999022
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,4095,0.45194239616394044
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,8191,0.7798704147338867
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,16383,1.5651840209960937
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,3,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,15,0.017348800599575043
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,31,0.017151999473571777
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,8,4,128,1,float16,float16,32767,3.0227455139160155
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,255,0.022163200378417968
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,1023,0.02677760124206543
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,511,0.02232320010662079
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,2047,0.02744320034980774
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,8191,0.033792001008987424
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,4095,0.028355199098587035
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,16383,0.04270080029964447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,1,128,1,float16,float16,32767,0.062105602025985716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,1,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,3,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,7,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,15,0.017560000717639922
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,31,0.017764799296855927
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,63,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,255,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,511,0.02282720059156418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,1023,0.027750399708747864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,2047,0.02805440127849579
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,8191,0.04152320027351379
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,4095,0.032767999172210696
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,16383,0.06169599890708923
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,2,128,1,float16,float16,32767,0.22169599533081055
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,7,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,63,0.018371200561523436
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,511,0.02462719976902008
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,255,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,2047,0.03292160034179688
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,1023,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,8191,0.061952000856399535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,4095,0.041468799114227295
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,16383,0.23014400005340577
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,8,4,128,1,float16,float16,32767,0.41395039558410646
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,1,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,3,0.020121599733829498
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,7,0.020632000267505647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,15,0.019864000380039215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,31,0.0201664000749588
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,63,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,127,0.020633600652217865
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,255,0.025497600436210632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,511,0.04162560105323791
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,1023,0.06553599834442139
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,2047,0.22215681076049804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,8191,0.7683072090148926
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,4095,0.4056575775146484
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,1,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,1,128,1,float16,float16,16383,1.5340543746948243
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,3,0.026470398902893065
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,7,0.026822400093078614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,15,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,31,0.02677760124206543
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,127,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,63,0.027084800601005554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,255,0.03650560081005096
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,511,0.06625279784202576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,1023,0.22031359672546386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,2047,0.4120063781738281
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,4095,0.7683072090148926
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,8191,1.5316479682922364
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,1,0.040038400888442995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,3,0.040243199467658995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,7,0.040191999077796935
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,15,0.040243199467658995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,2,128,1,float16,float16,16383,3.058790397644043
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,31,0.040243199467658995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,63,0.040191999077796935
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,127,0.04167680144309997
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,255,0.06072319746017456
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,511,0.22374238967895507
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,1023,0.415283203125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,2047,0.8728575706481934
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,4095,1.578547191619873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,8191,3.143116760253906
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,1,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,3,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,7,0.029388800263404846
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,15,0.028364801406860353
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,8,4,128,1,float16,float16,16383,6.257611083984375
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,31,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,63,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,127,0.030054399371147157
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,255,0.03824639916419983
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,511,0.06881279945373535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,1023,0.22363839149475098
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,2047,0.41507840156555176
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,1,128,1,float16,float16,4095,0.8600576400756836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,1,0.04116480052471161
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,3,0.04131839871406555
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,7,0.04177919924259186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,15,0.04248639941215515
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,31,0.04157440066337585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,127,0.043007999658584595
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,63,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,255,0.06215680241584778
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,511,0.22118399143218995
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,2047,0.7829504013061523
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,1023,0.4069375991821289
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,2,128,1,float16,float16,4095,1.5368703842163085
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,3,0.0679423987865448
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,1,0.0681984007358551
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,7,0.06799359917640686
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,15,0.06717439889907836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,31,0.06896640062332153
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,63,0.07024480104446411
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,255,0.22697598934173585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,127,0.06835200190544129
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,511,0.4148223876953125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,1023,0.800870418548584
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,1,0.04490239918231964
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,2047,1.6353727340698243
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,3,0.04480000138282776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,7,0.04485119879245758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,15,0.04433920085430145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,8,4,128,1,float16,float16,4095,3.238142395019531
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,31,0.04485119879245758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,63,0.04510720074176788
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,127,0.04618239998817444
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,511,0.22814719676971434
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,255,0.06584320068359376
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,1023,0.41436161994934084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,3,0.07152479887008667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,1,128,1,float16,float16,2047,0.8707584381103516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,1,0.07137280106544494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,7,0.07162879705429077
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,15,0.07188479900360108
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,31,0.07275519967079162
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,127,0.07239519953727722
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,255,0.20736000537872315
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,63,0.07311360239982605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,1023,0.7832575798034668
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,511,0.41011199951171873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,1,0.12031999826431275
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,2,128,1,float16,float16,2047,1.599283218383789
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,3,0.11960320472717285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,7,0.11929600238800049
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,31,0.12097760438919067
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,15,0.12026879787445069
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,127,0.22753279209136962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,63,0.12047359943389893
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,255,0.3970047950744629
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,511,0.7861199855804444
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,1023,1.5469039916992187
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,1,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,8,4,128,1,float16,float16,2047,3.1684608459472656
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,31,0.01796479970216751
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,63,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,255,0.022067199647426605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,1023,0.027750399708747864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,511,0.023039999604225158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,2047,0.028364801406860353
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,4095,0.03307519853115082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,8191,0.04183039963245392
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,16383,0.06220800280570984
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,1,128,1,float16,float16,32767,0.22225921154022216
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,3,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,127,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,255,0.021807999908924104
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,511,0.02359839975833893
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,1023,0.028313601016998292
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,2047,0.03281919956207276
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,4095,0.04147199988365173
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,8191,0.0609279990196228
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,32767,0.3999727964401245
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,2,128,1,float16,float16,16383,0.2229759931564331
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,1,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,3,0.017763200402259826
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,7,0.017611199617385866
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,15,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,31,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,127,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,255,0.021913599967956544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,1023,0.033024001121521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,511,0.027852800488471986
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,2047,0.04085760116577149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,4095,0.06174719929695129
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,8191,0.22973439693450928
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,16383,0.4137983798980713
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,8,4,128,1,float16,float16,32767,0.9216511726379395
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,3,0.07193440198898315
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,1,0.07183359861373902
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,7,0.07316480278968811
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,15,0.07192639708518982
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,31,0.07121919989585876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,63,0.07301120162010193
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,127,0.07352319955825806
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,511,0.4167679786682129
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,255,0.22507519721984864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,1,0.11914240121841431
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,3,0.12001279592514039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,7,0.11939840316772461
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,15,0.11990560293197632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,1,128,1,float16,float16,1023,0.7979519844055176
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,31,0.12149759531021118
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,63,0.12083200216293336
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,127,0.20843520164489746
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,255,0.39121921062469484
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,511,0.773632001876831
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,2,128,1,float16,float16,1023,1.5157759666442872
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,1,0.20899679660797119
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,3,0.21027839183807373
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,7,0.2116096019744873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,15,0.21048319339752197
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,31,0.21186559200286864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,63,0.22783999443054198
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,255,0.7723519802093506
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,127,0.43438081741333007
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,511,1.529190444946289
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,1,128,1,float16,float16,1,0.11924480199813843
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,8,4,128,1,float16,float16,1023,3.0608896255493163
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,1,128,1,float16,float16,3,0.11918560266494752
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,1,128,1,float16,float16,7,0.11857919692993164
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,1,128,1,float16,float16,15,0.11965440511703491
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,1,128,1,float16,float16,31,0.12247040271759033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,1,128,1,float16,float16,63,0.12108800411224366
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,1,128,1,float16,float16,127,0.2362368106842041
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,1,128,1,float16,float16,255,0.4059135913848877
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,1,128,1,float16,float16,511,0.8613887786865234
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,2,128,1,float16,float16,1,0.20869119167327882
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,2,128,1,float16,float16,3,0.20746240615844727
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,2,128,1,float16,float16,7,0.21063680648803712
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,2,128,1,float16,float16,15,0.21191680431365967
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,2,128,1,float16,float16,31,0.2136064052581787
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,2,128,1,float16,float16,63,0.23905279636383056
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,2,128,1,float16,float16,127,0.39813120365142823
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,2,128,1,float16,float16,255,0.7616511821746826
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,2,128,1,float16,float16,511,1.5022496223449706
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,4,128,1,float16,float16,1,0.3875839948654175
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,4,128,1,float16,float16,3,0.3881472110748291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,4,128,1,float16,float16,7,0.3933183908462524
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,4,128,1,float16,float16,15,0.39347200393676757
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,4,128,1,float16,float16,31,0.3974143981933594
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,4,128,1,float16,float16,63,0.4438015937805176
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,4,128,1,float16,float16,127,0.7687104225158692
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,4,128,1,float16,float16,255,1.5225775718688965
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,1,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,3,0.017399999499320983
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,8,4,128,1,float16,float16,511,3.120639991760254
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,15,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,7,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,63,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,127,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,255,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,511,0.023756800591945647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,1023,0.028364801406860353
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,2047,0.032767999172210696
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,8191,0.0611840009689331
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,16383,0.22282240390777588
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,4095,0.04264479875564575
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,1,128,1,float16,float16,32767,0.4071375846862793
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,1,0.017608000338077544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,7,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,15,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,63,0.01786080002784729
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,31,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,127,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,255,0.02227199971675873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,511,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,1023,0.032764801383018495
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,2047,0.04111360013484955
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,4095,0.06307839751243591
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,16383,0.4016096115112305
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,8191,0.2229759931564331
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,1,0.018119999766349794
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,2,128,1,float16,float16,32767,0.8948224067687989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,7,0.01781280040740967
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,31,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,127,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,255,0.02263039946556091
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,511,0.03317759931087494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,1023,0.04085760116577149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,2047,0.06394879817962647
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,4095,0.2308095932006836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,8191,0.4119552135467529
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,16383,0.7858672142028809
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,1,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,8,4,128,1,float16,float16,32767,1.548851203918457
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,3,0.018681600689888
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,7,0.018329599499702455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,15,0.018380799889564516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,31,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,63,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,127,0.019046400487422944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,511,0.034764799475669864
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,255,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,1023,0.04336639940738678
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,2047,0.06650879979133606
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,4095,0.2243583917617798
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,8191,0.4076543807983398
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,16383,0.7755775928497315
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,32767,1.5391231536865235
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,1,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,1,128,1,float16,float16,65535,3.0361087799072264
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,3,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,7,0.01884160041809082
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,15,0.01940480023622513
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,31,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,63,0.019763199985027312
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,127,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,255,0.024268800020217897
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,511,0.04080640077590943
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,2047,0.22108159065246583
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,1023,0.06379520297050476
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,4095,0.3999743938446045
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,8191,0.7611904144287109
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,16383,1.5281663894653321
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,32767,3.060838317871094
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,1,0.015513600409030914
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,7,0.01592320054769516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,3,0.015667200088500977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,15,0.01525759994983673
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,31,0.015667200088500977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,4,2,128,1,float16,float16,65535,6.079590225219727
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,63,0.015615999698638916
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,127,0.016072000563144683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,255,0.01966080069541931
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,511,0.021247999370098115
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,2047,0.024879999458789825
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,1023,0.025284799933433532
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,4095,0.025600001215934753
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,16383,0.030771198868751525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,8191,0.02748799920082092
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,32767,0.03609600067138672
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,1,0.01648640036582947
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,1,128,1,float16,float16,65535,0.04756479859352112
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,7,0.015820799767971037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,3,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,15,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,31,0.016128000617027283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,63,0.016076800227165223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,127,0.017100800573825837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,511,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,255,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,1023,0.025753599405288697
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,2047,0.026111999154090883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,4095,0.026726400852203368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,8191,0.028723201155662535
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,16383,0.03317759931087494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,32767,0.04213759899139404
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,4,2,128,1,float16,float16,65535,0.06225919723510742
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,1,0.01653759926557541
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,3,0.01592320054769516
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,7,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,15,0.016947199404239655
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,31,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,63,0.016332800686359405
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,127,0.01669120043516159
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,255,0.020531199872493744
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,511,0.02160640060901642
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,1023,0.025497600436210632
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,2047,0.026060798764228822
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,4095,0.027084800601005554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,8191,0.029337599873542786
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,32767,0.04433920085430145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,16383,0.03537920117378235
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,1,128,1,float16,float16,65535,0.06430400013923646
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,1,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,3,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,7,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,31,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,63,0.017603200674057008
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,511,0.02309119999408722
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,1023,0.026822400093078614
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,2047,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,4095,0.02892799973487854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,16383,0.042905598878860474
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,32767,0.06287360191345215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,8191,0.034560000896453856
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,4,2,128,1,float16,float16,65535,0.22225759029388428
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,1,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,7,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,15,0.01796640008687973
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,31,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,127,0.01812160015106201
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,63,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,255,0.022623999416828154
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,511,0.029235199093818665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,1023,0.03415040075778961
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,2047,0.04208639860153198
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,4095,0.06389759778976441
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,8191,0.22415359020233155
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,16383,0.4093440055847168
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,32767,0.9036800384521484
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,1,128,1,float16,float16,65535,1.7901056289672852
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,3,0.018585599958896637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,1,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,7,0.01802240014076233
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,15,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,31,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,63,0.01817599982023239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,255,0.023552000522613525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,127,0.01879040002822876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,511,0.035123199224472046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,1023,0.04203520119190216
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,2047,0.06466559767723083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,4095,0.2234879970550537
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,8191,0.40309758186340333
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,16383,0.7685599803924561
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,1,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,32767,1.4844415664672852
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,7,0.017254400253295898
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,3,0.017560000717639922
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,4,2,128,1,float16,float16,65535,3.0941183090209963
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,31,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,63,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,127,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,2047,0.028307199478149414
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,511,0.02314240038394928
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,1023,0.027084800601005554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,255,0.022118400037288665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,4095,0.029286399483680725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,8191,0.03430399894714355
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,16383,0.04510720074176788
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,32767,0.0652288019657135
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,1,0.01730560064315796
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,1,128,1,float16,float16,65535,0.22558720111846925
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,7,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,15,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,31,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,127,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,1023,0.02892799973487854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,2047,0.0289792001247406
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,511,0.02303680032491684
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,4095,0.033792001008987424
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,8191,0.04270080029964447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,16383,0.06359040141105651
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,32767,0.22282240390777588
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,4,2,128,1,float16,float16,65535,0.4007423877716064
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,1,0.01950719952583313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,3,0.01971199959516525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,7,0.020019200444221497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,15,0.020121599733829498
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,31,0.019968000054359437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,63,0.019968000054359437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,127,0.020838400721549986
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,255,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,1023,0.06502400040626526
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,511,0.04157440066337585
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,2047,0.22107200622558593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,4095,0.4133887767791748
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,8191,0.7688127994537354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,16383,1.5514623641967773
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,1,0.026419198513031004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,3,0.02662239968776703
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,7,0.02738879919052124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,15,0.026572799682617186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,1,128,1,float16,float16,32767,2.959206390380859
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,31,0.026624000072479247
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,63,0.027340799570083618
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,127,0.02754560112953186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,255,0.03660799860954285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,511,0.0667136013507843
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,1023,0.22087678909301758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,2047,0.41175041198730467
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,4095,0.7649792194366455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,8191,1.6135679244995118
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,1,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,16383,3.0593536376953123
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,7,0.028467199206352232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,3,0.02800000011920929
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,15,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,31,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,63,0.028620800375938414
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,4,2,128,1,float16,float16,32767,5.756415939331054
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,255,0.037939199805259706
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,127,0.02969599962234497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,511,0.06895679831504822
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,1023,0.2233344078063965
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,2047,0.41390080451965333
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,4095,0.7792640209197998
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,1,128,1,float16,float16,8191,1.547878360748291
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,1,0.04183039963245392
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,3,0.04095999896526337
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,15,0.04177919924259186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,7,0.0415695995092392
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,31,0.04188160002231598
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,63,0.04177919924259186
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,127,0.04270080029964447
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,255,0.06250879764556885
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,511,0.22056961059570312
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,1023,0.4058112144470215
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,2047,0.7830016136169433
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,4095,1.5305727958679198
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,1,0.04403199851512909
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,3,0.0449535995721817
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,7,0.04480000138282776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,4,2,128,1,float16,float16,8191,3.0638591766357424
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,15,0.0447488009929657
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,31,0.0449535995721817
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,63,0.045516800880432126
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,127,0.046024000644683837
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,255,0.06466559767723083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,1023,0.41333761215209963
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,511,0.22568960189819337
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,2047,0.8630784034729004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,1,0.07197920083999634
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,1,128,1,float16,float16,4095,1.5513600349426269
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,3,0.07137280106544494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,7,0.07218559980392455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,15,0.07132160067558288
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,31,0.07239360213279725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,63,0.07311360239982605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,127,0.07214080095291138
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,511,0.4079599857330322
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,255,0.22456319332122804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,1023,0.7809535980224609
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,3,0.017407999932765962
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,2047,1.5913472175598145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,1,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,4,2,128,1,float16,float16,4095,3.0274560928344725
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,7,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,31,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,255,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,511,0.023244799673557283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,1023,0.028467199206352232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,2047,0.029235199093818665
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,4095,0.03348479866981506
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,8191,0.042854401469230655
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,16383,0.0652288019657135
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,32767,0.22568960189819337
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,1,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,1,128,1,float16,float16,65535,0.41031680107116697
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,15,0.01930239945650101
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,7,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,31,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,63,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,255,0.022163200378417968
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,511,0.02447360008955002
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,1023,0.028508800268173217
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,2047,0.03374080061912536
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,4095,0.0421887993812561
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,8191,0.062105602025985716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,16383,0.22302720546722413
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,32767,0.4021183967590332
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,4,2,128,1,float16,float16,65535,0.9543168067932128
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,7,0.0719871997833252
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,1,0.07208960056304932
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,3,0.07198560237884521
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,15,0.07193599939346314
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,31,0.07157760262489318
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,63,0.07326719760894776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,127,0.07260000109672546
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,255,0.22973439693450928
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,511,0.41113600730895994
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,1023,0.7909376144409179
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,3,0.11842559576034546
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,1,0.11898880004882813
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,1,128,1,float16,float16,2047,1.6034303665161134
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,15,0.11929600238800049
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,7,0.11898880004882813
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,31,0.11929600238800049
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,63,0.12170239686965942
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,127,0.2240000009536743
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,255,0.38645761013031005
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,511,0.7671807765960693
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,1023,1.5055359840393066
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,1,0.118886399269104
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,3,0.1183743953704834
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,4,2,128,1,float16,float16,2047,3.050444793701172
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,15,0.12011519670486451
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,7,0.12011519670486451
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,63,0.12190719842910766
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,31,0.12093440294265748
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,255,0.3984895944595337
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,127,0.22947840690612792
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,1,0.20842719078063965
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,511,0.8540672302246094
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,3,0.2100224018096924
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,1,128,1,float16,float16,1023,1.552793598175049
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,7,0.2110975980758667
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,15,0.21068799495697021
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,31,0.21232640743255615
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,127,0.39244000911712645
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,63,0.23439040184020996
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,255,0.7510528087615966
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,511,1.5183792114257812
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,1,0.017560000717639922
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,4,2,128,1,float16,float16,1023,3.0953983306884765
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,15,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,63,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,31,0.017612800002098083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,255,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,511,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,1023,0.02903040051460266
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,2047,0.03363839983940124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,4095,0.04212960004806519
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,8191,0.0626688003540039
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,16383,0.22568960189819337
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,32767,0.41057281494140624
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,1,128,1,float16,float16,65535,0.9477631568908691
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,3,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,7,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,1,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,15,0.018529599905014037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,31,0.018278400599956512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,127,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,1023,0.033843201398849485
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,511,0.028825598955154418
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,255,0.021862399578094483
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,63,0.017715199291706084
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,2047,0.04193280041217804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,4095,0.06302719712257385
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,8191,0.2233344078063965
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,16383,0.4027391910552979
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,1,0.01955839991569519
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,32767,0.8960000038146972
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,3,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,7,0.018636800348758698
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,15,0.018688000738620758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,4,2,128,1,float16,float16,65535,1.742131233215332
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,31,0.018432000279426576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,63,0.018534399569034576
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,255,0.022937600314617158
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,127,0.018995200097560883
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,511,0.035067200660705566
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,2047,0.06553599834442139
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,1023,0.04321280121803284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,4095,0.24437758922576905
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,8191,0.4074495792388916
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,16383,0.7789567947387696
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,32767,1.4981632232666016
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,1,0.015820799767971037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,3,0.015820799767971037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,65535,2.968627166748047
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,15,0.015513600409030914
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,7,0.015718400478363037
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,64,1,2,1,128,1,float16,float16,131071,6.186700820922852
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,31,0.015974399447441102
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,63,0.015667200088500977
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,127,0.016998399794101716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,511,0.02160640060901642
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,255,0.020275199413299562
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,1023,0.02529279887676239
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,2047,0.025088000297546386
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,4095,0.02595840096473694
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,8191,0.027596798539161683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,16383,0.03128319978713989
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,32767,0.03640320003032684
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,65535,0.04863840043544769
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1,1,2,1,128,1,float16,float16,131071,0.06686720252037048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,1,0.016230399906635284
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,7,0.016484799981117248
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,3,0.015974399447441102
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,15,0.016179199516773223
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,31,0.016128000617027283
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,63,0.01658879965543747
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,127,0.01679359972476959
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,255,0.020428800582885744
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,511,0.021964800357818604
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,1023,0.026316800713539125
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,4095,0.02682879865169525
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,8191,0.0289792001247406
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,2047,0.025804799795150758
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,16383,0.03660799860954285
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,65535,0.06389759778976441
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,32767,0.045158401131629944
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2,1,2,1,128,1,float16,float16,131071,0.22466559410095216
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,1,0.01812160015106201
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,3,0.017761600017547608
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,15,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,31,0.018115200102329254
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,7,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,63,0.01812479943037033
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,255,0.02227199971675873
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,127,0.018483200669288637
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,511,0.029849600791931153
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,1023,0.03425279855728149
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,2047,0.04259839951992035
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,4095,0.06369280219078063
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,16383,0.4123136043548584
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,8191,0.22456319332122804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,32767,0.9057279586791992
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,65535,1.7908735275268555
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,3,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,1,0.017459200322628023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,7,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,32,1,2,1,128,1,float16,float16,131071,3.547443389892578
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,31,0.01791999936103821
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,63,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,127,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,255,0.022015999257564544
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,511,0.023449599742889404
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,1023,0.02784479856491089
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,2047,0.028415998816490172
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,4095,0.02887679934501648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,8191,0.03430399894714355
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,16383,0.046228799223899844
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,32767,0.06574079990386963
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,65535,0.2259455919265747
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,4,1,2,1,128,1,float16,float16,131071,0.41031680107116697
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,1,0.020735999941825865
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,3,0.020019200444221497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,7,0.019763199985027312
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,31,0.02022400051355362
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,15,0.020070399343967437
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,63,0.020019200444221497
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,127,0.020479999482631683
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,255,0.025036799907684325
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,511,0.04183039963245392
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,1023,0.06487039923667907
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,2047,0.2214911937713623
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,4095,0.4050432205200195
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,8191,0.7686655998229981
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,1,0.02815999984741211
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,16383,1.5354880332946776
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,7,0.027904000878334046
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,3,0.027699199318885804
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,32767,3.079217529296875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,15,0.02800639867782593
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,128,1,2,1,128,1,float16,float16,65535,5.831679916381836
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,31,0.028569599986076354
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,63,0.028512001037597656
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,255,0.03927040100097656
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,511,0.06840320229530335
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,127,0.02928000092506409
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,1023,0.2229248046875
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,2047,0.41333761215209963
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,4095,0.8582143783569336
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,8191,1.5369664192199708
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,1,0.04433920085430145
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,3,0.04490239918231964
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,7,0.04459519982337952
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,256,1,2,1,128,1,float16,float16,16383,3.0523391723632813
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,15,0.0444927990436554
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,31,0.04439040124416351
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,63,0.04526079893112182
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,127,0.0462336003780365
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,255,0.06446080207824707
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,511,0.22476799488067628
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,1023,0.4126719951629639
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,2047,0.8612863540649414
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,1,0.017764799296855927
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,4095,1.540403175354004
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,512,1,2,1,128,1,float16,float16,8191,3.0473215103149416
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,3,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,7,0.017558400332927705
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,15,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,31,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,63,0.01822720021009445
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,127,0.01807360053062439
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,511,0.02380799949169159
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,255,0.02176000028848648
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,1023,0.028467199206352232
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,4095,0.03343360126018524
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,2047,0.029329600930213928
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,8191,0.043315199017524716
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,16383,0.06650879979133606
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,32767,0.22712318897247313
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,65535,0.4104191780090332
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,1,0.07121919989585876
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,3,0.07080960273742676
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,7,0.07183039784431458
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,8,1,2,1,128,1,float16,float16,131071,0.7727615833282471
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,15,0.07137280106544494
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,31,0.07280640006065368
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,63,0.07352319955825806
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,127,0.07193599939346314
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,255,0.20935680866241455
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,511,0.40960001945495605
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,1023,0.7866367816925048
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,2047,1.5706111907958984
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,1,0.12047359943389893
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,1024,1,2,1,128,1,float16,float16,4095,3.047064018249512
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,3,0.11949280500411988
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,7,0.11796480417251587
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,15,0.11924480199813843
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,31,0.1212928056716919
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,127,0.2259455919265747
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,63,0.12149759531021118
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,255,0.43274240493774413
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,511,0.7712768077850342
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,1023,1.560934352874756
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,1,0.017664000391960144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,3,0.017561599612236023
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,2048,1,2,1,128,1,float16,float16,2047,3.1515647888183596
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,15,0.017510400712490083
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,7,0.017766399681568144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,31,0.017868800461292265
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,63,0.01797119975090027
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,127,0.017817600071430205
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,255,0.02181120067834854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,511,0.024780799448490144
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,1023,0.029388800263404846
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,2047,0.03363839983940124
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,4095,0.042956799268722534
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,8191,0.06399840116500854
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,16383,0.2268160104751587
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,32767,0.4109312057495117
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,65535,0.9481727600097656
SGLang,0.5.5.post3,NVIDIA L40S,generation_attention,flash_attention,16,1,2,1,128,1,float16,float16,131071,1.8729471206665038
