TBNilles/LEANN
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial commit

Browse Source
This commit is contained in:
yichuan520030910320
2025-06-30 09:05:05 +00:00
commit 46f6cc100b
1231 changed files with 278432 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

89
research/micro/data/transformer-batching-microbenchmarks.csv Normal file
View File

@@ -0,0 +1,89 @@
n,d,seqlen,bs,latency,h,flop,io,intensity,throughput,series
3,256,256,2048,0.009623501679245285,768,618475290624,167.48502132816208,3692720015.912285,64267177503366.266,dense
3,256,256,1024,0.004853848615384615,768,309237645312,166.15392854317415,1861151572.059558,63709783682138.234,dense
3,256,256,512,0.0024687246971962615,768,154618822656,163.57953256539062,945221081.3366361,62631051097597.516,dense
3,256,256,256,0.0012845360838052097,768,77309411328,157.64931990085577,490388486.1451936,60184694149645.54,dense
3,256,256,128,0.0006901147179878049,768,38654705664,147.57393422494675,261934506.70684624,56012000116019.945,dense
3,256,256,64,0.0003363830693015702,768,19327352832,153.1328437752606,126212981.84970059,57456378146882.51,dense
3,256,256,32,0.00018671159748991485,768,9663676416,141.10249365427362,68486928.65540518,51757237075334.75,dense
3,256,256,16,0.00012353640857142858,768,4831838208,111.40488993609125,43371868.24359184,39112665358133.98,dense
3,256,256,8,9.774760007849294e-05,768,2415919104,76.43260800265766,31608487.09906635,24715891766754.14,dense
3,256,256,4,6.672271167474822e-05,768,1207959552,64.82614227498455,18633833.660438772,18104173551704.773,dense
3,256,256,2,4.9758770289855074e-05,768,603979776,55.317122669351576,10918495.880745342,12138157202874.861,dense
3,256,1,2048,9.785507940251571e-05,768,2415919104,76.34865809334705,31643242.518371396,24688745017132.86,dense
3,256,1,1024,6.692813470149253e-05,768,1207959552,64.62717090938949,18691202.70936228,18048606275785.867,dense
3,256,1,512,4.9680950036205655e-05,768,603979776,55.40377142534654,10901419.893658841,12157170415618.898,dense
3,256,1,256,4.2781118741058655e-05,768,301989888,45.95672244805227,6571179.83862661,7058952568020.829,dense
3,256,1,128,5.0662328255350016e-05,768,150994944,31.046026784880404,4863583.512513602,2980418571348.519,dense
3,256,1,64,4.475009253945481e-05,768,75497472,30.75426042497223,2454862.219307235,1687090857598.4766,dense
3,256,1,32,4.51682671454219e-05,768,37748736,28.29313765537115,1334201.1218340008,835735758435.5786,dense
3,256,1,16,5.03585186661834e-05,768,18874368,24.401035466223117,773506.846712577,374799904761.1871,dense
3,256,1,8,5.023459565217391e-05,768,9437184,23.972005435021096,393675.19858030166,187862246674.45105,dense
3,256,1,4,5.053219391083726e-05,768,4718592,23.58765586356967,200044.97383259286,93377936614.54384,dense
3,256,1,2,4.4607398995335484e-05,768,2359296,26.58285456464288,88752.54515134107,52890239133.797226,dense
12,256,256,2048,0.14480779847058822,3072,9895604649984,44.620009282941716,221775046868.20184,68336130750540.26,dense
12,256,256,1024,0.07254347629166667,3072,4947802324992,44.664248332585096,110777691547.58836,68204648824643.82,dense
12,256,256,512,0.036310761444444443,3072,2473901162496,44.876147984203506,55127306456.13385,68131349056975.164,dense
12,256,256,256,0.01821551906896552,3072,1236950581248,45.24607467289738,27338295977.947884,67906414116709.98,dense
12,256,256,128,0.009229417903030302,3072,618475290624,45.67217092440895,13541622351.335684,67011299859001.46,dense
12,256,256,64,0.004754550595394737,3072,309237645312,46.31372736116993,6677019167.566916,65040352207320.695,dense
12,256,256,32,0.002405752659340659,3072,154618822656,49.68826015254682,3111777755.5766335,64270456921525.82,dense
12,256,256,16,0.0012287219045005488,3072,77309411328,56.323579604557374,1372594069.3184311,62918558743709.18,dense
12,256,256,8,0.0006206816149425287,3072,38654705664,70.95456179103653,544781120.315271,62277832520589.78,dense
12,256,256,4,0.0003875502697142857,3072,19327352832,81.16954743236613,238110885.71245712,49870569942445.75,dense
12,256,256,2,0.00027502018627941914,3072,9663676416,91.50537035282076,105607751.53129694,35138062215483.168,dense
12,256,1,2048,0.0006202853873290136,3072,38654705664,70.99988634205897,544433345.6784943,62317614526515.766,dense
12,256,1,1024,0.00038721467732724153,3072,19327352832,81.2398957010995,237904697.74985722,49913791918755.53,dense
12,256,1,512,0.000274364799,3072,9663676416,91.72395326121995,105356082.81599998,35221998052308.45,dense
12,256,1,256,0.00012488918589482266,3072,4831838208,176.31707535146046,27404255.647778228,38689003962834.75,dense
12,256,1,128,8.976711102514506e-05,3072,2415919104,227.78088507574267,10606329.425740216,26913187652026.21,dense
12,256,1,64,8.715176287471176e-05,3072,1207959552,225.59268282689945,5354604.31102229,13860414432884.701,dense
12,256,1,32,8.523013435114503e-05,3072,603979776,226.06539514085782,2671703.8033338524,7086458100741.991,dense
12,256,1,16,7.901561645904116e-05,3072,301989888,241.35704882952732,1251216.3595988373,3821901309300.556,dense
12,256,1,8,7.827949114210329e-05,3072,150994944,242.37091635608994,622991.1833900034,1928920867994.581,dense
12,256,1,4,7.779445951035782e-05,3072,75497472,243.25022783249054,310369.58391664835,970473636235.5986,dense
12,256,1,2,7.758845406626506e-05,3072,37748736,243.57933441822672,154975.11761480253,486525172518.07056,dense
3,256,256,2048,0.00507974918466899,768,206158430208,475.59810852303485,433471930.42508715,40584371927298.98,qk_init
3,256,256,1024,0.0025616677649325623,768,103079215104,471.5519977009198,218595649.27424532,40239103803811.82,qk_init
3,256,256,512,0.0013029336670480549,768,51539607552,463.55374128015677,111183672.92143403,39556585922573.38,qk_init
3,256,256,256,0.0006738189029345373,768,25769803776,448.1766342333362,57499213.050413854,38244406121244.69,qk_init
3,256,256,128,0.000358254672959467,768,12884901888,421.47375986100144,30571065.425874516,35965760841472.125,qk_init
3,256,256,64,0.0002007051105022831,768,6442450944,376.1611839930762,17126836.096194826,32099087700742.5,qk_init
3,256,256,32,0.00012189697230142565,768,3221225472,309.6773881032524,10401874.969721656,26425803784810.87,qk_init
3,256,256,16,8.453561698040722e-05,768,1610612736,223.2711923587723,7213705.982328083,19052475081281.902,qk_init
3,256,256,8,6.407660705009276e-05,768,805306368,147.2797083750448,5467870.468274581,12567868448003.822,qk_init
3,256,256,4,5.036328747284576e-05,768,402653184,93.69110391262903,4297667.197682838,7994974200544.344,qk_init
3,256,256,2,4.5488761135057476e-05,768,201326592,51.865470527877875,3881707.616858238,4425853485045.578,qk_init
12,256,256,2048,0.020202365999999996,3072,824633720832,478.3437947812648,1723935231.9999998,40818670488001.266,qk_init
12,256,256,1024,0.010124155888157895,3072,412316860416,477.2583770318811,863927969.1228071,40726048173387.19,qk_init
12,256,256,512,0.005085633937062937,3072,206158430208,475.04777848703077,433974095.9627039,40537410430893.29,qk_init
12,256,256,256,0.0025654916853281853,3072,103079215104,470.84913933193053,218921957.14800516,40179126556324.74,qk_init
12,256,256,128,0.0013045765704467354,3072,51539607552,462.9699702434292,111323867.34478809,39506770794105.96,qk_init
12,256,256,64,0.0006742801519939804,3072,25769803776,447.87005387442576,57538572.970153,38218244597284.33,qk_init
12,256,256,32,0.00035831976790671853,3072,12884901888,421.3971919051604,30576620.194706645,35959227042573.69,qk_init
12,256,256,16,0.0002005369068918302,3072,6442450944,376.4766953382971,17112482.721436176,32126011335534.68,qk_init
12,256,256,8,0.00012179187250509165,3072,3221225472,309.94462293386505,10392906.453767821,26448607823689.82,qk_init
12,256,256,4,8.452507263643351e-05,3072,1610612736,223.2990450204527,7212806.198308992,19054851841745.297,qk_init
12,256,256,2,6.412381767545489e-05,3072,805306368,147.17127491946468,5471899.108305484,12558615459794.32,qk_init
3,256,256,2048,0.0016183739398395718,768,805306368,811597824.0,0.9922480620155039,1265467.7325087283,qk_ar
3,256,256,1024,0.0008322699728813558,768,402653184,405798912.0,0.9922480620155039,1230369.9921491416,qk_ar
3,256,256,512,0.00043886859397590365,768,201326592,202899456.0,0.9922480620155039,1166636.2255762408,qk_ar
3,256,256,256,0.00024185948322147648,768,100663296,101449728.0,0.9922480620155039,1058465.8355760013,qk_ar
3,256,256,128,0.00014308985100166944,768,50331648,50724864.0,0.9922480620155039,894542.82818777,qk_ar
3,256,256,64,9.382939365815932e-05,768,25165824,25362432.0,0.9922480620155039,682089.028872613,qk_ar
3,256,256,32,6.856070612244899e-05,768,12582912,12681216.0,0.9922480620155039,466739.6503012703,qk_ar
3,256,256,16,5.452260553129549e-05,768,6291456,6340608.0,0.9922480620155039,293456.26174846216,qk_ar
3,256,256,8,4.608557533261417e-05,768,3145728,3170304.0,0.9922480620155039,173590.1080166944,qk_ar
3,256,256,4,4.386146957766642e-05,768,1572864,1585152.0,0.9922480620155039,91196.21477609445,qk_ar
3,256,256,2,4.330941094420601e-05,768,786432,792576.0,0.9922480620155039,46179.33969539622,qk_ar
12,256,256,2048,0.006347041645299144,3072,3221225472,3246391296.0,0.9922480620155039,322670.011392918,qk_ar
12,256,256,1024,0.0031943104467592586,3072,1610612736,1623195648.0,0.9922480620155039,320569.96872013,qk_ar
12,256,256,512,0.0016183416350267381,3072,805306368,811597824.0,0.9922480620155039,316373.2483416833,qk_ar
12,256,256,256,0.0008325934893977947,3072,402653184,405798912.0,0.9922480620155039,307472.9784221131,qk_ar
12,256,256,128,0.0004389725746987952,3072,201326592,202899456.0,0.9922480620155039,291589.9702568624,qk_ar
12,256,256,64,0.00024191767449664432,3072,100663296,101449728.0,0.9922480620155039,264552.8076159138,qk_ar
12,256,256,32,0.0001431546143572621,3072,50331648,50724864.0,0.9922480620155039,223534.53392804778,qk_ar
12,256,256,16,9.404283597678917e-05,3072,25165824,25362432.0,0.9922480620155039,170135.23501087292,qk_ar
12,256,256,8,6.855550037091989e-05,3072,12582912,12681216.0,0.9922480620155039,116693.773026467,qk_ar
12,256,256,4,5.4802094978165945e-05,3072,6291456,6340608.0,0.9922480620155039,72989.91036006316,qk_ar
12,256,256,2,4.608510707869206e-05,3072,3145728,3170304.0,0.9922480620155039,43397.96795057727,qk_ar
1 n d seqlen bs latency h flop io intensity throughput series
2 3 256 256 2048 0.009623501679245285 768 618475290624 167.48502132816208 3692720015.912285 64267177503366.266 dense
3 3 256 256 1024 0.004853848615384615 768 309237645312 166.15392854317415 1861151572.059558 63709783682138.234 dense
4 3 256 256 512 0.0024687246971962615 768 154618822656 163.57953256539062 945221081.3366361 62631051097597.516 dense
5 3 256 256 256 0.0012845360838052097 768 77309411328 157.64931990085577 490388486.1451936 60184694149645.54 dense
6 3 256 256 128 0.0006901147179878049 768 38654705664 147.57393422494675 261934506.70684624 56012000116019.945 dense
7 3 256 256 64 0.0003363830693015702 768 19327352832 153.1328437752606 126212981.84970059 57456378146882.51 dense
8 3 256 256 32 0.00018671159748991485 768 9663676416 141.10249365427362 68486928.65540518 51757237075334.75 dense
9 3 256 256 16 0.00012353640857142858 768 4831838208 111.40488993609125 43371868.24359184 39112665358133.98 dense
10 3 256 256 8 9.774760007849294e-05 768 2415919104 76.43260800265766 31608487.09906635 24715891766754.14 dense
11 3 256 256 4 6.672271167474822e-05 768 1207959552 64.82614227498455 18633833.660438772 18104173551704.773 dense
12 3 256 256 2 4.9758770289855074e-05 768 603979776 55.317122669351576 10918495.880745342 12138157202874.861 dense
13 3 256 1 2048 9.785507940251571e-05 768 2415919104 76.34865809334705 31643242.518371396 24688745017132.86 dense
14 3 256 1 1024 6.692813470149253e-05 768 1207959552 64.62717090938949 18691202.70936228 18048606275785.867 dense
15 3 256 1 512 4.9680950036205655e-05 768 603979776 55.40377142534654 10901419.893658841 12157170415618.898 dense
16 3 256 1 256 4.2781118741058655e-05 768 301989888 45.95672244805227 6571179.83862661 7058952568020.829 dense
17 3 256 1 128 5.0662328255350016e-05 768 150994944 31.046026784880404 4863583.512513602 2980418571348.519 dense
18 3 256 1 64 4.475009253945481e-05 768 75497472 30.75426042497223 2454862.219307235 1687090857598.4766 dense
19 3 256 1 32 4.51682671454219e-05 768 37748736 28.29313765537115 1334201.1218340008 835735758435.5786 dense
20 3 256 1 16 5.03585186661834e-05 768 18874368 24.401035466223117 773506.846712577 374799904761.1871 dense
21 3 256 1 8 5.023459565217391e-05 768 9437184 23.972005435021096 393675.19858030166 187862246674.45105 dense
22 3 256 1 4 5.053219391083726e-05 768 4718592 23.58765586356967 200044.97383259286 93377936614.54384 dense
23 3 256 1 2 4.4607398995335484e-05 768 2359296 26.58285456464288 88752.54515134107 52890239133.797226 dense
24 12 256 256 2048 0.14480779847058822 3072 9895604649984 44.620009282941716 221775046868.20184 68336130750540.26 dense
25 12 256 256 1024 0.07254347629166667 3072 4947802324992 44.664248332585096 110777691547.58836 68204648824643.82 dense
26 12 256 256 512 0.036310761444444443 3072 2473901162496 44.876147984203506 55127306456.13385 68131349056975.164 dense
27 12 256 256 256 0.01821551906896552 3072 1236950581248 45.24607467289738 27338295977.947884 67906414116709.98 dense
28 12 256 256 128 0.009229417903030302 3072 618475290624 45.67217092440895 13541622351.335684 67011299859001.46 dense
29 12 256 256 64 0.004754550595394737 3072 309237645312 46.31372736116993 6677019167.566916 65040352207320.695 dense
30 12 256 256 32 0.002405752659340659 3072 154618822656 49.68826015254682 3111777755.5766335 64270456921525.82 dense
31 12 256 256 16 0.0012287219045005488 3072 77309411328 56.323579604557374 1372594069.3184311 62918558743709.18 dense
32 12 256 256 8 0.0006206816149425287 3072 38654705664 70.95456179103653 544781120.315271 62277832520589.78 dense
33 12 256 256 4 0.0003875502697142857 3072 19327352832 81.16954743236613 238110885.71245712 49870569942445.75 dense
34 12 256 256 2 0.00027502018627941914 3072 9663676416 91.50537035282076 105607751.53129694 35138062215483.168 dense
35 12 256 1 2048 0.0006202853873290136 3072 38654705664 70.99988634205897 544433345.6784943 62317614526515.766 dense
36 12 256 1 1024 0.00038721467732724153 3072 19327352832 81.2398957010995 237904697.74985722 49913791918755.53 dense
37 12 256 1 512 0.000274364799 3072 9663676416 91.72395326121995 105356082.81599998 35221998052308.45 dense
38 12 256 1 256 0.00012488918589482266 3072 4831838208 176.31707535146046 27404255.647778228 38689003962834.75 dense
39 12 256 1 128 8.976711102514506e-05 3072 2415919104 227.78088507574267 10606329.425740216 26913187652026.21 dense
40 12 256 1 64 8.715176287471176e-05 3072 1207959552 225.59268282689945 5354604.31102229 13860414432884.701 dense
41 12 256 1 32 8.523013435114503e-05 3072 603979776 226.06539514085782 2671703.8033338524 7086458100741.991 dense
42 12 256 1 16 7.901561645904116e-05 3072 301989888 241.35704882952732 1251216.3595988373 3821901309300.556 dense
43 12 256 1 8 7.827949114210329e-05 3072 150994944 242.37091635608994 622991.1833900034 1928920867994.581 dense
44 12 256 1 4 7.779445951035782e-05 3072 75497472 243.25022783249054 310369.58391664835 970473636235.5986 dense
45 12 256 1 2 7.758845406626506e-05 3072 37748736 243.57933441822672 154975.11761480253 486525172518.07056 dense
46 3 256 256 2048 0.00507974918466899 768 206158430208 475.59810852303485 433471930.42508715 40584371927298.98 qk_init
47 3 256 256 1024 0.0025616677649325623 768 103079215104 471.5519977009198 218595649.27424532 40239103803811.82 qk_init
48 3 256 256 512 0.0013029336670480549 768 51539607552 463.55374128015677 111183672.92143403 39556585922573.38 qk_init
49 3 256 256 256 0.0006738189029345373 768 25769803776 448.1766342333362 57499213.050413854 38244406121244.69 qk_init
50 3 256 256 128 0.000358254672959467 768 12884901888 421.47375986100144 30571065.425874516 35965760841472.125 qk_init
51 3 256 256 64 0.0002007051105022831 768 6442450944 376.1611839930762 17126836.096194826 32099087700742.5 qk_init
52 3 256 256 32 0.00012189697230142565 768 3221225472 309.6773881032524 10401874.969721656 26425803784810.87 qk_init
53 3 256 256 16 8.453561698040722e-05 768 1610612736 223.2711923587723 7213705.982328083 19052475081281.902 qk_init
54 3 256 256 8 6.407660705009276e-05 768 805306368 147.2797083750448 5467870.468274581 12567868448003.822 qk_init
55 3 256 256 4 5.036328747284576e-05 768 402653184 93.69110391262903 4297667.197682838 7994974200544.344 qk_init
56 3 256 256 2 4.5488761135057476e-05 768 201326592 51.865470527877875 3881707.616858238 4425853485045.578 qk_init
57 12 256 256 2048 0.020202365999999996 3072 824633720832 478.3437947812648 1723935231.9999998 40818670488001.266 qk_init
58 12 256 256 1024 0.010124155888157895 3072 412316860416 477.2583770318811 863927969.1228071 40726048173387.19 qk_init
59 12 256 256 512 0.005085633937062937 3072 206158430208 475.04777848703077 433974095.9627039 40537410430893.29 qk_init
60 12 256 256 256 0.0025654916853281853 3072 103079215104 470.84913933193053 218921957.14800516 40179126556324.74 qk_init
61 12 256 256 128 0.0013045765704467354 3072 51539607552 462.9699702434292 111323867.34478809 39506770794105.96 qk_init
62 12 256 256 64 0.0006742801519939804 3072 25769803776 447.87005387442576 57538572.970153 38218244597284.33 qk_init
63 12 256 256 32 0.00035831976790671853 3072 12884901888 421.3971919051604 30576620.194706645 35959227042573.69 qk_init
64 12 256 256 16 0.0002005369068918302 3072 6442450944 376.4766953382971 17112482.721436176 32126011335534.68 qk_init
65 12 256 256 8 0.00012179187250509165 3072 3221225472 309.94462293386505 10392906.453767821 26448607823689.82 qk_init
66 12 256 256 4 8.452507263643351e-05 3072 1610612736 223.2990450204527 7212806.198308992 19054851841745.297 qk_init
67 12 256 256 2 6.412381767545489e-05 3072 805306368 147.17127491946468 5471899.108305484 12558615459794.32 qk_init
68 3 256 256 2048 0.0016183739398395718 768 805306368 811597824.0 0.9922480620155039 1265467.7325087283 qk_ar
69 3 256 256 1024 0.0008322699728813558 768 402653184 405798912.0 0.9922480620155039 1230369.9921491416 qk_ar
70 3 256 256 512 0.00043886859397590365 768 201326592 202899456.0 0.9922480620155039 1166636.2255762408 qk_ar
71 3 256 256 256 0.00024185948322147648 768 100663296 101449728.0 0.9922480620155039 1058465.8355760013 qk_ar
72 3 256 256 128 0.00014308985100166944 768 50331648 50724864.0 0.9922480620155039 894542.82818777 qk_ar
73 3 256 256 64 9.382939365815932e-05 768 25165824 25362432.0 0.9922480620155039 682089.028872613 qk_ar
74 3 256 256 32 6.856070612244899e-05 768 12582912 12681216.0 0.9922480620155039 466739.6503012703 qk_ar
75 3 256 256 16 5.452260553129549e-05 768 6291456 6340608.0 0.9922480620155039 293456.26174846216 qk_ar
76 3 256 256 8 4.608557533261417e-05 768 3145728 3170304.0 0.9922480620155039 173590.1080166944 qk_ar
77 3 256 256 4 4.386146957766642e-05 768 1572864 1585152.0 0.9922480620155039 91196.21477609445 qk_ar
78 3 256 256 2 4.330941094420601e-05 768 786432 792576.0 0.9922480620155039 46179.33969539622 qk_ar
79 12 256 256 2048 0.006347041645299144 3072 3221225472 3246391296.0 0.9922480620155039 322670.011392918 qk_ar
80 12 256 256 1024 0.0031943104467592586 3072 1610612736 1623195648.0 0.9922480620155039 320569.96872013 qk_ar
81 12 256 256 512 0.0016183416350267381 3072 805306368 811597824.0 0.9922480620155039 316373.2483416833 qk_ar
82 12 256 256 256 0.0008325934893977947 3072 402653184 405798912.0 0.9922480620155039 307472.9784221131 qk_ar
83 12 256 256 128 0.0004389725746987952 3072 201326592 202899456.0 0.9922480620155039 291589.9702568624 qk_ar
84 12 256 256 64 0.00024191767449664432 3072 100663296 101449728.0 0.9922480620155039 264552.8076159138 qk_ar
85 12 256 256 32 0.0001431546143572621 3072 50331648 50724864.0 0.9922480620155039 223534.53392804778 qk_ar
86 12 256 256 16 9.404283597678917e-05 3072 25165824 25362432.0 0.9922480620155039 170135.23501087292 qk_ar
87 12 256 256 8 6.855550037091989e-05 3072 12582912 12681216.0 0.9922480620155039 116693.773026467 qk_ar
88 12 256 256 4 5.4802094978165945e-05 3072 6291456 6340608.0 0.9922480620155039 72989.91036006316 qk_ar
89 12 256 256 2 4.608510707869206e-05 3072 3145728 3170304.0 0.9922480620155039 43397.96795057727 qk_ar