Skip to content

Instantly share code, notes, and snippets.

@younesbelkada
Created November 10, 2022 11:20
Show Gist options
  • Save younesbelkada/1886b368943354b468f77bcea7aadc6f to your computer and use it in GitHub Desktop.
Save younesbelkada/1886b368943354b468f77bcea7aadc6f to your computer and use it in GitHub Desktop.

bert-base-uncased - A100 - half - cuda

batch_size seq_len pad_percentage HF_time BT_time Speedup
8 64 0 0.012947250976562501 0.007366779174804688 1.7575185395598296
8 64 0.1 0.012887490234375 0.0072092669677734375 1.7876283805252486
8 64 0.2 0.012887449951171874 0.007392645263671876 1.7432799074645124
8 64 0.5 0.012756295166015626 0.007139061889648438 1.7868307297506574
8 64 0.75 0.0132109521484375 0.007268699951171876 1.8175123800931694
8 128 0 0.012552437744140625 0.0075138049316406256 1.6705833939449668
8 128 0.1 0.0132246533203125 0.007377489624023438 1.7925682033152373
8 128 0.2 0.01270349853515625 0.007361699829101562 1.7256202820085118
8 128 0.5 0.01253486572265625 0.0073076123046875 1.715316193582917
8 128 0.75 0.01280079833984375 0.007162613525390625 1.7871686493298042
8 256 0 0.013096918945312501 0.0073260644531249995 1.7877154956950851
8 256 0.1 0.012815031738281251 0.007538012084960938 1.7000545493749575
8 256 0.2 0.0127824072265625 0.007293112182617187 1.75266839539762
8 256 0.5 0.012925665283203125 0.007354347534179687 1.7575543205064037
8 256 0.75 0.012753940429687501 0.007219097900390626 1.766694482560956
batch_size seq_len pad_percentage HF_time BT_time Speedup
16 64 0 0.012791541748046874 0.0075867138671875 1.6860451009455133
16 64 0.1 0.012824945068359375 0.007612374877929687 1.6847495392722085
16 64 0.2 0.01257017333984375 0.007708118896484375 1.6307705561699273
16 64 0.5 0.012674600830078124 0.0073746228027343754 1.7186778455134835
16 64 0.75 0.013131120605468751 0.007327129516601563 1.7921234469401286
16 128 0 0.0125453515625 0.0074241845703125 1.689795215041635
16 128 0.1 0.012719411621093751 0.00749285400390625 1.697538963719664
16 128 0.2 0.01282830322265625 0.007500534057617188 1.7103186418610328
16 128 0.5 0.012582399902343751 0.007418777465820313 1.6960206665199498
16 128 0.75 0.01282303955078125 0.007292477416992188 1.758392767991617
16 256 0 0.013372744140625002 0.00832458740234375 1.6064152485034833
16 256 0.1 0.012801024169921875 0.007841935424804687 1.6323807168101871
16 256 0.2 0.013163704833984375 0.0075897857666015625 1.7343974176333854
16 256 0.5 0.013022576904296874 0.0077078936767578125 1.6895117460642748
16 256 0.75 0.01305870361328125 0.007338618774414063 1.779449786765071
batch_size seq_len pad_percentage HF_time BT_time Speedup
64 64 0 0.013192867431640625 0.00828737548828125 1.591923456382057
64 64 0.1 0.01270013916015625 0.008382853393554688 1.5150138698502096
64 64 0.2 0.012997877197265625 0.008232857666015624 1.578780749595551
64 64 0.5 0.012849664306640625 0.008375623779296874 1.5341740084365802
64 64 0.75 0.013100318603515626 0.008229109497070312 1.5919485101248851
64 128 0 0.014951875000000002 0.013786091308593752 1.0845623074235367
64 128 0.1 0.014791700439453126 0.012987801513671874 1.1388917842548134
64 128 0.2 0.014881505126953125 0.012005662841796876 1.2395404837743906
64 128 0.5 0.01511077880859375 0.008753684692382812 1.7262192253443496
64 128 0.75 0.015181844482421875 0.008237485961914063 1.8430191629600325
64 256 0 0.032153271484374996 0.0269648681640625 1.1924134502992807
64 256 0.1 0.03160266845703125 0.0255294677734375 1.2378898274531478
64 256 0.2 0.032148212890625 0.023026831054687502 1.3961197185263878
64 256 0.5 0.0324914794921875 0.01525843994140625 2.1294103209081428
64 256 0.75 0.032503767089843746 0.0089327001953125 3.638739281421348
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment