List your company

Language Modeling on PG19 T5 (val)

15.31PPLXPrimer
15.266815.558415.8516.1416Sep 17, 2021

Evaluation results

MethodLinks
PrimerParams=110M, Steps=920K, Baseline Compute=1M2021.0915.311.81
Primer-EZParams=110M, Steps=896K, Baseline Compute=1M2021.0915.371.98
Transformer++Params=110M, Steps=998K, Baseline Compute=1M2021.0915.641.2
Vanilla TransformerParams=110M, Steps=1M, Baseline Compute=1M2021.0915.83—
Transformer+GELUParams=110M, Steps=998K, Baseline Compute=1M2021.0915.840.95
Primer-EZParams=110M, Steps=471K, Baseline Compute=525K2021.0915.841.74
PrimerParams=110M, Steps=483K, Baseline Compute=525K2021.0915.961.68
Transformer++Params=110M, Steps=524K, Baseline Compute=525K2021.0916.151.18
Transformer+GELUParams=110M, Steps=524K, Baseline Compute=525K2021.0916.351.01
Vanilla TransformerParams=110M, Steps=525K, Baseline Compute=525K2021.0916.39—