List your company
ResearchFormerly WizWand

Language Modeling on CNN Daily Mail (test)

16.84PerplexityDropout
16.79217.11617.4417.764Sep 19, 2025
Updated 6mo ago

Evaluation results

MethodLinks
DropoutPercentage of influential data=20%, Model backbone=GPT-22025.0916.84
DataInfPercentage of influential data=20%, Model backbone=GPT-22025.0916.87
LOGRAPercentage of influential data=20%, Model backbone=GPT-22025.0916.89
Hessian-freePercentage of influential data=20%, Model backbone=GPT-22025.0916.9
Hessian-freePercentage of influential data=15%, Model backbone=GPT-22025.0916.91
DataInfPercentage of influential data=15%, Model backbone=GPT-22025.0916.91
FJLTPercentage of influential data=20%, Model backbone=GPT-22025.0916.91
DropoutPercentage of influential data=15%, Model backbone=GPT-22025.0916.95
FJLTPercentage of influential data=15%, Model backbone=GPT-22025.0916.98
LOGRAPercentage of influential data=15%, Model backbone=GPT-22025.0917.03
Random RankingPercentage of influential data=20%, Model backbone=GPT-22025.0917.05
LiSSAPercentage of influential data=20%, Model backbone=GPT-22025.0917.05
Random RankingPercentage of influential data=15%, Model backbone=GPT-22025.0917.1
DropoutPercentage of influential data=10%, Model backbone=GPT-22025.0917.1
LiSSAPercentage of influential data=15%, Model backbone=GPT-22025.0917.11
DataInfPercentage of influential data=10%, Model backbone=GPT-22025.0917.11
Hessian-freePercentage of influential data=10%, Model backbone=GPT-22025.0917.13
FJLTPercentage of influential data=10%, Model backbone=GPT-22025.0917.14
LOGRAPercentage of influential data=10%, Model backbone=GPT-22025.0917.24
FJLTPercentage of influential data=5%, Model backbone=GPT-22025.0917.25
DropoutPercentage of influential data=5%, Model backbone=GPT-22025.0917.27
LiSSAPercentage of influential data=10%, Model backbone=GPT-22025.0917.31
Random RankingPercentage of influential data=10%, Model backbone=GPT-22025.0917.38
LiSSAPercentage of influential data=5%, Model backbone=GPT-22025.0917.4
Random RankingPercentage of influential data=5%, Model backbone=GPT-22025.0917.43
Hessian-freePercentage of influential data=5%, Model backbone=GPT-22025.0917.43
LOGRAPercentage of influential data=5%, Model backbone=GPT-22025.0917.65
DataInfPercentage of influential data=5%, Model backbone=GPT-22025.0918.04