List your company
ResearchFormerly WizWand

Language Modeling on Pile uncopyrighted (test)

3.608Worst Log-PerplexityFGDRO-KL-Adam
3.426524.651515.87657.10149Oct 8, 2024
Updated 8mo ago

Evaluation results

MethodLinks
FGDRO-KL-AdamModel=GPT2, Local iterations=200K, Batch size=322024.103.6082.653
FedAdamModel=GPT2, Local iterations=200K, Batch size=322024.107.2426.479
FGDRO-KLModel=GPT2, Local iterations=200K, Batch size=322024.107.9326.664
SCAFFOLDModel=GPT2, Local iterations=200K, Batch size=322024.107.9756.901
DRFAModel=GPT2, Local iterations=200K, Batch size=322024.108.0146.702
DR-DSGDModel=GPT2, Local iterations=200K, Batch size=322024.108.0236.693
FedProxModel=GPT2, Local iterations=200K, Batch size=322024.108.0796.724
FedAvgModel=GPT2, Local iterations=200K, Batch size=322024.108.0856.785
FGDRO-CVaRModel=GPT2, Local iterations=200K, Batch size=322024.108.1456.907