← Back to Architecture

EMA

Architecture
Used in
14 PRs
Best BPB
1.0597
Avg BPB
1.2640

Hyperparameters Across PRs

pr_numberparameters
307{"enabled":1,"decay":0.997}
338{"decay":0.997}
535{"decay":0.997}
610
617
831
1004{"decay":0.997}
1044
1104
1480
1654{"decay":0.996}
1670{"decay":0.9965}
1842
1913{"decay":0.9965}