← Back to Regularization

label smoothing

Regularization
Used in
7 PRs
Best BPB
0.8503
Avg BPB
1.0816

Hyperparameters Across PRs

pr_numberparameters
375{"value":0.05}
667{"value":0.05}
1124{"value":0}
1368{"value":0.1}
1380
1602
1702{"bpb_weighted_loss":true,"weight_power":0.5,"weight_clip":2}