← Back to LR Schedule
cosine decay with linear warmup
LR ScheduleUsed in
3 PRs
Best BPB
1.0574
Avg BPB
1.1451
Submissions
Hyperparameters Across PRs
| pr_number | parameters |
|---|---|
| 131 | {"warmup_steps":200,"min_lr_ratio":0.05} |
| 612 | {"warmup_steps":20,"warmdown_start_step":7000,"total_steps":12000} |
| 684 | {"start_lr":0.0005,"end_lr":0.00002,"warmup":"1 epoch"} |