← Back to LR Schedule
warmup + warmdown + cosine decay
LR ScheduleUsed in
1 PRs
Best BPB
1.1532
Avg BPB
1.1532
Submissions
Hyperparameters Across PRs
| pr_number | parameters |
|---|---|
| 456 | {"warmup_steps":20,"warmdown_steps":3000,"total_steps":5200} |