configs/model/lr_scheduler/warmup_cosine_decay.yaml

_partial_: true
_target_: utils.lr_scheduler.WarmupCosineDecayLR
warmup_steps: 500
total_steps: ${trainer.max_steps}