CosineAnnealingLR in the PyTorch documentation (source code)

= CosineAnnealingLR in the PyTorch documentation
{c}

https://docs.pytorch.org/docs/2.15/generated/torch.optim.lr_scheduler.CosineAnnealingLR.html

Primary implementation documentation: class description, closed-form <cosine learning rate schedule>, `T_max` parameter and explicit note distinguishing the implementation without restarts from <SGDR>. The example advances the schedule after training and validation.