lightning_template.utils.optim

lightning_template.utils.optim#

Submodules#

Classes#

KeepAndLinearlyDecayLrScheduler

Keep lr first then decay learning rate linearly in optimizer.

WarmupScheduler

Warm-up(increasing) learning rate in optimizer.

Functions#

get_configure_optimizers_method(optim_config)

Package Contents#

lightning_template.utils.optim.get_configure_optimizers_method(optim_config)#
class lightning_template.utils.optim.KeepAndLinearlyDecayLrScheduler(optimizer, keep_epochs, decay_epochs, last_epoch=-1, verbose=False)#

Bases: torch.optim.lr_scheduler._LRScheduler

Keep lr first then decay learning rate linearly in optimizer.

Parameters:

  • optimizer (Optimizer) – Wrapped optimizer.

  • keep_epochs – target keep epoch.

  • decay_epochs – target decay epoch.

keep_epochs#
decay_epochs#
get_lr()#

Compute learning rate using chainable form of the scheduler.

class lightning_template.utils.optim.WarmupScheduler(*args, warmup_iters, warmup_ratio=0.1, warmup_mode='linear', **kwargs)#

Bases: torch.optim.lr_scheduler._LRScheduler

Warm-up(increasing) learning rate in optimizer.

Parameters:

  • optimizer (Optimizer) – Wrapped optimizer.

  • warmup_iters – target warm up epoch.

warmup_iters#
warmup_ratio = 0.1#
warmup_mode = 'linear'#
get_lr()#

Compute learning rate using chainable form of the scheduler.

Contents