noam

allennlp.training.learning_rate_schedulers.noam

[SOURCE]

---

NoamLR

@LearningRateScheduler.register("noam")
class NoamLR(LearningRateScheduler):
 | def __init__(
 |     self,
 |     optimizer: torch.optim.Optimizer,
 |     model_size: int,
 |     warmup_steps: int,
 |     factor: float = 1.0,
 |     last_epoch: int = -1
 | ) -> None

Implements the Noam Learning rate schedule. This corresponds to increasing the learning rate linearly for the first warmup_steps training steps, and decreasing it thereafter proportionally to the inverse square root of the step number, scaled by the inverse square root of the dimensionality of the model. Time will tell if this is just madness or it's actually important.

Registered as a LearningRateScheduler with name "noam".

Parameters

• optimizer : torch.optim.Optimizer
  This argument does not get an entry in a configuration file for the object.
• model_size : int
  The hidden size parameter which dominates the number of parameters in your model.
• warmup_steps : int
  The number of steps to linearly increase the learning rate.
• factor : float, optional (default = 1.0)
  The overall scale factor for the learning rate decay.

step

class NoamLR(LearningRateScheduler):
 | ...
 | @overrides
 | def step(self, metric: float = None) -> None

step_batch

class NoamLR(LearningRateScheduler):
 | ...
 | def step_batch(self, batch_num_total: int = None) -> None

get_values

class NoamLR(LearningRateScheduler):
 | ...
 | def get_values(self)