ScheduleWrapper

class opennmt.schedules.ScheduleWrapper(schedule, step_start=0, step_duration=1, minimum_learning_rate=0)[source]

Wrapper to augment a learning rate scheduler behavior.

Inherits from: keras.src.optimizers.schedules.learning_rate_schedule.LearningRateSchedule

__init__(schedule, step_start=0, step_duration=1, minimum_learning_rate=0)[source]

Initializes the decay function.

Parameters
  • schedule – A tf.keras.optimizers.schedules.LearningRateSchedule.

  • step_duration – The number of training steps that make 1 decay step.

  • start_step – Start decay after this many steps.

  • minimum_learning_rate – Do not decay past this learning rate value.