starling.models.vae.KLDWeightScheduler

class KLDWeightScheduler[source]

Methods

`__init__`
`configure`	Configure the scheduler with training step information
`get_weight`	Get current KLD weight for a given step (or max weight if no warmup)

Attributes

`current_weight`	Get the current KLD weight
`max_weight`	Get the maximum KLD weight

__init__(max_weight: float, warmup_fraction: float = None, scheduler_type='cyclical')[source]

configure(total_training_steps: int)[source]: Configure the scheduler with training step information

get_weight(step=None) → float[source]: Get current KLD weight for a given step (or max weight if no warmup)