ml.lr_schedulers.linear_no_decay

Defines a linear warmup scheduler without decay.

class ml.lr_schedulers.linear_no_decay.LinearNoDecayLRSchedulerConfig(name: str = '???', warmup_steps: int = 1000)[source]

class ml.lr_schedulers.linear_no_decay.LinearNoDecayLRScheduler(config: BaseConfigT)[source]

get_lr_scale(state: State) → float[source]

Given a state, returns the current learning rate.