espnet.nets.pytorch_backend.transformer.optimizer.NoamOpt

source

class espnet.nets.pytorch_backend.transformer.optimizer.NoamOpt(model_size, factor, warmup, optimizer)

Bases: object

Optim wrapper that implements rate.

Construct an NoamOpt object.

load_state_dict(state_dict)

Load state_dict.

property param_groups

Return param_groups.

rate(step=None)

Implement lrate above.

state_dict()

Return state_dict.

step()

Update parameters and rate.

zero_grad()

Reset gradient.