espnet.nets.chainer_backend.rnn.training.CustomParallelUpdater

source

class espnet.nets.chainer_backend.rnn.training.CustomParallelUpdater(train_iters, optimizer, converter, devices, accum_grad=1)

Bases: MultiprocessParallelUpdater

Custom Parallel Updater for chainer.

Defines the main update routine.

• Parameters:
  • train_iter (iterator | dict *[*str , iterator ]) – Dataset iterator for the training dataset. It can also be a dictionary that maps strings to iterators. If this is just an iterator, then the iterator is registered by the name 'main'.
  • optimizer (optimizer | dict *[*str , optimizer ]) – Optimizer to update parameters. It can also be a dictionary that maps strings to optimizers. If this is just an optimizer, then the optimizer is registered by the name 'main'.
  • converter (espnet.asr.chainer_backend.asr.CustomConverter) – Converter function to build input arrays. Each batch extracted by the main iterator and the device option are passed to this function. chainer.dataset.concat_examples() is used by default.
  • device (torch.device) – Device to which the training data is sent. Negative value indicates the host memory (CPU).
  • accum_grad (int) – The number of gradient accumulation. if set to 2, the network parameters will be updated once in twice, i.e. actual batchsize will be doubled.

Initialize Custom Parallel Updater.

update()

Update optimizers.

update_core()

Execute main update routine of the custom parallel updater.