init error handler

rnn_type must be a class inheriting from torch.nn.RNNBase

Extended version of the definition in torchtext.data.Iterator. Added yield_raw_example behaviour to yield a torchtext.data.Example

Compute the loss. The args must match :func:`self._make_shard_state()`. Args: batch: the current batch. outp

Split apart word features (like POS/NER tags) from the tokens. Args: string (str): A string with ``tok_delim`` joining tokens and

Using input feed by concatenating input with attention vectors.

Args: dictionaries with disjoint keys. Returns: a single dictionary that has the union of these keys.

See base class for args description.

See base class method `ReportMgrBase.report_step`.

See base class method `ReportMgrBase.report_training`.

See StdRNNDecoder._run_forward_pass() for description of arguments and return values.

DecodeStrategy subclasses should override :func:`advance()`. Advance is used to update ``self.alive_seq``, ``self.is_finished``, and,

Select next tokens randomly from the top k possible next tokens. Args: log_probs (FloatTensor): Shaped ``(batch_size, vocab_size)

All-reduce and rescale tensors in chunks of the specified size. Args: tensors: list of Tensors to all-reduce rescale_denom: denom

Apply mask

Sort using duration time of the sound spectrogram.

Pad and batch a sequence of images.

Transform from 3D to 2D, apply linear and return initial size

Returns a LossCompute subclass which wraps around an nn.Module subclass (such as nn.NLLLoss) which defines the loss criterion. The LossComput

Build the fields for all data sides. Args: train_dataset_files: a list of train dataset pt file. fields (dict[str, Field]): field

Returns zero as penalty

Our summary penalty.

GNMT coverage re-ranking score. See "Google's Neural Machine Translation System" :cite:`wu2016google`. ``cov`` is expected to be size

Decorator for critical section (mutually exclusive code)

Timeout function that frees GPU memory. Moves the model to CPU or unloads it; depending on attr`self.on_timemout` value

Embedding look-up table.

error listener

A standard exponential decay, scaling the learning rate by :obj:`rate` every :obj:`decay_steps` steps.

Return whether an example is an acceptable length. If used with a dataset as ``filter_pred``, use :func:`partial()` for all keyword arguments

Keep a progress bar of frames finished.

Args: inputs (FloatTensor): ``(batch_size, input_len, model_dim)`` Returns: (FloatTensor, FloatTensor):

sparsemax: normalizing sparse transform (a la softmax) Parameters: input (Tensor): any shape dim: dimension along whi

Args: source (FloatTensor): query vectors ``(batch, tgt_len, dim)`` memory_bank (FloatTensor): source vectors ``(batch,

input (FloatTensor): ``(n, num_classes)``. target (LongTensor): ``(n,)``, the indices of the target classes