The original version: "End-to-End Automatic Speech Translation of Audiobooks" (https://arxiv.org/abs/1802.04200)

Build the :class:`~fairseq.criterions.FairseqCriterion` instance for this task. Args: cfg (omegaconf.DictConfig)

Build a new model instance.

Helper function to build shared embeddings for a set of languages after checking that all dicts corresponding to those languages are

Retrieves all checkpoints found in `path` directory. Checkpoints are identified by matching filename to the specified pattern. If the pattern

For every centroid m, computes ||M - m[None, :]||_2 Remarks: - We rely on PyTorch's broadcast

Returns whether the most recent epoch iterator has been exhausted

Similar to *forward* but only return features. Includes several features from "Jointly Learning to Align and Translate with

Get a single smoothed value.

Return an EpochBatchIterator over the training set for a given epoch.

Get the waveform and sample rate of a 16-bit mono-channel WAV or FLAC. Args: path_or_fp (str or BinaryIO): the path or file-like object

Whether the iterator has been exhausted.

Adjust the learning rate at the end of the epoch.

Update the learning rate after each update.

Helper to get index of mask symbol

Parse the configuration of lambda coefficient (for scheduling). x = "3" # lambda will be a constant equal to x x = "0:1,

Do we require PathManager to access given path?

Register a classification head.

Usage:: x = torch.Tensor([1]) y = torch.Tensor([2]) tensors, packed_non_tensors = split_non_tensors((x, y, None, 3))

Called at the end of each step.

Returns a counter of the number of times each token is used in a constraint.

unfold T x B x C to T x B x C x K

Transforms *one row* of a packed constraint tensor (e.g., for one sentence in the batch) into a list of constraint tensors.

Updates the running statistics when quantizing a new layer.

Upgrade old state dicts to work with newer code.