Weight-normalized Conv1d layer optimized for decoding

Weight-normalized Conv1d layer optimized for decoding

Length of this dataset is the sum of individual datasets

Does the heavy lifting for creating a batch from the input list of examples. The logic is as follows: 1. Mask the input b

Convert a list of 2d frames into a padded 3d tensor Args: frames (list): list of 2d frames of size L[i]*f_dim. Where L[i] is

Flatten a nested dictionary.

Turn the convolution filters into band matrices and do matrix multiplication. This is faster when the sequence is short, but less memory effic

Turn the convolution filters into band matrices and do matrix multiplication. This is faster when the sequence is short, but less memory effic

The conventional implementation of convolutions. Unfolding the input by having a window shifting to the right.

The conventional implementation of convolutions. Unfolding the input by having a window shifting to the right.

Go through a single document and genrate sentence paris from it

Args: append_eos: if True, each input sentence in the source tokens tensor will have an EOS appended to the end.

Args: append_eos: if True, each input sentence in the source tokens tensor will have an EOS appended to the end.

Build a map from languages to ids. These ids are used as segment labels for cross-lingual LM training.

Give a list of evenly cut blocks/sentences, pair these sentences with 50% consecutive sentences and 50% random sentences. Thi

Sample among the smallest set of elements whose cumulative probability mass exceeds p. See `"The Curious Case of Neural Text Degeneration"

Build index mapping block indices to the underlying dataset indices

Split and transpose encoder outputs. This is cached when doing incremental inference.

Split and transpose encoder outputs.

Trancate a pair of sentence to limit total length under max_num_tokens Logics: 1. Truncate longer sentence 2.

Unflatten a flattened dictionary into a nested dictionary.

Helper for upgrading old model checkpoints.

Add task-specific arguments to the parser.

Add model-specific arguments to the parser.

Loads a pre-existing dictionary from a text file and adds its symbols to this instance.

Return the source and target datasets for masked LM training.

Backtranslate a list of samples. Given an input (*samples*) of the form: [{'id': 1, 'source': 'hallo welt'}] this will return:

Computes the sum of gradients of the given tensor w.r.t. graph leaves. Compared to :func:`fairseq.optim.FairseqOptimizer.backward`, this

Computes the sum of gradients of the given tensor w.r.t. graph leaves. Compared to :func:`fairseq.optim.FairseqOptimizer.backward`, this