Implement up to the computation of the gate

Train/validate loss statistics.

A two-layer Feed-Forward-Network.

Decoder built on CNN, which consists of resduial convolutional layers, with ConvMultiStepAttention.

Encoder built on CNN.

Generator module that additionally considers copying words directly from the source.

A simple network container. Parameters are a list of modules. Inputs are a 3d Variable whose last dimension is the same length as the

Words embeddings dictionary for encoder/decoder. Args: word_vec_size (int): size of the dictionary of embeddings. position_e

Encoder recurrent neural network for Images.

Stardard RNN decoder, with Input Feed and Attention.

Implementation of the matrix-tree theorem for computing marginals of non-projective dependency parsing. This attention layer is used in the pa

A trivial encoder without RNN, just takes mean as final state.

The encoder + decoder Neural Machine Translation Model.

The standard RNN encoder.

Stardard RNN decoder, with Attention. Currently no 'coverage_attn' and 'copy_attn' support.

The Transformer decoder from "Attention is All You Need".

The Transformer encoder from "Attention is All You Need".

Apply the context gate to both contexts

Copy Generator Loss Computation.

DecoderState is a base class for models, used during translation for storing translation states.

EncoderBase class for sharing code among various encoder.

Google NMT ranking score from Wu et al.

Luong Attention. Global attention takes a matrix and a query vector. It then computes a parameterized convex combination of the matrix

Layer normalization module

This is the loss criterion base class. Users can implement their own loss computation strategy by making subclass of this one. Users need

A really bare-bones argparse help formatter that generates valid markdown. This will generate something like: usage # **section heading**:

Multi-Head Attention module from "Attention is All You Need".

Standard NMT Loss Computation.

Defines a dataset for machine translation. An ONMTDataset is a collection that supports iteration over its examples. The parent class sup

RNN decoder base class.

Apply the context gate only to the source context

Our own implementation of stacked LSTM. Needed for the decoder, because we do input feeding.

Apply the context gate only to the target context