10.6 The Encoder–Decoder Architecture

In general sequence-to-sequence problems like machine translation (Section 10.5), inputs and outputs are of varying lengths that are unaligned. The standard approach to handling this sort of data is to design an encoder–decoder architecture (Figure 10.6.1) consisting of two major components: an encoder that takes a variable-length sequence as input, and a decoder that acts as a conditional language model, taking in the encoded input and the leftwards context of the target sequence and predicting the subsequent token in the target sequence.

Figure 10.6.1: The encoder–decoder architecture.

Let’s take machine translation from English to French as an example. Given an input sequence in English: “They”, “are”, “watching”, “.”, this encoder–decoder architecture first encodes the variable-length input into a state, then decodes the state to generate the translated sequence, token by token, as output: “Ils”, “regardent”, “.”. Since the encoder–decoder architecture forms the basis of different sequence-to-sequence models in subsequent sections, this section will convert this architecture into an interface that will be implemented later.

from d2l import torch as d2l
from torch import nn

from d2l import tensorflow as d2l
import tensorflow as tf

from d2l import jax as d2l
from flax import linen as nn

from d2l import mxnet as d2l
from mxnet.gluon import nn

10.6.1 Encoder

In the encoder interface, we just specify that the encoder takes variable-length sequences as input X. The implementation will be provided by any model that inherits this base Encoder class.

class Encoder(nn.Module):
    """The base encoder interface for the encoder--decoder architecture."""
    def __init__(self):
        super().__init__()

    # Later there can be additional arguments (e.g., length excluding padding)
    def forward(self, X, *args):
        raise NotImplementedError

class Encoder(tf.keras.layers.Layer):
    """The base encoder interface for the encoder--decoder architecture."""
    def __init__(self):
        super().__init__()

    # Later there can be additional arguments (e.g., length excluding padding)
    def call(self, X, *args):
        raise NotImplementedError

class Encoder(nn.Module):
    """The base encoder interface for the encoder--decoder architecture."""
    def setup(self):
        raise NotImplementedError

    # Later there can be additional arguments (e.g., length excluding padding)
    def __call__(self, X, *args):
        raise NotImplementedError

class Encoder(nn.Block):
    """The base encoder interface for the encoder--decoder architecture."""
    def __init__(self):
        super().__init__()

    # Later there can be additional arguments (e.g., length excluding padding)
    def forward(self, X, *args):
        raise NotImplementedError

10.6.2 Decoder

In the following decoder interface, we add an additional init_state method to convert the encoder output (enc_all_outputs) into the encoded state. Note that this step may require extra inputs, such as the valid length of the input, which was explained in Section 10.5. To generate a variable-length sequence token by token, every time the decoder may map an input (e.g., the generated token at the previous time step) and the encoded state into an output token at the current time step.

class Decoder(nn.Module):
    """The base decoder interface for the encoder--decoder architecture."""
    def __init__(self):
        super().__init__()

    # Later there can be additional arguments (e.g., length excluding padding)
    def init_state(self, enc_all_outputs, *args):
        raise NotImplementedError

    def forward(self, X, state):
        raise NotImplementedError

class Decoder(tf.keras.layers.Layer):
    """The base decoder interface for the encoder--decoder architecture."""
    def __init__(self):
        super().__init__()

    # Later there can be additional arguments (e.g., length excluding padding)
    def init_state(self, enc_all_outputs, *args):
        raise NotImplementedError

    def call(self, X, state):
        raise NotImplementedError

class Decoder(nn.Module):
    """The base decoder interface for the encoder--decoder architecture."""
    def setup(self):
        raise NotImplementedError

    # Later there can be additional arguments (e.g., length excluding padding)
    def init_state(self, enc_all_outputs, *args):
        raise NotImplementedError

    def __call__(self, X, state):
        raise NotImplementedError

class Decoder(nn.Block):
    """The base decoder interface for the encoder--decoder architecture."""
    def __init__(self):
        super().__init__()

    # Later there can be additional arguments (e.g., length excluding padding)
    def init_state(self, enc_all_outputs, *args):
        raise NotImplementedError

    def forward(self, X, state):
        raise NotImplementedError

10.6.3 Putting the Encoder and Decoder Together

In the forward propagation, the output of the encoder is used to produce the encoded state, and this state will be further used by the decoder as one of its input.

class EncoderDecoder(d2l.Classifier):
    """The base class for the encoder--decoder architecture."""
    def __init__(self, encoder, decoder):
        super().__init__()
        self.encoder = encoder
        self.decoder = decoder

    def forward(self, enc_X, dec_X, *args):
        enc_all_outputs = self.encoder(enc_X, *args)
        dec_state = self.decoder.init_state(enc_all_outputs, *args)
        # Return decoder output only
        return self.decoder(dec_X, dec_state)[0]

class EncoderDecoder(d2l.Classifier):
    """The base class for the encoder--decoder architecture."""
    def __init__(self, encoder, decoder):
        super().__init__()
        self.encoder = encoder
        self.decoder = decoder

    def call(self, enc_X, dec_X, *args):
        enc_all_outputs = self.encoder(enc_X, *args, training=True)
        dec_state = self.decoder.init_state(enc_all_outputs, *args)
        # Return decoder output only
        return self.decoder(dec_X, dec_state, training=True)[0]

class EncoderDecoder(d2l.Classifier):
    """The base class for the encoder--decoder architecture."""
    encoder: nn.Module
    decoder: nn.Module
    training: bool

    def __call__(self, enc_X, dec_X, *args):
        enc_all_outputs = self.encoder(enc_X, *args, training=self.training)
        dec_state = self.decoder.init_state(enc_all_outputs, *args)
        # Return decoder output only
        return self.decoder(dec_X, dec_state, training=self.training)[0]

class EncoderDecoder(d2l.Classifier):
    """The base class for the encoder--decoder architecture."""
    def __init__(self, encoder, decoder):
        super().__init__()
        self.encoder = encoder
        self.decoder = decoder

    def forward(self, enc_X, dec_X, *args):
        enc_all_outputs = self.encoder(enc_X, *args)
        dec_state = self.decoder.init_state(enc_all_outputs, *args)
        # Return decoder output only
        return self.decoder(dec_X, dec_state)[0]

In the next section, we will see how to apply RNNs to design sequence-to-sequence models based on this encoder–decoder architecture.

10.6.4 Summary

Encoder-decoder architectures can handle inputs and outputs that both consist of variable-length sequences and thus are suitable for sequence-to-sequence problems such as machine translation. The encoder takes a variable-length sequence as input and transforms it into a state with a fixed shape. The decoder maps the encoded state of a fixed shape to a variable-length sequence.

10.6.5 Exercises

Suppose that we use neural networks to implement the encoder–decoder architecture. Do the encoder and the decoder have to be the same type of neural network?
Besides machine translation, can you think of another application where the encoder–decoder architecture can be applied?

Discussions