Transformer model for language understanding

View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook

This tutorial trains a Transformer model to translate Portuguese to English. This is an advanced example that assumes knowledge of text generation and attention.

The core idea behind the Transformer model is self-attention—the ability to attend to different positions of the input sequence to compute a representation of that sequence. Transformer creates stacks of self-attention layers and is explained below in the sections Scaled dot product attention and Multi-head attention.

A transformer model handles variable-sized input using stacks of self-attention layers instead of RNNs or CNNs. This general architecture has a number of advantages:

  • It make no assumptions about the temporal/spatial relationships across the data. This is ideal for processing a set of objects (for example, StarCraft units).
  • Layer outputs can be calculated in parallel, instead of a series like an RNN.
  • Distant items can affect each other's output without passing through many RNN-steps, or convolution layers (see Scene Memory Transformer for example).
  • It can learn long-range dependencies. This is a challenge in many sequence tasks.

The downsides of this architecture are:

  • For a time-series, the output for a time-step is calculated from the entire history instead of only the inputs and current hidden-state. This may be less efficient.
  • If the input does have a temporal/spatial relationship, like text, some positional encoding must be added or the model will effectively see a bag of words.

After training the model in this notebook, you will be able to input a Portuguese sentence and return the English translation.

Attention heatmap

# Pin matplotlib version to 3.2.2 since in the latest version
# transformer.ipynb fails with the following error:
# https://stackoverflow.com/questions/62953704/valueerror-the-number-of-fixedlocator-locations-5-usually-from-a-call-to-set
pip install -q matplotlib==3.2.2

import tensorflow_datasets as tfds
import tensorflow as tf

import time
import numpy as np
import matplotlib.pyplot as plt

Setup input pipeline

Use TFDS to load the Portugese-English translation dataset from the TED Talks Open Translation Project.

This dataset contains approximately 50000 training examples, 1100 validation examples, and 2000 test examples.

examples, metadata = tfds.load('ted_hrlr_translate/pt_to_en', with_info=True,
                               as_supervised=True)
train_examples, val_examples = examples['train'], examples['validation']

Downloading and preparing dataset 124.94 MiB (download: 124.94 MiB, generated: Unknown size, total: 124.94 MiB) to /home/kbuilder/tensorflow_datasets/ted_hrlr_translate/pt_to_en/1.0.0...
Dataset ted_hrlr_translate downloaded and prepared to /home/kbuilder/tensorflow_datasets/ted_hrlr_translate/pt_to_en/1.0.0. Subsequent calls will reuse this data.

Create a custom subwords tokenizer from the training dataset. 

tokenizer_en = tfds.deprecated.text.SubwordTextEncoder.build_from_corpus(
    (en.numpy() for pt, en in train_examples), target_vocab_size=2**13)

tokenizer_pt = tfds.deprecated.text.SubwordTextEncoder.build_from_corpus(
    (pt.numpy() for pt, en in train_examples), target_vocab_size=2**13)

sample_string = 'Transformer is awesome.'

tokenized_string = tokenizer_en.encode(sample_string)
print ('Tokenized string is {}'.format(tokenized_string))

original_string = tokenizer_en.decode(tokenized_string)
print ('The original string: {}'.format(original_string))

assert original_string == sample_string

Tokenized string is [7915, 1248, 7946, 7194, 13, 2799, 7877]
The original string: Transformer is awesome.

The tokenizer encodes the string by breaking it into subwords if the word is not in its dictionary.

for ts in tokenized_string:
  print ('{} ----> {}'.format(ts, tokenizer_en.decode([ts])))

7915 ----> T
1248 ----> ran
7946 ----> s
7194 ----> former 
13 ----> is 
2799 ----> awesome
7877 ----> .


BUFFER_SIZE = 20000
BATCH_SIZE = 64

Add a start and end token to the input and target. 

def encode(lang1, lang2):
  lang1 = [tokenizer_pt.vocab_size] + tokenizer_pt.encode(
      lang1.numpy()) + [tokenizer_pt.vocab_size+1]

  lang2 = [tokenizer_en.vocab_size] + tokenizer_en.encode(
      lang2.numpy()) + [tokenizer_en.vocab_size+1]

  return lang1, lang2

You want to use Dataset.map to apply this function to each element of the dataset. Dataset.map runs in graph mode.

  • Graph tensors do not have a value.
  • In graph mode you can only use TensorFlow Ops and functions.

So you can't .map this function directly: You need to wrap it in a tf.py_function. The tf.py_function will pass regular tensors (with a value and a .numpy() method to access it), to the wrapped python function.

def tf_encode(pt, en):
  result_pt, result_en = tf.py_function(encode, [pt, en], [tf.int64, tf.int64])
  result_pt.set_shape([None])
  result_en.set_shape([None])

  return result_pt, result_en

MAX_LENGTH = 40

def filter_max_length(x, y, max_length=MAX_LENGTH):
  return tf.logical_and(tf.size(x) <= max_length,
                        tf.size(y) <= max_length)

train_dataset = train_examples.map(tf_encode)
train_dataset = train_dataset.filter(filter_max_length)
# cache the dataset to memory to get a speedup while reading from it.
train_dataset = train_dataset.cache()
train_dataset = train_dataset.shuffle(BUFFER_SIZE).padded_batch(BATCH_SIZE)
train_dataset = train_dataset.prefetch(tf.data.AUTOTUNE)


val_dataset = val_examples.map(tf_encode)
val_dataset = val_dataset.filter(filter_max_length).padded_batch(BATCH_SIZE)

pt_batch, en_batch = next(iter(val_dataset))
pt_batch, en_batch

(<tf.Tensor: shape=(64, 38), dtype=int64, numpy=
 array([[8214,  342, 3032, ...,    0,    0,    0],
        [8214,   95,  198, ...,    0,    0,    0],
        [8214, 4479, 7990, ...,    0,    0,    0],
        ...,
        [8214,  584,   12, ...,    0,    0,    0],
        [8214,   59, 1548, ...,    0,    0,    0],
        [8214,  118,   34, ...,    0,    0,    0]])>,
 <tf.Tensor: shape=(64, 40), dtype=int64, numpy=
 array([[8087,   98,   25, ...,    0,    0,    0],
        [8087,   12,   20, ...,    0,    0,    0],
        [8087,   12, 5453, ...,    0,    0,    0],
        ...,
        [8087,   18, 2059, ...,    0,    0,    0],
        [8087,   16, 1436, ...,    0,    0,    0],
        [8087,   15,   57, ...,    0,    0,    0]])>)

Positional encoding

Since this model doesn't contain any recurrence or convolution, positional encoding is added to give the model some information about the relative position of the words in the sentence.

The positional encoding vector is added to the embedding vector. Embeddings represent a token in a d-dimensional space where tokens with similar meaning will be closer to each other. But the embeddings do not encode the relative position of words in a sentence. So after adding the positional encoding, words will be closer to each other based on the similarity of their meaning and their position in the sentence, in the d-dimensional space.

See the notebook on positional encoding to learn more about it. The formula for calculating the positional encoding is as follows:

$$\Large{PE_{(pos, 2i)} = sin(pos / 10000^{2i / d_{model} })} $$
$$\Large{PE_{(pos, 2i+1)} = cos(pos / 10000^{2i / d_{model} })} $$ 
def get_angles(pos, i, d_model):
  angle_rates = 1 / np.power(10000, (2 * (i//2)) / np.float32(d_model))
  return pos * angle_rates

def positional_encoding(position, d_model):
  angle_rads = get_angles(np.arange(position)[:, np.newaxis],
                          np.arange(d_model)[np.newaxis, :],
                          d_model)

  # apply sin to even indices in the array; 2i
  angle_rads[:, 0::2] = np.sin(angle_rads[:, 0::2])

  # apply cos to odd indices in the array; 2i+1
  angle_rads[:, 1::2] = np.cos(angle_rads[:, 1::2])

  pos_encoding = angle_rads[np.newaxis, ...]

  return tf.cast(pos_encoding, dtype=tf.float32)

pos_encoding = positional_encoding(50, 512)
print (pos_encoding.shape)

plt.pcolormesh(pos_encoding[0], cmap='RdBu')
plt.xlabel('Depth')
plt.xlim((0, 512))
plt.ylabel('Position')
plt.colorbar()
plt.show()

(1, 50, 512)

png

Masking

Mask all the pad tokens in the batch of sequence. It ensures that the model does not treat padding as the input. The mask indicates where pad value 0 is present: it outputs a 1 at those locations, and a 0 otherwise.

def create_padding_mask(seq):
  seq = tf.cast(tf.math.equal(seq, 0), tf.float32)

  # add extra dimensions to add the padding
  # to the attention logits.
  return seq[:, tf.newaxis, tf.newaxis, :]  # (batch_size, 1, 1, seq_len)

x = tf.constant([[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]])
create_padding_mask(x)

<tf.Tensor: shape=(3, 1, 1, 5), dtype=float32, numpy=
array([[[[0., 0., 1., 1., 0.]]],


       [[[0., 0., 0., 1., 1.]]],


       [[[1., 1., 1., 0., 0.]]]], dtype=float32)>

The look-ahead mask is used to mask the future tokens in a sequence. In other words, the mask indicates which entries should not be used.

This means that to predict the third word, only the first and second word will be used. Similarly to predict the fourth word, only the first, second and the third word will be used and so on.

def create_look_ahead_mask(size):
  mask = 1 - tf.linalg.band_part(tf.ones((size, size)), -1, 0)
  return mask  # (seq_len, seq_len)

x = tf.random.uniform((1, 3))
temp = create_look_ahead_mask(x.shape[1])
temp

<tf.Tensor: shape=(3, 3), dtype=float32, numpy=
array([[0., 1., 1.],
       [0., 0., 1.],
       [0., 0., 0.]], dtype=float32)>

Scaled dot product attention

scaled_dot_product_attention

The attention function used by the transformer takes three inputs: Q (query), K (key), V (value). The equation used to calculate the attention weights is:

$$\Large{Attention(Q, K, V) = softmax_k(\frac{QK^T}{\sqrt{d_k} }) V} $$

The dot-product attention is scaled by a factor of square root of the depth. This is done because for large values of depth, the dot product grows large in magnitude pushing the softmax function where it has small gradients resulting in a very hard softmax.

For example, consider that Q and K have a mean of 0 and variance of 1. Their matrix multiplication will have a mean of 0 and variance of dk. Hence, square root of dk is used for scaling (and not any other number) because the matmul of Q and K should have a mean of 0 and variance of 1, and you get a gentler softmax.

The mask is multiplied with -1e9 (close to negative infinity). This is done because the mask is summed with the scaled matrix multiplication of Q and K and is applied immediately before a softmax. The goal is to zero out these cells, and large negative inputs to softmax are near zero in the output.

def scaled_dot_product_attention(q, k, v, mask):
  """Calculate the attention weights.
  q, k, v must have matching leading dimensions.
  k, v must have matching penultimate dimension, i.e.: seq_len_k = seq_len_v.
  The mask has different shapes depending on its type(padding or look ahead) 
  but it must be broadcastable for addition.

  Args:
    q: query shape == (..., seq_len_q, depth)
    k: key shape == (..., seq_len_k, depth)
    v: value shape == (..., seq_len_v, depth_v)
    mask: Float tensor with shape broadcastable 
          to (..., seq_len_q, seq_len_k). Defaults to None.

  Returns:
    output, attention_weights
  """

  matmul_qk = tf.matmul(q, k, transpose_b=True)  # (..., seq_len_q, seq_len_k)

  # scale matmul_qk
  dk = tf.cast(tf.shape(k)[-1], tf.float32)
  scaled_attention_logits = matmul_qk / tf.math.sqrt(dk)

  # add the mask to the scaled tensor.
  if mask is not None:
    scaled_attention_logits += (mask * -1e9)  

  # softmax is normalized on the last axis (seq_len_k) so that the scores
  # add up to 1.
  attention_weights = tf.nn.softmax(scaled_attention_logits, axis=-1)  # (..., seq_len_q, seq_len_k)

  output = tf.matmul(attention_weights, v)  # (..., seq_len_q, depth_v)

  return output, attention_weights

As the softmax normalization is done on K, its values decide the amount of importance given to Q.

The output represents the multiplication of the attention weights and the V (value) vector. This ensures that the words you want to focus on are kept as-is and the irrelevant words are flushed out.

def print_out(q, k, v):
  temp_out, temp_attn = scaled_dot_product_attention(
      q, k, v, None)
  print ('Attention weights are:')
  print (temp_attn)
  print ('Output is:')
  print (temp_out)

np.set_printoptions(suppress=True)

temp_k = tf.constant([[10,0,0],
                      [0,10,0],
                      [0,0,10],
                      [0,0,10]], dtype=tf.float32)  # (4, 3)

temp_v = tf.constant([[   1,0],
                      [  10,0],
                      [ 100,5],
                      [1000,6]], dtype=tf.float32)  # (4, 2)

# This `query` aligns with the second `key`,
# so the second `value` is returned.
temp_q = tf.constant([[0, 10, 0]], dtype=tf.float32)  # (1, 3)
print_out(temp_q, temp_k, temp_v)

Attention weights are:
tf.Tensor([[0. 1. 0. 0.]], shape=(1, 4), dtype=float32)
Output is:
tf.Tensor([[10.  0.]], shape=(1, 2), dtype=float32)


# This query aligns with a repeated key (third and fourth), 
# so all associated values get averaged.
temp_q = tf.constant([[0, 0, 10]], dtype=tf.float32)  # (1, 3)
print_out(temp_q, temp_k, temp_v)

Attention weights are:
tf.Tensor([[0.  0.  0.5 0.5]], shape=(1, 4), dtype=float32)
Output is:
tf.Tensor([[550.    5.5]], shape=(1, 2), dtype=float32)


# This query aligns equally with the first and second key, 
# so their values get averaged.
temp_q = tf.constant([[10, 10, 0]], dtype=tf.float32)  # (1, 3)
print_out(temp_q, temp_k, temp_v)

Attention weights are:
tf.Tensor([[0.5 0.5 0.  0. ]], shape=(1, 4), dtype=float32)
Output is:
tf.Tensor([[5.5 0. ]], shape=(1, 2), dtype=float32)

Pass all the queries together.

temp_q = tf.constant([[0, 0, 10], [0, 10, 0], [10, 10, 0]], dtype=tf.float32)  # (3, 3)
print_out(temp_q, temp_k, temp_v)

Attention weights are:
tf.Tensor(
[[0.  0.  0.5 0.5]
 [0.  1.  0.  0. ]
 [0.5 0.5 0.  0. ]], shape=(3, 4), dtype=float32)
Output is:
tf.Tensor(
[[550.    5.5]
 [ 10.    0. ]
 [  5.5   0. ]], shape=(3, 2), dtype=float32)

Multi-head attention

multi-head attention

Multi-head attention consists of four parts:

  • Linear layers and split into heads.
  • Scaled dot-product attention.
  • Concatenation of heads.
  • Final linear layer.

Each multi-head attention block gets three inputs; Q (query), K (key), V (value). These are put through linear (Dense) layers and split up into multiple heads.

The scaled_dot_product_attention defined above is applied to each head (broadcasted for efficiency). An appropriate mask must be used in the attention step. The attention output for each head is then concatenated (using tf.transpose, and tf.reshape) and put through a final Dense layer.

Instead of one single attention head, Q, K, and V are split into multiple heads because it allows the model to jointly attend to information at different positions from different representational spaces. After the split each head has a reduced dimensionality, so the total computation cost is the same as a single head attention with full dimensionality.

class MultiHeadAttention(tf.keras.layers.Layer):
  def __init__(self, d_model, num_heads):
    super(MultiHeadAttention, self).__init__()
    self.num_heads = num_heads
    self.d_model = d_model

    assert d_model % self.num_heads == 0

    self.depth = d_model // self.num_heads

    self.wq = tf.keras.layers.Dense(d_model)
    self.wk = tf.keras.layers.Dense(d_model)
    self.wv = tf.keras.layers.Dense(d_model)

    self.dense = tf.keras.layers.Dense(d_model)

  def split_heads(self, x, batch_size):
    """Split the last dimension into (num_heads, depth).
    Transpose the result such that the shape is (batch_size, num_heads, seq_len, depth)
    """
    x = tf.reshape(x, (batch_size, -1, self.num_heads, self.depth))
    return tf.transpose(x, perm=[0, 2, 1, 3])

  def call(self, v, k, q, mask):
    batch_size = tf.shape(q)[0]

    q = self.wq(q)  # (batch_size, seq_len, d_model)
    k = self.wk(k)  # (batch_size, seq_len, d_model)
    v = self.wv(v)  # (batch_size, seq_len, d_model)

    q = self.split_heads(q, batch_size)  # (batch_size, num_heads, seq_len_q, depth)
    k = self.split_heads(k, batch_size)  # (batch_size, num_heads, seq_len_k, depth)
    v = self.split_heads(v, batch_size)  # (batch_size, num_heads, seq_len_v, depth)

    # scaled_attention.shape == (batch_size, num_heads, seq_len_q, depth)
    # attention_weights.shape == (batch_size, num_heads, seq_len_q, seq_len_k)
    scaled_attention, attention_weights = scaled_dot_product_attention(
        q, k, v, mask)

    scaled_attention = tf.transpose(scaled_attention, perm=[0, 2, 1, 3])  # (batch_size, seq_len_q, num_heads, depth)

    concat_attention = tf.reshape(scaled_attention, 
                                  (batch_size, -1, self.d_model))  # (batch_size, seq_len_q, d_model)

    output = self.dense(concat_attention)  # (batch_size, seq_len_q, d_model)

    return output, attention_weights

Create a MultiHeadAttention layer to try out. At each location in the sequence, y, the MultiHeadAttention runs all 8 attention heads across all other locations in the sequence, returning a new vector of the same length at each location.

temp_mha = MultiHeadAttention(d_model=512, num_heads=8)
y = tf.random.uniform((1, 60, 512))  # (batch_size, encoder_sequence, d_model)
out, attn = temp_mha(y, k=y, q=y, mask=None)
out.shape, attn.shape

(TensorShape([1, 60, 512]), TensorShape([1, 8, 60, 60]))

Point wise feed forward network

Point wise feed forward network consists of two fully-connected layers with a ReLU activation in between.

def point_wise_feed_forward_network(d_model, dff):
  return tf.keras.Sequential([
      tf.keras.layers.Dense(dff, activation='relu'),  # (batch_size, seq_len, dff)
      tf.keras.layers.Dense(d_model)  # (batch_size, seq_len, d_model)
  ])

sample_ffn = point_wise_feed_forward_network(512, 2048)
sample_ffn(tf.random.uniform((64, 50, 512))).shape

TensorShape([64, 50, 512])

Encoder and decoder

transformer

The transformer model follows the same general pattern as a standard sequence to sequence with attention model.

  • The input sentence is passed through N encoder layers that generates an output for each word/token in the sequence.
  • The decoder attends on the encoder's output and its own input (self-attention) to predict the next word.

Encoder layer

Each encoder layer consists of sublayers:

  1. Multi-head attention (with padding mask)
  2. Point wise feed forward networks.

Each of these sublayers has a residual connection around it followed by a layer normalization. Residual connections help in avoiding the vanishing gradient problem in deep networks.

The output of each sublayer is LayerNorm(x + Sublayer(x)). The normalization is done on the d_model (last) axis. There are N encoder layers in the transformer.

class EncoderLayer(tf.keras.layers.Layer):
  def __init__(self, d_model, num_heads, dff, rate=0.1):
    super(EncoderLayer, self).__init__()

    self.mha = MultiHeadAttention(d_model, num_heads)
    self.ffn = point_wise_feed_forward_network(d_model, dff)

    self.layernorm1 = tf.keras.layers.LayerNormalization(epsilon=1e-6)
    self.layernorm2 = tf.keras.layers.LayerNormalization(epsilon=1e-6)

    self.dropout1 = tf.keras.layers.Dropout(rate)
    self.dropout2 = tf.keras.layers.Dropout(rate)

  def call(self, x, training, mask):

    attn_output, _ = self.mha(x, x, x, mask)  # (batch_size, input_seq_len, d_model)
    attn_output = self.dropout1(attn_output, training=training)
    out1 = self.layernorm1(x + attn_output)  # (batch_size, input_seq_len, d_model)

    ffn_output = self.ffn(out1)  # (batch_size, input_seq_len, d_model)
    ffn_output = self.dropout2(ffn_output, training=training)
    out2 = self.layernorm2(out1 + ffn_output)  # (batch_size, input_seq_len, d_model)

    return out2

sample_encoder_layer = EncoderLayer(512, 8, 2048)

sample_encoder_layer_output = sample_encoder_layer(
    tf.random.uniform((64, 43, 512)), False, None)

sample_encoder_layer_output.shape  # (batch_size, input_seq_len, d_model)

TensorShape([64, 43, 512])

Decoder layer

Each decoder layer consists of sublayers:

  1. Masked multi-head attention (with look ahead mask and padding mask)
  2. Multi-head attention (with padding mask). V (value) and K (key) receive the encoder output as inputs. Q (query) receives the output from the masked multi-head attention sublayer.
  3. Point wise feed forward networks

Each of these sublayers has a residual connection around it followed by a layer normalization. The output of each sublayer is LayerNorm(x + Sublayer(x)). The normalization is done on the d_model (last) axis.

There are N decoder layers in the transformer.

As Q receives the output from decoder's first attention block, and K receives the encoder output, the attention weights represent the importance given to the decoder's input based on the encoder's output. In other words, the decoder predicts the next word by looking at the encoder output and self-attending to its own output. See the demonstration above in the scaled dot product attention section.

class DecoderLayer(tf.keras.layers.Layer):
  def __init__(self, d_model, num_heads, dff, rate=0.1):
    super(DecoderLayer, self).__init__()

    self.mha1 = MultiHeadAttention(d_model, num_heads)
    self.mha2 = MultiHeadAttention(d_model, num_heads)

    self.ffn = point_wise_feed_forward_network(d_model, dff)

    self.layernorm1 = tf.keras.layers.LayerNormalization(epsilon=1e-6)
    self.layernorm2 = tf.keras.layers.LayerNormalization(epsilon=1e-6)
    self.layernorm3 = tf.keras.layers.LayerNormalization(epsilon=1e-6)

    self.dropout1 = tf.keras.layers.Dropout(rate)
    self.dropout2 = tf.keras.layers.Dropout(rate)
    self.dropout3 = tf.keras.layers.Dropout(rate)


  def call(self, x, enc_output, training, 
           look_ahead_mask, padding_mask):
    # enc_output.shape == (batch_size, input_seq_len, d_model)

    attn1, attn_weights_block1 = self.mha1(x, x, x, look_ahead_mask)  # (batch_size, target_seq_len, d_model)
    attn1 = self.dropout1(attn1, training=training)
    out1 = self.layernorm1(attn1 + x)

    attn2, attn_weights_block2 = self.mha2(
        enc_output, enc_output, out1, padding_mask)  # (batch_size, target_seq_len, d_model)
    attn2 = self.dropout2(attn2, training=training)
    out2 = self.layernorm2(attn2 + out1)  # (batch_size, target_seq_len, d_model)

    ffn_output = self.ffn(out2)  # (batch_size, target_seq_len, d_model)
    ffn_output = self.dropout3(ffn_output, training=training)
    out3 = self.layernorm3(ffn_output + out2)  # (batch_size, target_seq_len, d_model)

    return out3, attn_weights_block1, attn_weights_block2

sample_decoder_layer = DecoderLayer(512, 8, 2048)

sample_decoder_layer_output, _, _ = sample_decoder_layer(
    tf.random.uniform((64, 50, 512)), sample_encoder_layer_output, 
    False, None, None)

sample_decoder_layer_output.shape  # (batch_size, target_seq_len, d_model)

TensorShape([64, 50, 512])

Encoder

The Encoder consists of:

  1. Input Embedding
  2. Positional Encoding
  3. N encoder layers

The input is put through an embedding which is summed with the positional encoding. The output of this summation is the input to the encoder layers. The output of the encoder is the input to the decoder.

class Encoder(tf.keras.layers.Layer):
  def __init__(self, num_layers, d_model, num_heads, dff, input_vocab_size,
               maximum_position_encoding, rate=0.1):
    super(Encoder, self).__init__()

    self.d_model = d_model
    self.num_layers = num_layers

    self.embedding = tf.keras.layers.Embedding(input_vocab_size, d_model)
    self.pos_encoding = positional_encoding(maximum_position_encoding, 
                                            self.d_model)


    self.enc_layers = [EncoderLayer(d_model, num_heads, dff, rate) 
                       for _ in range(num_layers)]

    self.dropout = tf.keras.layers.Dropout(rate)

  def call(self, x, training, mask):

    seq_len = tf.shape(x)[1]

    # adding embedding and position encoding.
    x = self.embedding(x)  # (batch_size, input_seq_len, d_model)
    x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32))
    x += self.pos_encoding[:, :seq_len, :]

    x = self.dropout(x, training=training)

    for i in range(self.num_layers):
      x = self.enc_layers[i](x, training, mask)

    return x  # (batch_size, input_seq_len, d_model)

sample_encoder = Encoder(num_layers=2, d_model=512, num_heads=8, 
                         dff=2048, input_vocab_size=8500,
                         maximum_position_encoding=10000)
temp_input = tf.random.uniform((64, 62), dtype=tf.int64, minval=0, maxval=200)

sample_encoder_output = sample_encoder(temp_input, training=False, mask=None)

print (sample_encoder_output.shape)  # (batch_size, input_seq_len, d_model)

(64, 62, 512)

Decoder

The Decoder consists of:

  1. Output Embedding
  2. Positional Encoding
  3. N decoder layers

The target is put through an embedding which is summed with the positional encoding. The output of this summation is the input to the decoder layers. The output of the decoder is the input to the final linear layer.

class Decoder(tf.keras.layers.Layer):
  def __init__(self, num_layers, d_model, num_heads, dff, target_vocab_size,
               maximum_position_encoding, rate=0.1):
    super(Decoder, self).__init__()

    self.d_model = d_model
    self.num_layers = num_layers

    self.embedding = tf.keras.layers.Embedding(target_vocab_size, d_model)
    self.pos_encoding = positional_encoding(maximum_position_encoding, d_model)

    self.dec_layers = [DecoderLayer(d_model, num_heads, dff, rate) 
                       for _ in range(num_layers)]
    self.dropout = tf.keras.layers.Dropout(rate)

  def call(self, x, enc_output, training, 
           look_ahead_mask, padding_mask):

    seq_len = tf.shape(x)[1]
    attention_weights = {}

    x = self.embedding(x)  # (batch_size, target_seq_len, d_model)
    x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32))
    x += self.pos_encoding[:, :seq_len, :]

    x = self.dropout(x, training=training)

    for i in range(self.num_layers):
      x, block1, block2 = self.dec_layers[i](x, enc_output, training,
                                             look_ahead_mask, padding_mask)

      attention_weights['decoder_layer{}_block1'.format(i+1)] = block1
      attention_weights['decoder_layer{}_block2'.format(i+1)] = block2

    # x.shape == (batch_size, target_seq_len, d_model)
    return x, attention_weights

sample_decoder = Decoder(num_layers=2, d_model=512, num_heads=8, 
                         dff=2048, target_vocab_size=8000,
                         maximum_position_encoding=5000)
temp_input = tf.random.uniform((64, 26), dtype=tf.int64, minval=0, maxval=200)

output, attn = sample_decoder(temp_input, 
                              enc_output=sample_encoder_output, 
                              training=False,
                              look_ahead_mask=None, 
                              padding_mask=None)

output.shape, attn['decoder_layer2_block2'].shape

(TensorShape([64, 26, 512]), TensorShape([64, 8, 26, 62]))

Create the Transformer

Transformer consists of the encoder, decoder and a final linear layer. The output of the decoder is the input to the linear layer and its output is returned.

class Transformer(tf.keras.Model):
  def __init__(self, num_layers, d_model, num_heads, dff, input_vocab_size, 
               target_vocab_size, pe_input, pe_target, rate=0.1):
    super(Transformer, self).__init__()

    self.encoder = Encoder(num_layers, d_model, num_heads, dff, 
                           input_vocab_size, pe_input, rate)

    self.decoder = Decoder(num_layers, d_model, num_heads, dff, 
                           target_vocab_size, pe_target, rate)

    self.final_layer = tf.keras.layers.Dense(target_vocab_size)

  def call(self, inp, tar, training, enc_padding_mask, 
           look_ahead_mask, dec_padding_mask):

    enc_output = self.encoder(inp, training, enc_padding_mask)  # (batch_size, inp_seq_len, d_model)

    # dec_output.shape == (batch_size, tar_seq_len, d_model)
    dec_output, attention_weights = self.decoder(
        tar, enc_output, training, look_ahead_mask, dec_padding_mask)

    final_output = self.final_layer(dec_output)  # (batch_size, tar_seq_len, target_vocab_size)

    return final_output, attention_weights

sample_transformer = Transformer(
    num_layers=2, d_model=512, num_heads=8, dff=2048, 
    input_vocab_size=8500, target_vocab_size=8000, 
    pe_input=10000, pe_target=6000)

temp_input = tf.random.uniform((64, 38), dtype=tf.int64, minval=0, maxval=200)
temp_target = tf.random.uniform((64, 36), dtype=tf.int64, minval=0, maxval=200)

fn_out, _ = sample_transformer(temp_input, temp_target, training=False, 
                               enc_padding_mask=None, 
                               look_ahead_mask=None,
                               dec_padding_mask=None)

fn_out.shape  # (batch_size, tar_seq_len, target_vocab_size)

TensorShape([64, 36, 8000])

Set hyperparameters

To keep this example small and relatively fast, the values for num_layers, d_model, and dff have been reduced.

The values used in the base model of transformer were; num_layers=6, d_model = 512, dff = 2048. See the paper for all the other versions of the transformer.

num_layers = 4
d_model = 128
dff = 512
num_heads = 8

input_vocab_size = tokenizer_pt.vocab_size + 2
target_vocab_size = tokenizer_en.vocab_size + 2
dropout_rate = 0.1

Optimizer

Use the Adam optimizer with a custom learning rate scheduler according to the formula in the paper.

$$\Large{lrate = d_{model}^{-0.5} * min(step{\_}num^{-0.5}, step{\_}num * warmup{\_}steps^{-1.5})}$$ 
class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):
  def __init__(self, d_model, warmup_steps=4000):
    super(CustomSchedule, self).__init__()

    self.d_model = d_model
    self.d_model = tf.cast(self.d_model, tf.float32)

    self.warmup_steps = warmup_steps

  def __call__(self, step):
    arg1 = tf.math.rsqrt(step)
    arg2 = step * (self.warmup_steps ** -1.5)

    return tf.math.rsqrt(self.d_model) * tf.math.minimum(arg1, arg2)

learning_rate = CustomSchedule(d_model)

optimizer = tf.keras.optimizers.Adam(learning_rate, beta_1=0.9, beta_2=0.98, 
                                     epsilon=1e-9)

temp_learning_rate_schedule = CustomSchedule(d_model)

plt.plot(temp_learning_rate_schedule(tf.range(40000, dtype=tf.float32)))
plt.ylabel("Learning Rate")
plt.xlabel("Train Step")

Text(0.5, 0, 'Train Step')

png

Loss and metrics

Since the target sequences are padded, it is important to apply a padding mask when calculating the loss.

loss_object = tf.keras.losses.SparseCategoricalCrossentropy(
    from_logits=True, reduction='none')

def loss_function(real, pred):
  mask = tf.math.logical_not(tf.math.equal(real, 0))
  loss_ = loss_object(real, pred)

  mask = tf.cast(mask, dtype=loss_.dtype)
  loss_ *= mask

  return tf.reduce_sum(loss_)/tf.reduce_sum(mask)


def accuracy_function(real, pred):
  accuracies = tf.equal(real, tf.argmax(pred, axis=2))

  mask = tf.math.logical_not(tf.math.equal(real, 0))
  accuracies = tf.math.logical_and(mask, accuracies)

  accuracies = tf.cast(accuracies, dtype=tf.float32)
  mask = tf.cast(mask, dtype=tf.float32)
  return tf.reduce_sum(accuracies)/tf.reduce_sum(mask)

train_loss = tf.keras.metrics.Mean(name='train_loss')
train_accuracy = tf.keras.metrics.Mean(name='train_accuracy')

Training and checkpointing

transformer = Transformer(num_layers, d_model, num_heads, dff,
                          input_vocab_size, target_vocab_size, 
                          pe_input=input_vocab_size, 
                          pe_target=target_vocab_size,
                          rate=dropout_rate)

def create_masks(inp, tar):
  # Encoder padding mask
  enc_padding_mask = create_padding_mask(inp)

  # Used in the 2nd attention block in the decoder.
  # This padding mask is used to mask the encoder outputs.
  dec_padding_mask = create_padding_mask(inp)

  # Used in the 1st attention block in the decoder.
  # It is used to pad and mask future tokens in the input received by 
  # the decoder.
  look_ahead_mask = create_look_ahead_mask(tf.shape(tar)[1])
  dec_target_padding_mask = create_padding_mask(tar)
  combined_mask = tf.maximum(dec_target_padding_mask, look_ahead_mask)

  return enc_padding_mask, combined_mask, dec_padding_mask

Create the checkpoint path and the checkpoint manager. This will be used to save checkpoints every n epochs.

checkpoint_path = "./checkpoints/train"

ckpt = tf.train.Checkpoint(transformer=transformer,
                           optimizer=optimizer)

ckpt_manager = tf.train.CheckpointManager(ckpt, checkpoint_path, max_to_keep=5)

# if a checkpoint exists, restore the latest checkpoint.
if ckpt_manager.latest_checkpoint:
  ckpt.restore(ckpt_manager.latest_checkpoint)
  print ('Latest checkpoint restored!!')

The target is divided into tar_inp and tar_real. tar_inp is passed as an input to the decoder. tar_real is that same input shifted by 1: At each location in tar_input, tar_real contains the next token that should be predicted.

For example, sentence = "SOS A lion in the jungle is sleeping EOS"

tar_inp = "SOS A lion in the jungle is sleeping"

tar_real = "A lion in the jungle is sleeping EOS"

The transformer is an auto-regressive model: it makes predictions one part at a time, and uses its output so far to decide what to do next.

During training this example uses teacher-forcing (like in the text generation tutorial). Teacher forcing is passing the true output to the next time step regardless of what the model predicts at the current time step.

As the transformer predicts each word, self-attention allows it to look at the previous words in the input sequence to better predict the next word.

To prevent the model from peeking at the expected output the model uses a look-ahead mask.

EPOCHS = 20

# The @tf.function trace-compiles train_step into a TF graph for faster
# execution. The function specializes to the precise shape of the argument
# tensors. To avoid re-tracing due to the variable sequence lengths or variable
# batch sizes (the last batch is smaller), use input_signature to specify
# more generic shapes.

train_step_signature = [
    tf.TensorSpec(shape=(None, None), dtype=tf.int64),
    tf.TensorSpec(shape=(None, None), dtype=tf.int64),
]

@tf.function(input_signature=train_step_signature)
def train_step(inp, tar):
  tar_inp = tar[:, :-1]
  tar_real = tar[:, 1:]

  enc_padding_mask, combined_mask, dec_padding_mask = create_masks(inp, tar_inp)

  with tf.GradientTape() as tape:
    predictions, _ = transformer(inp, tar_inp, 
                                 True, 
                                 enc_padding_mask, 
                                 combined_mask, 
                                 dec_padding_mask)
    loss = loss_function(tar_real, predictions)

  gradients = tape.gradient(loss, transformer.trainable_variables)    
  optimizer.apply_gradients(zip(gradients, transformer.trainable_variables))

  train_loss(loss)
  train_accuracy(accuracy_function(tar_real, predictions))

Portuguese is used as the input language and English is the target language.

for epoch in range(EPOCHS):
  start = time.time()

  train_loss.reset_states()
  train_accuracy.reset_states()

  # inp -> portuguese, tar -> english
  for (batch, (inp, tar)) in enumerate(train_dataset):
    train_step(inp, tar)

    if batch % 50 == 0:
      print ('Epoch {} Batch {} Loss {:.4f} Accuracy {:.4f}'.format(
          epoch + 1, batch, train_loss.result(), train_accuracy.result()))

  if (epoch + 1) % 5 == 0:
    ckpt_save_path = ckpt_manager.save()
    print ('Saving checkpoint for epoch {} at {}'.format(epoch+1,
                                                         ckpt_save_path))

  print ('Epoch {} Loss {:.4f} Accuracy {:.4f}'.format(epoch + 1, 
                                                train_loss.result(), 
                                                train_accuracy.result()))

  print ('Time taken for 1 epoch: {} secs\n'.format(time.time() - start))

Epoch 1 Batch 0 Loss 9.0325 Accuracy 0.0000
Epoch 1 Batch 50 Loss 8.9631 Accuracy 0.0029
Epoch 1 Batch 100 Loss 8.8625 Accuracy 0.0285
Epoch 1 Batch 150 Loss 8.7535 Accuracy 0.0396
Epoch 1 Batch 200 Loss 8.6247 Accuracy 0.0465
Epoch 1 Batch 250 Loss 8.4712 Accuracy 0.0555
Epoch 1 Batch 300 Loss 8.2956 Accuracy 0.0660
Epoch 1 Batch 350 Loss 8.1095 Accuracy 0.0753
Epoch 1 Batch 400 Loss 7.9295 Accuracy 0.0823
Epoch 1 Batch 450 Loss 7.7644 Accuracy 0.0879
Epoch 1 Batch 500 Loss 7.6171 Accuracy 0.0927
Epoch 1 Batch 550 Loss 7.4840 Accuracy 0.0983
Epoch 1 Batch 600 Loss 7.3578 Accuracy 0.1048
Epoch 1 Batch 650 Loss 7.2364 Accuracy 0.1117
Epoch 1 Batch 700 Loss 7.1232 Accuracy 0.1185
Epoch 1 Loss 7.1188 Accuracy 0.1187
Time taken for 1 epoch: 54.0901415348053 secs

Epoch 2 Batch 0 Loss 5.5828 Accuracy 0.1969
Epoch 2 Batch 50 Loss 5.4646 Accuracy 0.2218
Epoch 2 Batch 100 Loss 5.4122 Accuracy 0.2249
Epoch 2 Batch 150 Loss 5.3780 Accuracy 0.2278
Epoch 2 Batch 200 Loss 5.3342 Accuracy 0.2315
Epoch 2 Batch 250 Loss 5.2924 Accuracy 0.2352
Epoch 2 Batch 300 Loss 5.2508 Accuracy 0.2393
Epoch 2 Batch 350 Loss 5.2157 Accuracy 0.2424
Epoch 2 Batch 400 Loss 5.1816 Accuracy 0.2459
Epoch 2 Batch 450 Loss 5.1515 Accuracy 0.2492
Epoch 2 Batch 500 Loss 5.1203 Accuracy 0.2523
Epoch 2 Batch 550 Loss 5.0920 Accuracy 0.2552
Epoch 2 Batch 600 Loss 5.0661 Accuracy 0.2579
Epoch 2 Batch 650 Loss 5.0418 Accuracy 0.2604
Epoch 2 Batch 700 Loss 5.0186 Accuracy 0.2627
Epoch 2 Loss 5.0179 Accuracy 0.2628
Time taken for 1 epoch: 30.69964909553528 secs

Epoch 3 Batch 0 Loss 4.6354 Accuracy 0.3023
Epoch 3 Batch 50 Loss 4.6234 Accuracy 0.3012
Epoch 3 Batch 100 Loss 4.6082 Accuracy 0.3021
Epoch 3 Batch 150 Loss 4.6073 Accuracy 0.3030
Epoch 3 Batch 200 Loss 4.5980 Accuracy 0.3043
Epoch 3 Batch 250 Loss 4.5887 Accuracy 0.3054
Epoch 3 Batch 300 Loss 4.5753 Accuracy 0.3069
Epoch 3 Batch 350 Loss 4.5617 Accuracy 0.3084
Epoch 3 Batch 400 Loss 4.5476 Accuracy 0.3099
Epoch 3 Batch 450 Loss 4.5342 Accuracy 0.3112
Epoch 3 Batch 500 Loss 4.5243 Accuracy 0.3122
Epoch 3 Batch 550 Loss 4.5141 Accuracy 0.3134
Epoch 3 Batch 600 Loss 4.5010 Accuracy 0.3148
Epoch 3 Batch 650 Loss 4.4917 Accuracy 0.3158
Epoch 3 Batch 700 Loss 4.4797 Accuracy 0.3170
Epoch 3 Loss 4.4795 Accuracy 0.3170
Time taken for 1 epoch: 30.95038342475891 secs

Epoch 4 Batch 0 Loss 3.9974 Accuracy 0.3544
Epoch 4 Batch 50 Loss 4.2098 Accuracy 0.3431
Epoch 4 Batch 100 Loss 4.1827 Accuracy 0.3466
Epoch 4 Batch 150 Loss 4.1836 Accuracy 0.3471
Epoch 4 Batch 200 Loss 4.1799 Accuracy 0.3473
Epoch 4 Batch 250 Loss 4.1675 Accuracy 0.3486
Epoch 4 Batch 300 Loss 4.1581 Accuracy 0.3501
Epoch 4 Batch 350 Loss 4.1447 Accuracy 0.3516
Epoch 4 Batch 400 Loss 4.1289 Accuracy 0.3538
Epoch 4 Batch 450 Loss 4.1132 Accuracy 0.3559
Epoch 4 Batch 500 Loss 4.0977 Accuracy 0.3581
Epoch 4 Batch 550 Loss 4.0811 Accuracy 0.3601
Epoch 4 Batch 600 Loss 4.0648 Accuracy 0.3623
Epoch 4 Batch 650 Loss 4.0508 Accuracy 0.3644
Epoch 4 Batch 700 Loss 4.0378 Accuracy 0.3662
Epoch 4 Loss 4.0372 Accuracy 0.3663
Time taken for 1 epoch: 30.76306676864624 secs

Epoch 5 Batch 0 Loss 3.7628 Accuracy 0.3862
Epoch 5 Batch 50 Loss 3.7014 Accuracy 0.4030
Epoch 5 Batch 100 Loss 3.6951 Accuracy 0.4051
Epoch 5 Batch 150 Loss 3.6715 Accuracy 0.4084
Epoch 5 Batch 200 Loss 3.6619 Accuracy 0.4105
Epoch 5 Batch 250 Loss 3.6462 Accuracy 0.4131
Epoch 5 Batch 300 Loss 3.6403 Accuracy 0.4145
Epoch 5 Batch 350 Loss 3.6238 Accuracy 0.4168
Epoch 5 Batch 400 Loss 3.6115 Accuracy 0.4186
Epoch 5 Batch 450 Loss 3.5990 Accuracy 0.4200
Epoch 5 Batch 500 Loss 3.5902 Accuracy 0.4215
Epoch 5 Batch 550 Loss 3.5806 Accuracy 0.4228
Epoch 5 Batch 600 Loss 3.5705 Accuracy 0.4240
Epoch 5 Batch 650 Loss 3.5612 Accuracy 0.4252
Epoch 5 Batch 700 Loss 3.5494 Accuracy 0.4267
Saving checkpoint for epoch 5 at ./checkpoints/train/ckpt-1
Epoch 5 Loss 3.5493 Accuracy 0.4267
Time taken for 1 epoch: 31.12636709213257 secs

Epoch 6 Batch 0 Loss 3.2444 Accuracy 0.4605
Epoch 6 Batch 50 Loss 3.2155 Accuracy 0.4619
Epoch 6 Batch 100 Loss 3.2090 Accuracy 0.4631
Epoch 6 Batch 150 Loss 3.2024 Accuracy 0.4633
Epoch 6 Batch 200 Loss 3.1966 Accuracy 0.4648
Epoch 6 Batch 250 Loss 3.1929 Accuracy 0.4655
Epoch 6 Batch 300 Loss 3.1881 Accuracy 0.4664
Epoch 6 Batch 350 Loss 3.1832 Accuracy 0.4674
Epoch 6 Batch 400 Loss 3.1761 Accuracy 0.4685
Epoch 6 Batch 450 Loss 3.1699 Accuracy 0.4688
Epoch 6 Batch 500 Loss 3.1650 Accuracy 0.4696
Epoch 6 Batch 550 Loss 3.1592 Accuracy 0.4705
Epoch 6 Batch 600 Loss 3.1530 Accuracy 0.4717
Epoch 6 Batch 650 Loss 3.1471 Accuracy 0.4723
Epoch 6 Batch 700 Loss 3.1409 Accuracy 0.4731
Epoch 6 Loss 3.1406 Accuracy 0.4731
Time taken for 1 epoch: 30.89075517654419 secs

Epoch 7 Batch 0 Loss 2.7938 Accuracy 0.5259
Epoch 7 Batch 50 Loss 2.8191 Accuracy 0.5065
Epoch 7 Batch 100 Loss 2.8050 Accuracy 0.5093
Epoch 7 Batch 150 Loss 2.8151 Accuracy 0.5085
Epoch 7 Batch 200 Loss 2.8113 Accuracy 0.5093
Epoch 7 Batch 250 Loss 2.8064 Accuracy 0.5101
Epoch 7 Batch 300 Loss 2.8001 Accuracy 0.5109
Epoch 7 Batch 350 Loss 2.7923 Accuracy 0.5123
Epoch 7 Batch 400 Loss 2.7791 Accuracy 0.5138
Epoch 7 Batch 450 Loss 2.7733 Accuracy 0.5146
Epoch 7 Batch 500 Loss 2.7688 Accuracy 0.5154
Epoch 7 Batch 550 Loss 2.7602 Accuracy 0.5166
Epoch 7 Batch 600 Loss 2.7542 Accuracy 0.5175
Epoch 7 Batch 650 Loss 2.7497 Accuracy 0.5184
Epoch 7 Batch 700 Loss 2.7439 Accuracy 0.5193
Epoch 7 Loss 2.7436 Accuracy 0.5194
Time taken for 1 epoch: 31.79271173477173 secs

Epoch 8 Batch 0 Loss 2.5171 Accuracy 0.5501
Epoch 8 Batch 50 Loss 2.4147 Accuracy 0.5562
Epoch 8 Batch 100 Loss 2.4221 Accuracy 0.5555
Epoch 8 Batch 150 Loss 2.4131 Accuracy 0.5580
Epoch 8 Batch 200 Loss 2.4173 Accuracy 0.5570
Epoch 8 Batch 250 Loss 2.4214 Accuracy 0.5569
Epoch 8 Batch 300 Loss 2.4305 Accuracy 0.5555
Epoch 8 Batch 350 Loss 2.4305 Accuracy 0.5555
Epoch 8 Batch 400 Loss 2.4273 Accuracy 0.5560
Epoch 8 Batch 450 Loss 2.4261 Accuracy 0.5563
Epoch 8 Batch 500 Loss 2.4245 Accuracy 0.5567
Epoch 8 Batch 550 Loss 2.4233 Accuracy 0.5572
Epoch 8 Batch 600 Loss 2.4182 Accuracy 0.5582
Epoch 8 Batch 650 Loss 2.4162 Accuracy 0.5585
Epoch 8 Batch 700 Loss 2.4147 Accuracy 0.5589
Epoch 8 Loss 2.4139 Accuracy 0.5590
Time taken for 1 epoch: 31.114221572875977 secs

Epoch 9 Batch 0 Loss 2.0796 Accuracy 0.6034
Epoch 9 Batch 50 Loss 2.1062 Accuracy 0.5978
Epoch 9 Batch 100 Loss 2.1268 Accuracy 0.5934
Epoch 9 Batch 150 Loss 2.1465 Accuracy 0.5914
Epoch 9 Batch 200 Loss 2.1525 Accuracy 0.5908
Epoch 9 Batch 250 Loss 2.1578 Accuracy 0.5902
Epoch 9 Batch 300 Loss 2.1590 Accuracy 0.5899
Epoch 9 Batch 350 Loss 2.1592 Accuracy 0.5898
Epoch 9 Batch 400 Loss 2.1598 Accuracy 0.5899
Epoch 9 Batch 450 Loss 2.1614 Accuracy 0.5898
Epoch 9 Batch 500 Loss 2.1632 Accuracy 0.5897
Epoch 9 Batch 550 Loss 2.1625 Accuracy 0.5899
Epoch 9 Batch 600 Loss 2.1655 Accuracy 0.5897
Epoch 9 Batch 650 Loss 2.1659 Accuracy 0.5896
Epoch 9 Batch 700 Loss 2.1671 Accuracy 0.5896
Epoch 9 Loss 2.1671 Accuracy 0.5896
Time taken for 1 epoch: 31.18158221244812 secs

Epoch 10 Batch 0 Loss 1.8438 Accuracy 0.6431
Epoch 10 Batch 50 Loss 1.9301 Accuracy 0.6210
Epoch 10 Batch 100 Loss 1.9432 Accuracy 0.6182
Epoch 10 Batch 150 Loss 1.9503 Accuracy 0.6175
Epoch 10 Batch 200 Loss 1.9521 Accuracy 0.6180
Epoch 10 Batch 250 Loss 1.9540 Accuracy 0.6173
Epoch 10 Batch 300 Loss 1.9551 Accuracy 0.6174
Epoch 10 Batch 350 Loss 1.9615 Accuracy 0.6167
Epoch 10 Batch 400 Loss 1.9619 Accuracy 0.6165
Epoch 10 Batch 450 Loss 1.9636 Accuracy 0.6167
Epoch 10 Batch 500 Loss 1.9685 Accuracy 0.6160
Epoch 10 Batch 550 Loss 1.9700 Accuracy 0.6159
Epoch 10 Batch 600 Loss 1.9750 Accuracy 0.6153
Epoch 10 Batch 650 Loss 1.9787 Accuracy 0.6150
Epoch 10 Batch 700 Loss 1.9803 Accuracy 0.6147
Saving checkpoint for epoch 10 at ./checkpoints/train/ckpt-2
Epoch 10 Loss 1.9804 Accuracy 0.6147
Time taken for 1 epoch: 31.05565071105957 secs

Epoch 11 Batch 0 Loss 1.6855 Accuracy 0.6649
Epoch 11 Batch 50 Loss 1.7798 Accuracy 0.6409
Epoch 11 Batch 100 Loss 1.7849 Accuracy 0.6394
Epoch 11 Batch 150 Loss 1.7887 Accuracy 0.6387
Epoch 11 Batch 200 Loss 1.7986 Accuracy 0.6376
Epoch 11 Batch 250 Loss 1.8039 Accuracy 0.6371
Epoch 11 Batch 300 Loss 1.8087 Accuracy 0.6361
Epoch 11 Batch 350 Loss 1.8122 Accuracy 0.6354
Epoch 11 Batch 400 Loss 1.8157 Accuracy 0.6352
Epoch 11 Batch 450 Loss 1.8157 Accuracy 0.6352
Epoch 11 Batch 500 Loss 1.8214 Accuracy 0.6344
Epoch 11 Batch 550 Loss 1.8249 Accuracy 0.6340
Epoch 11 Batch 600 Loss 1.8275 Accuracy 0.6339
Epoch 11 Batch 650 Loss 1.8327 Accuracy 0.6333
Epoch 11 Batch 700 Loss 1.8357 Accuracy 0.6329
Epoch 11 Loss 1.8357 Accuracy 0.6329
Time taken for 1 epoch: 31.027227878570557 secs

Epoch 12 Batch 0 Loss 1.6143 Accuracy 0.6664
Epoch 12 Batch 50 Loss 1.6444 Accuracy 0.6597
Epoch 12 Batch 100 Loss 1.6483 Accuracy 0.6591
Epoch 12 Batch 150 Loss 1.6596 Accuracy 0.6571
Epoch 12 Batch 200 Loss 1.6646 Accuracy 0.6567
Epoch 12 Batch 250 Loss 1.6728 Accuracy 0.6555
Epoch 12 Batch 300 Loss 1.6742 Accuracy 0.6554
Epoch 12 Batch 350 Loss 1.6771 Accuracy 0.6547
Epoch 12 Batch 400 Loss 1.6812 Accuracy 0.6543
Epoch 12 Batch 450 Loss 1.6868 Accuracy 0.6532
Epoch 12 Batch 500 Loss 1.6895 Accuracy 0.6529
Epoch 12 Batch 550 Loss 1.6949 Accuracy 0.6522
Epoch 12 Batch 600 Loss 1.7010 Accuracy 0.6515
Epoch 12 Batch 650 Loss 1.7086 Accuracy 0.6504
Epoch 12 Batch 700 Loss 1.7139 Accuracy 0.6499
Epoch 12 Loss 1.7141 Accuracy 0.6499
Time taken for 1 epoch: 30.62126898765564 secs

Epoch 13 Batch 0 Loss 1.4501 Accuracy 0.6824
Epoch 13 Batch 50 Loss 1.5419 Accuracy 0.6747
Epoch 13 Batch 100 Loss 1.5441 Accuracy 0.6730
Epoch 13 Batch 150 Loss 1.5583 Accuracy 0.6700
Epoch 13 Batch 200 Loss 1.5641 Accuracy 0.6695
Epoch 13 Batch 250 Loss 1.5682 Accuracy 0.6690
Epoch 13 Batch 300 Loss 1.5721 Accuracy 0.6683
Epoch 13 Batch 350 Loss 1.5766 Accuracy 0.6680
Epoch 13 Batch 400 Loss 1.5813 Accuracy 0.6676
Epoch 13 Batch 450 Loss 1.5896 Accuracy 0.6664
Epoch 13 Batch 500 Loss 1.5922 Accuracy 0.6661
Epoch 13 Batch 550 Loss 1.5978 Accuracy 0.6654
Epoch 13 Batch 600 Loss 1.6030 Accuracy 0.6646
Epoch 13 Batch 650 Loss 1.6082 Accuracy 0.6639
Epoch 13 Batch 700 Loss 1.6134 Accuracy 0.6633
Epoch 13 Loss 1.6134 Accuracy 0.6634
Time taken for 1 epoch: 30.817842721939087 secs

Epoch 14 Batch 0 Loss 1.3038 Accuracy 0.7251
Epoch 14 Batch 50 Loss 1.4288 Accuracy 0.6897
Epoch 14 Batch 100 Loss 1.4507 Accuracy 0.6881
Epoch 14 Batch 150 Loss 1.4641 Accuracy 0.6851
Epoch 14 Batch 200 Loss 1.4735 Accuracy 0.6847
Epoch 14 Batch 250 Loss 1.4773 Accuracy 0.6839
Epoch 14 Batch 300 Loss 1.4825 Accuracy 0.6828
Epoch 14 Batch 350 Loss 1.4890 Accuracy 0.6819
Epoch 14 Batch 400 Loss 1.4937 Accuracy 0.6808
Epoch 14 Batch 450 Loss 1.4998 Accuracy 0.6798
Epoch 14 Batch 500 Loss 1.5044 Accuracy 0.6792
Epoch 14 Batch 550 Loss 1.5087 Accuracy 0.6785
Epoch 14 Batch 600 Loss 1.5150 Accuracy 0.6773
Epoch 14 Batch 650 Loss 1.5209 Accuracy 0.6766
Epoch 14 Batch 700 Loss 1.5267 Accuracy 0.6759
Epoch 14 Loss 1.5264 Accuracy 0.6760
Time taken for 1 epoch: 30.76487922668457 secs

Epoch 15 Batch 0 Loss 1.3602 Accuracy 0.7020
Epoch 15 Batch 50 Loss 1.3786 Accuracy 0.6978
Epoch 15 Batch 100 Loss 1.3675 Accuracy 0.6994
Epoch 15 Batch 150 Loss 1.3881 Accuracy 0.6961
Epoch 15 Batch 200 Loss 1.3986 Accuracy 0.6947
Epoch 15 Batch 250 Loss 1.4042 Accuracy 0.6937
Epoch 15 Batch 300 Loss 1.4095 Accuracy 0.6929
Epoch 15 Batch 350 Loss 1.4175 Accuracy 0.6917
Epoch 15 Batch 400 Loss 1.4202 Accuracy 0.6913
Epoch 15 Batch 450 Loss 1.4256 Accuracy 0.6906
Epoch 15 Batch 500 Loss 1.4271 Accuracy 0.6904
Epoch 15 Batch 550 Loss 1.4337 Accuracy 0.6895
Epoch 15 Batch 600 Loss 1.4401 Accuracy 0.6885
Epoch 15 Batch 650 Loss 1.4460 Accuracy 0.6876
Epoch 15 Batch 700 Loss 1.4515 Accuracy 0.6868
Saving checkpoint for epoch 15 at ./checkpoints/train/ckpt-3
Epoch 15 Loss 1.4519 Accuracy 0.6868
Time taken for 1 epoch: 30.835017681121826 secs

Epoch 16 Batch 0 Loss 1.1559 Accuracy 0.7455
Epoch 16 Batch 50 Loss 1.2853 Accuracy 0.7131
Epoch 16 Batch 100 Loss 1.3070 Accuracy 0.7094
Epoch 16 Batch 150 Loss 1.3140 Accuracy 0.7080
Epoch 16 Batch 200 Loss 1.3281 Accuracy 0.7059
Epoch 16 Batch 250 Loss 1.3378 Accuracy 0.7040
Epoch 16 Batch 300 Loss 1.3457 Accuracy 0.7029
Epoch 16 Batch 350 Loss 1.3470 Accuracy 0.7029
Epoch 16 Batch 400 Loss 1.3514 Accuracy 0.7023
Epoch 16 Batch 450 Loss 1.3575 Accuracy 0.7011
Epoch 16 Batch 500 Loss 1.3623 Accuracy 0.7004
Epoch 16 Batch 550 Loss 1.3688 Accuracy 0.6993
Epoch 16 Batch 600 Loss 1.3738 Accuracy 0.6985
Epoch 16 Batch 650 Loss 1.3789 Accuracy 0.6978
Epoch 16 Batch 700 Loss 1.3835 Accuracy 0.6972
Epoch 16 Loss 1.3842 Accuracy 0.6970
Time taken for 1 epoch: 30.86835026741028 secs

Epoch 17 Batch 0 Loss 1.1976 Accuracy 0.7252
Epoch 17 Batch 50 Loss 1.2320 Accuracy 0.7229
Epoch 17 Batch 100 Loss 1.2459 Accuracy 0.7198
Epoch 17 Batch 150 Loss 1.2615 Accuracy 0.7170
Epoch 17 Batch 200 Loss 1.2671 Accuracy 0.7155
Epoch 17 Batch 250 Loss 1.2781 Accuracy 0.7137
Epoch 17 Batch 300 Loss 1.2837 Accuracy 0.7124
Epoch 17 Batch 350 Loss 1.2856 Accuracy 0.7121
Epoch 17 Batch 400 Loss 1.2926 Accuracy 0.7109
Epoch 17 Batch 450 Loss 1.2976 Accuracy 0.7103
Epoch 17 Batch 500 Loss 1.3044 Accuracy 0.7092
Epoch 17 Batch 550 Loss 1.3103 Accuracy 0.7080
Epoch 17 Batch 600 Loss 1.3158 Accuracy 0.7072
Epoch 17 Batch 650 Loss 1.3188 Accuracy 0.7069
Epoch 17 Batch 700 Loss 1.3237 Accuracy 0.7062
Epoch 17 Loss 1.3240 Accuracy 0.7062
Time taken for 1 epoch: 31.70491647720337 secs

Epoch 18 Batch 0 Loss 1.3016 Accuracy 0.7197
Epoch 18 Batch 50 Loss 1.1798 Accuracy 0.7305
Epoch 18 Batch 100 Loss 1.1901 Accuracy 0.7291
Epoch 18 Batch 150 Loss 1.1978 Accuracy 0.7277
Epoch 18 Batch 200 Loss 1.2066 Accuracy 0.7257
Epoch 18 Batch 250 Loss 1.2141 Accuracy 0.7243
Epoch 18 Batch 300 Loss 1.2225 Accuracy 0.7227
Epoch 18 Batch 350 Loss 1.2288 Accuracy 0.7215
Epoch 18 Batch 400 Loss 1.2353 Accuracy 0.7203
Epoch 18 Batch 450 Loss 1.2398 Accuracy 0.7195
Epoch 18 Batch 500 Loss 1.2441 Accuracy 0.7186
Epoch 18 Batch 550 Loss 1.2516 Accuracy 0.7174
Epoch 18 Batch 600 Loss 1.2581 Accuracy 0.7164
Epoch 18 Batch 650 Loss 1.2648 Accuracy 0.7154
Epoch 18 Batch 700 Loss 1.2693 Accuracy 0.7148
Epoch 18 Loss 1.2694 Accuracy 0.7148
Time taken for 1 epoch: 30.78923749923706 secs

Epoch 19 Batch 0 Loss 1.0897 Accuracy 0.7513
Epoch 19 Batch 50 Loss 1.1161 Accuracy 0.7400
Epoch 19 Batch 100 Loss 1.1371 Accuracy 0.7369
Epoch 19 Batch 150 Loss 1.1537 Accuracy 0.7339
Epoch 19 Batch 200 Loss 1.1645 Accuracy 0.7322
Epoch 19 Batch 250 Loss 1.1728 Accuracy 0.7307
Epoch 19 Batch 300 Loss 1.1812 Accuracy 0.7293
Epoch 19 Batch 350 Loss 1.1851 Accuracy 0.7283
Epoch 19 Batch 400 Loss 1.1890 Accuracy 0.7277
Epoch 19 Batch 450 Loss 1.1922 Accuracy 0.7273
Epoch 19 Batch 500 Loss 1.1964 Accuracy 0.7265
Epoch 19 Batch 550 Loss 1.2038 Accuracy 0.7251
Epoch 19 Batch 600 Loss 1.2107 Accuracy 0.7238
Epoch 19 Batch 650 Loss 1.2152 Accuracy 0.7232
Epoch 19 Batch 700 Loss 1.2193 Accuracy 0.7224
Epoch 19 Loss 1.2199 Accuracy 0.7223
Time taken for 1 epoch: 31.050918340682983 secs

Epoch 20 Batch 0 Loss 1.0248 Accuracy 0.7491
Epoch 20 Batch 50 Loss 1.1020 Accuracy 0.7435
Epoch 20 Batch 100 Loss 1.1042 Accuracy 0.7420
Epoch 20 Batch 150 Loss 1.1102 Accuracy 0.7407
Epoch 20 Batch 200 Loss 1.1199 Accuracy 0.7385
Epoch 20 Batch 250 Loss 1.1271 Accuracy 0.7371
Epoch 20 Batch 300 Loss 1.1352 Accuracy 0.7358
Epoch 20 Batch 350 Loss 1.1401 Accuracy 0.7349
Epoch 20 Batch 400 Loss 1.1419 Accuracy 0.7348
Epoch 20 Batch 450 Loss 1.1476 Accuracy 0.7338
Epoch 20 Batch 500 Loss 1.1544 Accuracy 0.7325
Epoch 20 Batch 550 Loss 1.1585 Accuracy 0.7320
Epoch 20 Batch 600 Loss 1.1650 Accuracy 0.7309
Epoch 20 Batch 650 Loss 1.1699 Accuracy 0.7301
Epoch 20 Batch 700 Loss 1.1764 Accuracy 0.7291
Saving checkpoint for epoch 20 at ./checkpoints/train/ckpt-4
Epoch 20 Loss 1.1765 Accuracy 0.7290
Time taken for 1 epoch: 30.766892671585083 secs

Evaluate

The following steps are used for evaluation:

  • Encode the input sentence using the Portuguese tokenizer (tokenizer_pt). Moreover, add the start and end token so the input is equivalent to what the model is trained with. This is the encoder input.
  • The decoder input is the start token == tokenizer_en.vocab_size.
  • Calculate the padding masks and the look ahead masks.
  • The decoder then outputs the predictions by looking at the encoder output and its own output (self-attention).
  • Select the last word and calculate the argmax of that.
  • Concatentate the predicted word to the decoder input as pass it to the decoder.
  • In this approach, the decoder predicts the next word based on the previous words it predicted.
def evaluate(inp_sentence):
  start_token = [tokenizer_pt.vocab_size]
  end_token = [tokenizer_pt.vocab_size + 1]

  # inp sentence is portuguese, hence adding the start and end token
  inp_sentence = start_token + tokenizer_pt.encode(inp_sentence) + end_token
  encoder_input = tf.expand_dims(inp_sentence, 0)

  # as the target is english, the first word to the transformer should be the
  # english start token.
  decoder_input = [tokenizer_en.vocab_size]
  output = tf.expand_dims(decoder_input, 0)

  for i in range(MAX_LENGTH):
    enc_padding_mask, combined_mask, dec_padding_mask = create_masks(
        encoder_input, output)

    # predictions.shape == (batch_size, seq_len, vocab_size)
    predictions, attention_weights = transformer(encoder_input, 
                                                 output,
                                                 False,
                                                 enc_padding_mask,
                                                 combined_mask,
                                                 dec_padding_mask)

    # select the last word from the seq_len dimension
    predictions = predictions[: ,-1:, :]  # (batch_size, 1, vocab_size)

    predicted_id = tf.cast(tf.argmax(predictions, axis=-1), tf.int32)

    # return the result if the predicted_id is equal to the end token
    if predicted_id == tokenizer_en.vocab_size+1:
      return tf.squeeze(output, axis=0), attention_weights

    # concatentate the predicted_id to the output which is given to the decoder
    # as its input.
    output = tf.concat([output, predicted_id], axis=-1)

  return tf.squeeze(output, axis=0), attention_weights

def plot_attention_weights(attention, sentence, result, layer):
  fig = plt.figure(figsize=(16, 8))

  sentence = tokenizer_pt.encode(sentence)

  attention = tf.squeeze(attention[layer], axis=0)

  for head in range(attention.shape[0]):
    ax = fig.add_subplot(2, 4, head+1)

    # plot the attention weights
    ax.matshow(attention[head][:-1, :], cmap='viridis')

    fontdict = {'fontsize': 10}

    ax.set_xticks(range(len(sentence)+2))
    ax.set_yticks(range(len(result)))

    ax.set_ylim(len(result)-1.5, -0.5)

    ax.set_xticklabels(
        ['<start>']+[tokenizer_pt.decode([i]) for i in sentence]+['<end>'], 
        fontdict=fontdict, rotation=90)

    ax.set_yticklabels([tokenizer_en.decode([i]) for i in result 
                        if i < tokenizer_en.vocab_size], 
                       fontdict=fontdict)

    ax.set_xlabel('Head {}'.format(head+1))

  plt.tight_layout()
  plt.show()

def translate(sentence, plot=''):
  result, attention_weights = evaluate(sentence)

  predicted_sentence = tokenizer_en.decode([i for i in result 
                                            if i < tokenizer_en.vocab_size])  

  print('Input: {}'.format(sentence))
  print('Predicted translation: {}'.format(predicted_sentence))

  if plot:
    plot_attention_weights(attention_weights, sentence, result, plot)

translate("este é um problema que temos que resolver.")
print ("Real translation: this is a problem we have to solve .")

Input: este é um problema que temos que resolver.
Predicted translation: this is a problem that we have to solve one .
Real translation: this is a problem we have to solve .


translate("os meus vizinhos ouviram sobre esta ideia.")
print ("Real translation: and my neighboring homes heard about this idea .")

Input: os meus vizinhos ouviram sobre esta ideia.
Predicted translation: my neighbors heard about this idea .
Real translation: and my neighboring homes heard about this idea .


translate("vou então muito rapidamente partilhar convosco algumas histórias de algumas coisas mágicas que aconteceram.")
print ("Real translation: so i 'll just share with you some stories very quickly of some magical things that have happened .")

Input: vou então muito rapidamente partilhar convosco algumas histórias de algumas coisas mágicas que aconteceram.
Predicted translation: so i 'm going to really quickly share with you some magic stories that happened to happen .
Real translation: so i 'll just share with you some stories very quickly of some magical things that have happened .

You can pass different layers and attention blocks of the decoder to the plot parameter.

translate("este é o primeiro livro que eu fiz.", plot='decoder_layer4_block2')
print ("Real translation: this is the first book i've ever done.")

Input: este é o primeiro livro que eu fiz.
Predicted translation: this is the first book i made .

png

Real translation: this is the first book i've ever done.

Summary

In this tutorial, you learned about positional encoding, multi-head attention, the importance of masking and how to create a transformer.

Try using a different dataset to train the transformer. You can also create the base transformer or transformer XL by changing the hyperparameters above. You can also use the layers defined here to create BERT and train state of the art models. Futhermore, you can implement beam search to get better predictions.