RSVP for your your local TensorFlow Everywhere event today!

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 a Portuguese to English dataset. 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 makes 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

Setup

pip install -q tensorflow_datasets
pip install -q tensorflow_text

import collections
import logging
import os
import pathlib
import re
import string
import sys
import time

import numpy as np
import matplotlib.pyplot as plt

import tensorflow_datasets as tfds
import tensorflow_text as text
import tensorflow as tf

logging.getLogger('tensorflow').setLevel(logging.ERROR)  # suppress warnings

Download the Dataset

Use TensorFlow datasets to load the Portuguese-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']

The tf.data.Dataset object returned by TensorFlow datasets yields pairs of text examples:

for pt_examples, en_examples in train_examples.batch(3).take(1):
  for pt in pt_examples.numpy():
    print(pt.decode('utf-8'))

  print()

  for en in en_examples.numpy():
    print(en.decode('utf-8'))

e quando melhoramos a procura , tiramos a única vantagem da impressão , que é a serendipidade .
mas e se estes fatores fossem ativos ?
mas eles não tinham a curiosidade de me testar .

and when you improve searchability , you actually take away the one advantage of print , which is serendipity .
but what if it were active ?
but they did n't test for curiosity .

Text tokenization & detokenization

You can't train a model directly on text. The text needs to be converted some numeric representation first. Typically you convert the text to sequences of token IDs, which are as indexes into an embedding.

One popular implementation is demonstrated in the Subword tokenizer tutorial builds subword tokenizers (text.BertTokenizer) optimized for this dataset and exports them in a saved_model.

Download and unzip and import the saved_model:

model_name = "ted_hrlr_translate_pt_en_converter"
tf.keras.utils.get_file(
    f"{model_name}.zip",
    f"https://storage.googleapis.com/download.tensorflow.org/models/{model_name}.zip",
    cache_dir='.', cache_subdir='', extract=True
)

Downloading data from https://storage.googleapis.com/download.tensorflow.org/models/ted_hrlr_translate_pt_en_converter.zip
188416/184801 [==============================] - 0s 0us/step

'./ted_hrlr_translate_pt_en_converter.zip'

tokenizers = tf.saved_model.load(model_name)

The tf.saved_model contains two text tokenizers, one for English and one for Portugese. Both have the same methods:

[item for item in dir(tokenizers.en) if not item.startswith('_')]

['detokenize',
 'get_reserved_tokens',
 'get_vocab_path',
 'get_vocab_size',
 'lookup',
 'tokenize',
 'tokenizer',
 'vocab']

The tokenize method converts a batch of strings to a padded-batch of token IDs. This method splits punctuation, lowercases and unicode-normalizes the input before tokenizing. That standardization is not visible here because the input data is already standardized.

for en in en_examples.numpy():
  print(en.decode('utf-8'))

and when you improve searchability , you actually take away the one advantage of print , which is serendipity .
but what if it were active ?
but they did n't test for curiosity .


encoded = tokenizers.en.tokenize(en_examples)

for row in encoded.to_list():
  print(row)

[2, 72, 117, 79, 1259, 1491, 2362, 13, 79, 150, 184, 311, 71, 103, 2308, 74, 2679, 13, 148, 80, 55, 4840, 1434, 2423, 540, 15, 3]
[2, 87, 90, 107, 76, 129, 1852, 30, 3]
[2, 87, 83, 149, 50, 9, 56, 664, 85, 2512, 15, 3]

The detokenize method attempts to convert these token IDs back to human readable text: 

round_trip = tokenizers.en.detokenize(encoded)
for line in round_trip.numpy():
  print(line.decode('utf-8'))

and when you improve searchability , you actually take away the one advantage of print , which is serendipity .
but what if it were active ?
but they did n ' t test for curiosity .

The lower level lookup method converts from token-IDs to token text:

tokens = tokenizers.en.lookup(encoded)
tokens

<tf.RaggedTensor [[b'[START]', b'and', b'when', b'you', b'improve', b'search', b'##ability', b',', b'you', b'actually', b'take', b'away', b'the', b'one', b'advantage', b'of', b'print', b',', b'which', b'is', b's', b'##ere', b'##nd', b'##ip', b'##ity', b'.', b'[END]'], [b'[START]', b'but', b'what', b'if', b'it', b'were', b'active', b'?', b'[END]'], [b'[START]', b'but', b'they', b'did', b'n', b"'", b't', b'test', b'for', b'curiosity', b'.', b'[END]']]>

Here you can see the "subword" aspect of the tokenizers. The word "searchability" is decomposed into "search ##ability" and the word "serindipity" into "s ##ere ##nd ##ip ##ity"

Setup input pipeline

To build an input pipeline suitable for training you'll apply some transformations to the dataset.

This function will be used to encode the batches of raw text:

def tokenize_pairs(pt, en):
    pt = tokenizers.pt.tokenize(pt)
    # Convert from ragged to dense, padding with zeros.
    pt = pt.to_tensor()

    en = tokenizers.en.tokenize(en)
    # Convert from ragged to dense, padding with zeros.
    en = en.to_tensor()
    return pt, en

Here's a simple input pipeline that processes, shuffles and batches the data:

BUFFER_SIZE = 20000
BATCH_SIZE = 64

def make_batches(ds):
  return (
      ds
      .cache()
      .shuffle(BUFFER_SIZE)
      .batch(BATCH_SIZE)
      .map(tokenize_pairs, num_parallel_calls=tf.data.AUTOTUNE)
      .prefetch(tf.data.AUTOTUNE))

train_batches = make_batches(train_examples)
val_batches = make_batches(val_examples)

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)

n, d = 2048, 512
pos_encoding = positional_encoding(n, d)
print(pos_encoding.shape)
pos_encoding = pos_encoding[0]

# Juggle the dimensions for the plot
pos_encoding = tf.reshape(pos_encoding, (n, d//2, 2))
pos_encoding = tf.transpose(pos_encoding, (2,1,0))
pos_encoding = tf.reshape(pos_encoding, (d, n))

plt.pcolormesh(pos_encoding, cmap='RdBu')
plt.ylabel('Depth')
plt.xlabel('Position')
plt.colorbar()
plt.show()

(1, 2048, 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. So the square root of dk is used for scaling so you get a consistent variance regardless of the value of dk. If the variance is too low the output may be too flat to optimize effectively. If the variance is too high the softmax may saturate at initilization making it dificult to learn.

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.tokenizer = 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.tokenizer(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
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=num_layers,
    d_model=d_model,
    num_heads=num_heads,
    dff=dff,
    input_vocab_size=tokenizers.pt.get_vocab_size(),
    target_vocab_size=tokenizers.en.get_vocab_size(), 
    pe_input=1000, 
    pe_target=1000,
    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_batches):
    train_step(inp, tar)

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

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

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

  print(f'Time taken for 1 epoch: {time.time() - start:.2f} secs\n')

Epoch 1 Batch 0 Loss 8.8523 Accuracy 0.0000
Epoch 1 Batch 50 Loss 8.7917 Accuracy 0.0099
Epoch 1 Batch 100 Loss 8.6921 Accuracy 0.0342
Epoch 1 Batch 150 Loss 8.5787 Accuracy 0.0436
Epoch 1 Batch 200 Loss 8.4363 Accuracy 0.0492
Epoch 1 Batch 250 Loss 8.2659 Accuracy 0.0542
Epoch 1 Batch 300 Loss 8.0744 Accuracy 0.0613
Epoch 1 Batch 350 Loss 7.8773 Accuracy 0.0684
Epoch 1 Batch 400 Loss 7.6901 Accuracy 0.0747
Epoch 1 Batch 450 Loss 7.5251 Accuracy 0.0804
Epoch 1 Batch 500 Loss 7.3783 Accuracy 0.0860
Epoch 1 Batch 550 Loss 7.2470 Accuracy 0.0913
Epoch 1 Batch 600 Loss 7.1266 Accuracy 0.0973
Epoch 1 Batch 650 Loss 7.0135 Accuracy 0.1036
Epoch 1 Batch 700 Loss 6.9068 Accuracy 0.1099
Epoch 1 Batch 750 Loss 6.8071 Accuracy 0.1157
Epoch 1 Batch 800 Loss 6.7141 Accuracy 0.1214
Epoch 1 Loss 6.6986 Accuracy 0.1222
Time taken for 1 epoch: 67.44 secs

Epoch 2 Batch 0 Loss 5.4622 Accuracy 0.1953
Epoch 2 Batch 50 Loss 5.2542 Accuracy 0.2099
Epoch 2 Batch 100 Loss 5.1996 Accuracy 0.2164
Epoch 2 Batch 150 Loss 5.1635 Accuracy 0.2216
Epoch 2 Batch 200 Loss 5.1462 Accuracy 0.2228
Epoch 2 Batch 250 Loss 5.1150 Accuracy 0.2262
Epoch 2 Batch 300 Loss 5.0863 Accuracy 0.2291
Epoch 2 Batch 350 Loss 5.0669 Accuracy 0.2312
Epoch 2 Batch 400 Loss 5.0468 Accuracy 0.2330
Epoch 2 Batch 450 Loss 5.0242 Accuracy 0.2350
Epoch 2 Batch 500 Loss 5.0009 Accuracy 0.2369
Epoch 2 Batch 550 Loss 4.9802 Accuracy 0.2387
Epoch 2 Batch 600 Loss 4.9624 Accuracy 0.2402
Epoch 2 Batch 650 Loss 4.9460 Accuracy 0.2416
Epoch 2 Batch 700 Loss 4.9283 Accuracy 0.2432
Epoch 2 Batch 750 Loss 4.9127 Accuracy 0.2444
Epoch 2 Batch 800 Loss 4.8953 Accuracy 0.2459
Epoch 2 Loss 4.8917 Accuracy 0.2462
Time taken for 1 epoch: 51.49 secs

Epoch 3 Batch 0 Loss 4.7124 Accuracy 0.2655
Epoch 3 Batch 50 Loss 4.6124 Accuracy 0.2690
Epoch 3 Batch 100 Loss 4.5969 Accuracy 0.2700
Epoch 3 Batch 150 Loss 4.5705 Accuracy 0.2729
Epoch 3 Batch 200 Loss 4.5594 Accuracy 0.2746
Epoch 3 Batch 250 Loss 4.5453 Accuracy 0.2761
Epoch 3 Batch 300 Loss 4.5357 Accuracy 0.2768
Epoch 3 Batch 350 Loss 4.5212 Accuracy 0.2782
Epoch 3 Batch 400 Loss 4.5079 Accuracy 0.2796
Epoch 3 Batch 450 Loss 4.4938 Accuracy 0.2814
Epoch 3 Batch 500 Loss 4.4783 Accuracy 0.2833
Epoch 3 Batch 550 Loss 4.4650 Accuracy 0.2848
Epoch 3 Batch 600 Loss 4.4505 Accuracy 0.2864
Epoch 3 Batch 650 Loss 4.4354 Accuracy 0.2883
Epoch 3 Batch 700 Loss 4.4190 Accuracy 0.2902
Epoch 3 Batch 750 Loss 4.4029 Accuracy 0.2920
Epoch 3 Batch 800 Loss 4.3870 Accuracy 0.2941
Epoch 3 Loss 4.3842 Accuracy 0.2943
Time taken for 1 epoch: 51.46 secs

Epoch 4 Batch 0 Loss 4.1454 Accuracy 0.3202
Epoch 4 Batch 50 Loss 4.0550 Accuracy 0.3283
Epoch 4 Batch 100 Loss 4.0330 Accuracy 0.3337
Epoch 4 Batch 150 Loss 4.0241 Accuracy 0.3353
Epoch 4 Batch 200 Loss 4.0000 Accuracy 0.3379
Epoch 4 Batch 250 Loss 3.9906 Accuracy 0.3391
Epoch 4 Batch 300 Loss 3.9751 Accuracy 0.3412
Epoch 4 Batch 350 Loss 3.9580 Accuracy 0.3433
Epoch 4 Batch 400 Loss 3.9447 Accuracy 0.3454
Epoch 4 Batch 450 Loss 3.9286 Accuracy 0.3474
Epoch 4 Batch 500 Loss 3.9123 Accuracy 0.3496
Epoch 4 Batch 550 Loss 3.8971 Accuracy 0.3517
Epoch 4 Batch 600 Loss 3.8823 Accuracy 0.3537
Epoch 4 Batch 650 Loss 3.8694 Accuracy 0.3556
Epoch 4 Batch 700 Loss 3.8562 Accuracy 0.3575
Epoch 4 Batch 750 Loss 3.8425 Accuracy 0.3593
Epoch 4 Batch 800 Loss 3.8306 Accuracy 0.3609
Epoch 4 Loss 3.8278 Accuracy 0.3613
Time taken for 1 epoch: 51.82 secs

Epoch 5 Batch 0 Loss 3.5380 Accuracy 0.3838
Epoch 5 Batch 50 Loss 3.5084 Accuracy 0.3961
Epoch 5 Batch 100 Loss 3.5043 Accuracy 0.3986
Epoch 5 Batch 150 Loss 3.4923 Accuracy 0.4000
Epoch 5 Batch 200 Loss 3.4812 Accuracy 0.4019
Epoch 5 Batch 250 Loss 3.4678 Accuracy 0.4042
Epoch 5 Batch 300 Loss 3.4593 Accuracy 0.4060
Epoch 5 Batch 350 Loss 3.4513 Accuracy 0.4069
Epoch 5 Batch 400 Loss 3.4459 Accuracy 0.4075
Epoch 5 Batch 450 Loss 3.4375 Accuracy 0.4083
Epoch 5 Batch 500 Loss 3.4280 Accuracy 0.4095
Epoch 5 Batch 550 Loss 3.4161 Accuracy 0.4111
Epoch 5 Batch 600 Loss 3.4038 Accuracy 0.4130
Epoch 5 Batch 650 Loss 3.3965 Accuracy 0.4140
Epoch 5 Batch 700 Loss 3.3886 Accuracy 0.4152
Epoch 5 Batch 750 Loss 3.3797 Accuracy 0.4161
Epoch 5 Batch 800 Loss 3.3698 Accuracy 0.4174
Saving checkpoint for epoch 5 at ./checkpoints/train/ckpt-1
Epoch 5 Loss 3.3683 Accuracy 0.4175
Time taken for 1 epoch: 53.01 secs

Epoch 6 Batch 0 Loss 3.1906 Accuracy 0.4351
Epoch 6 Batch 50 Loss 3.1223 Accuracy 0.4430
Epoch 6 Batch 100 Loss 3.0932 Accuracy 0.4475
Epoch 6 Batch 150 Loss 3.0904 Accuracy 0.4482
Epoch 6 Batch 200 Loss 3.0847 Accuracy 0.4489
Epoch 6 Batch 250 Loss 3.0772 Accuracy 0.4505
Epoch 6 Batch 300 Loss 3.0762 Accuracy 0.4505
Epoch 6 Batch 350 Loss 3.0696 Accuracy 0.4516
Epoch 6 Batch 400 Loss 3.0620 Accuracy 0.4524
Epoch 6 Batch 450 Loss 3.0489 Accuracy 0.4544
Epoch 6 Batch 500 Loss 3.0417 Accuracy 0.4556
Epoch 6 Batch 550 Loss 3.0318 Accuracy 0.4568
Epoch 6 Batch 600 Loss 3.0215 Accuracy 0.4583
Epoch 6 Batch 650 Loss 3.0122 Accuracy 0.4597
Epoch 6 Batch 700 Loss 3.0039 Accuracy 0.4611
Epoch 6 Batch 750 Loss 2.9967 Accuracy 0.4620
Epoch 6 Batch 800 Loss 2.9880 Accuracy 0.4634
Epoch 6 Loss 2.9880 Accuracy 0.4634
Time taken for 1 epoch: 51.59 secs

Epoch 7 Batch 0 Loss 2.8409 Accuracy 0.4786
Epoch 7 Batch 50 Loss 2.7417 Accuracy 0.4890
Epoch 7 Batch 100 Loss 2.7202 Accuracy 0.4938
Epoch 7 Batch 150 Loss 2.7067 Accuracy 0.4968
Epoch 7 Batch 200 Loss 2.7116 Accuracy 0.4972
Epoch 7 Batch 250 Loss 2.7108 Accuracy 0.4978
Epoch 7 Batch 300 Loss 2.7008 Accuracy 0.4992
Epoch 7 Batch 350 Loss 2.6981 Accuracy 0.4999
Epoch 7 Batch 400 Loss 2.6911 Accuracy 0.5010
Epoch 7 Batch 450 Loss 2.6845 Accuracy 0.5018
Epoch 7 Batch 500 Loss 2.6826 Accuracy 0.5020
Epoch 7 Batch 550 Loss 2.6797 Accuracy 0.5024
Epoch 7 Batch 600 Loss 2.6740 Accuracy 0.5032
Epoch 7 Batch 650 Loss 2.6724 Accuracy 0.5035
Epoch 7 Batch 700 Loss 2.6673 Accuracy 0.5044
Epoch 7 Batch 750 Loss 2.6628 Accuracy 0.5048
Epoch 7 Batch 800 Loss 2.6597 Accuracy 0.5053
Epoch 7 Loss 2.6587 Accuracy 0.5054
Time taken for 1 epoch: 51.14 secs

Epoch 8 Batch 0 Loss 2.3882 Accuracy 0.5387
Epoch 8 Batch 50 Loss 2.4424 Accuracy 0.5315
Epoch 8 Batch 100 Loss 2.4410 Accuracy 0.5327
Epoch 8 Batch 150 Loss 2.4445 Accuracy 0.5326
Epoch 8 Batch 200 Loss 2.4409 Accuracy 0.5337
Epoch 8 Batch 250 Loss 2.4420 Accuracy 0.5334
Epoch 8 Batch 300 Loss 2.4367 Accuracy 0.5345
Epoch 8 Batch 350 Loss 2.4359 Accuracy 0.5345
Epoch 8 Batch 400 Loss 2.4336 Accuracy 0.5350
Epoch 8 Batch 450 Loss 2.4320 Accuracy 0.5353
Epoch 8 Batch 500 Loss 2.4286 Accuracy 0.5361
Epoch 8 Batch 550 Loss 2.4250 Accuracy 0.5366
Epoch 8 Batch 600 Loss 2.4256 Accuracy 0.5365
Epoch 8 Batch 650 Loss 2.4243 Accuracy 0.5369
Epoch 8 Batch 700 Loss 2.4229 Accuracy 0.5371
Epoch 8 Batch 750 Loss 2.4211 Accuracy 0.5374
Epoch 8 Batch 800 Loss 2.4216 Accuracy 0.5374
Epoch 8 Loss 2.4208 Accuracy 0.5375
Time taken for 1 epoch: 50.51 secs

Epoch 9 Batch 0 Loss 2.0652 Accuracy 0.5853
Epoch 9 Batch 50 Loss 2.2172 Accuracy 0.5631
Epoch 9 Batch 100 Loss 2.2405 Accuracy 0.5601
Epoch 9 Batch 150 Loss 2.2480 Accuracy 0.5591
Epoch 9 Batch 200 Loss 2.2527 Accuracy 0.5587
Epoch 9 Batch 250 Loss 2.2496 Accuracy 0.5594
Epoch 9 Batch 300 Loss 2.2522 Accuracy 0.5587
Epoch 9 Batch 350 Loss 2.2539 Accuracy 0.5585
Epoch 9 Batch 400 Loss 2.2556 Accuracy 0.5584
Epoch 9 Batch 450 Loss 2.2529 Accuracy 0.5591
Epoch 9 Batch 500 Loss 2.2488 Accuracy 0.5599
Epoch 9 Batch 550 Loss 2.2471 Accuracy 0.5603
Epoch 9 Batch 600 Loss 2.2460 Accuracy 0.5604
Epoch 9 Batch 650 Loss 2.2441 Accuracy 0.5610
Epoch 9 Batch 700 Loss 2.2427 Accuracy 0.5613
Epoch 9 Batch 750 Loss 2.2420 Accuracy 0.5615
Epoch 9 Batch 800 Loss 2.2410 Accuracy 0.5618
Epoch 9 Loss 2.2399 Accuracy 0.5620
Time taken for 1 epoch: 50.38 secs

Epoch 10 Batch 0 Loss 2.1567 Accuracy 0.5736
Epoch 10 Batch 50 Loss 2.1007 Accuracy 0.5797
Epoch 10 Batch 100 Loss 2.0843 Accuracy 0.5824
Epoch 10 Batch 150 Loss 2.0873 Accuracy 0.5824
Epoch 10 Batch 200 Loss 2.0926 Accuracy 0.5818
Epoch 10 Batch 250 Loss 2.1019 Accuracy 0.5800
Epoch 10 Batch 300 Loss 2.1035 Accuracy 0.5797
Epoch 10 Batch 350 Loss 2.1043 Accuracy 0.5796
Epoch 10 Batch 400 Loss 2.1020 Accuracy 0.5802
Epoch 10 Batch 450 Loss 2.0987 Accuracy 0.5810
Epoch 10 Batch 500 Loss 2.0979 Accuracy 0.5813
Epoch 10 Batch 550 Loss 2.0952 Accuracy 0.5817
Epoch 10 Batch 600 Loss 2.0935 Accuracy 0.5820
Epoch 10 Batch 650 Loss 2.0944 Accuracy 0.5821
Epoch 10 Batch 700 Loss 2.0973 Accuracy 0.5818
Epoch 10 Batch 750 Loss 2.0963 Accuracy 0.5821
Epoch 10 Batch 800 Loss 2.0992 Accuracy 0.5817
Saving checkpoint for epoch 10 at ./checkpoints/train/ckpt-2
Epoch 10 Loss 2.0997 Accuracy 0.5816
Time taken for 1 epoch: 51.77 secs

Epoch 11 Batch 0 Loss 1.9663 Accuracy 0.6060
Epoch 11 Batch 50 Loss 1.9794 Accuracy 0.5981
Epoch 11 Batch 100 Loss 1.9783 Accuracy 0.5989
Epoch 11 Batch 150 Loss 1.9630 Accuracy 0.6013
Epoch 11 Batch 200 Loss 1.9593 Accuracy 0.6020
Epoch 11 Batch 250 Loss 1.9603 Accuracy 0.6027
Epoch 11 Batch 300 Loss 1.9651 Accuracy 0.6014
Epoch 11 Batch 350 Loss 1.9653 Accuracy 0.6014
Epoch 11 Batch 400 Loss 1.9676 Accuracy 0.6012
Epoch 11 Batch 450 Loss 1.9677 Accuracy 0.6011
Epoch 11 Batch 500 Loss 1.9697 Accuracy 0.6008
Epoch 11 Batch 550 Loss 1.9714 Accuracy 0.6006
Epoch 11 Batch 600 Loss 1.9796 Accuracy 0.5994
Epoch 11 Batch 650 Loss 1.9793 Accuracy 0.5995
Epoch 11 Batch 700 Loss 1.9812 Accuracy 0.5992
Epoch 11 Batch 750 Loss 1.9838 Accuracy 0.5989
Epoch 11 Batch 800 Loss 1.9859 Accuracy 0.5986
Epoch 11 Loss 1.9867 Accuracy 0.5984
Time taken for 1 epoch: 52.12 secs

Epoch 12 Batch 0 Loss 1.7489 Accuracy 0.6390
Epoch 12 Batch 50 Loss 1.8619 Accuracy 0.6154
Epoch 12 Batch 100 Loss 1.8582 Accuracy 0.6177
Epoch 12 Batch 150 Loss 1.8589 Accuracy 0.6175
Epoch 12 Batch 200 Loss 1.8625 Accuracy 0.6162
Epoch 12 Batch 250 Loss 1.8678 Accuracy 0.6149
Epoch 12 Batch 300 Loss 1.8726 Accuracy 0.6145
Epoch 12 Batch 350 Loss 1.8725 Accuracy 0.6146
Epoch 12 Batch 400 Loss 1.8760 Accuracy 0.6141
Epoch 12 Batch 450 Loss 1.8779 Accuracy 0.6136
Epoch 12 Batch 500 Loss 1.8765 Accuracy 0.6138
Epoch 12 Batch 550 Loss 1.8779 Accuracy 0.6137
Epoch 12 Batch 600 Loss 1.8793 Accuracy 0.6135
Epoch 12 Batch 650 Loss 1.8818 Accuracy 0.6131
Epoch 12 Batch 700 Loss 1.8846 Accuracy 0.6127
Epoch 12 Batch 750 Loss 1.8859 Accuracy 0.6126
Epoch 12 Batch 800 Loss 1.8873 Accuracy 0.6125
Epoch 12 Loss 1.8875 Accuracy 0.6124
Time taken for 1 epoch: 50.93 secs

Epoch 13 Batch 0 Loss 1.7549 Accuracy 0.6243
Epoch 13 Batch 50 Loss 1.7603 Accuracy 0.6322
Epoch 13 Batch 100 Loss 1.7573 Accuracy 0.6317
Epoch 13 Batch 150 Loss 1.7789 Accuracy 0.6283
Epoch 13 Batch 200 Loss 1.7789 Accuracy 0.6284
Epoch 13 Batch 250 Loss 1.7758 Accuracy 0.6290
Epoch 13 Batch 300 Loss 1.7837 Accuracy 0.6274
Epoch 13 Batch 350 Loss 1.7873 Accuracy 0.6271
Epoch 13 Batch 400 Loss 1.7883 Accuracy 0.6271
Epoch 13 Batch 450 Loss 1.7900 Accuracy 0.6271
Epoch 13 Batch 500 Loss 1.7902 Accuracy 0.6271
Epoch 13 Batch 550 Loss 1.7937 Accuracy 0.6267
Epoch 13 Batch 600 Loss 1.7965 Accuracy 0.6262
Epoch 13 Batch 650 Loss 1.8002 Accuracy 0.6256
Epoch 13 Batch 700 Loss 1.8041 Accuracy 0.6252
Epoch 13 Batch 750 Loss 1.8072 Accuracy 0.6247
Epoch 13 Batch 800 Loss 1.8090 Accuracy 0.6246
Epoch 13 Loss 1.8080 Accuracy 0.6248
Time taken for 1 epoch: 50.19 secs

Epoch 14 Batch 0 Loss 1.8098 Accuracy 0.6300
Epoch 14 Batch 50 Loss 1.6760 Accuracy 0.6449
Epoch 14 Batch 100 Loss 1.6889 Accuracy 0.6430
Epoch 14 Batch 150 Loss 1.6983 Accuracy 0.6419
Epoch 14 Batch 200 Loss 1.7033 Accuracy 0.6404
Epoch 14 Batch 250 Loss 1.7102 Accuracy 0.6396
Epoch 14 Batch 300 Loss 1.7142 Accuracy 0.6388
Epoch 14 Batch 350 Loss 1.7223 Accuracy 0.6376
Epoch 14 Batch 400 Loss 1.7229 Accuracy 0.6372
Epoch 14 Batch 450 Loss 1.7237 Accuracy 0.6373
Epoch 14 Batch 500 Loss 1.7242 Accuracy 0.6371
Epoch 14 Batch 550 Loss 1.7281 Accuracy 0.6364
Epoch 14 Batch 600 Loss 1.7293 Accuracy 0.6363
Epoch 14 Batch 650 Loss 1.7310 Accuracy 0.6360
Epoch 14 Batch 700 Loss 1.7336 Accuracy 0.6356
Epoch 14 Batch 750 Loss 1.7372 Accuracy 0.6351
Epoch 14 Batch 800 Loss 1.7375 Accuracy 0.6352
Epoch 14 Loss 1.7370 Accuracy 0.6353
Time taken for 1 epoch: 49.40 secs

Epoch 15 Batch 0 Loss 1.6549 Accuracy 0.6411
Epoch 15 Batch 50 Loss 1.6351 Accuracy 0.6485
Epoch 15 Batch 100 Loss 1.6318 Accuracy 0.6497
Epoch 15 Batch 150 Loss 1.6383 Accuracy 0.6493
Epoch 15 Batch 200 Loss 1.6434 Accuracy 0.6485
Epoch 15 Batch 250 Loss 1.6455 Accuracy 0.6478
Epoch 15 Batch 300 Loss 1.6462 Accuracy 0.6476
Epoch 15 Batch 350 Loss 1.6535 Accuracy 0.6464
Epoch 15 Batch 400 Loss 1.6570 Accuracy 0.6460
Epoch 15 Batch 450 Loss 1.6636 Accuracy 0.6448
Epoch 15 Batch 500 Loss 1.6683 Accuracy 0.6441
Epoch 15 Batch 550 Loss 1.6689 Accuracy 0.6441
Epoch 15 Batch 600 Loss 1.6702 Accuracy 0.6441
Epoch 15 Batch 650 Loss 1.6733 Accuracy 0.6438
Epoch 15 Batch 700 Loss 1.6747 Accuracy 0.6439
Epoch 15 Batch 750 Loss 1.6752 Accuracy 0.6440
Epoch 15 Batch 800 Loss 1.6762 Accuracy 0.6440
Saving checkpoint for epoch 15 at ./checkpoints/train/ckpt-3
Epoch 15 Loss 1.6764 Accuracy 0.6441
Time taken for 1 epoch: 49.83 secs

Epoch 16 Batch 0 Loss 1.6530 Accuracy 0.6543
Epoch 16 Batch 50 Loss 1.5770 Accuracy 0.6594
Epoch 16 Batch 100 Loss 1.5704 Accuracy 0.6605
Epoch 16 Batch 150 Loss 1.5830 Accuracy 0.6588
Epoch 16 Batch 200 Loss 1.5934 Accuracy 0.6573
Epoch 16 Batch 250 Loss 1.5939 Accuracy 0.6574
Epoch 16 Batch 300 Loss 1.5968 Accuracy 0.6569
Epoch 16 Batch 350 Loss 1.5974 Accuracy 0.6566
Epoch 16 Batch 400 Loss 1.5987 Accuracy 0.6560
Epoch 16 Batch 450 Loss 1.6013 Accuracy 0.6555
Epoch 16 Batch 500 Loss 1.6052 Accuracy 0.6549
Epoch 16 Batch 550 Loss 1.6075 Accuracy 0.6547
Epoch 16 Batch 600 Loss 1.6097 Accuracy 0.6545
Epoch 16 Batch 650 Loss 1.6140 Accuracy 0.6537
Epoch 16 Batch 700 Loss 1.6163 Accuracy 0.6534
Epoch 16 Batch 750 Loss 1.6195 Accuracy 0.6529
Epoch 16 Batch 800 Loss 1.6221 Accuracy 0.6526
Epoch 16 Loss 1.6227 Accuracy 0.6525
Time taken for 1 epoch: 50.37 secs

Epoch 17 Batch 0 Loss 1.6708 Accuracy 0.6313
Epoch 17 Batch 50 Loss 1.5137 Accuracy 0.6694
Epoch 17 Batch 100 Loss 1.5265 Accuracy 0.6680
Epoch 17 Batch 150 Loss 1.5309 Accuracy 0.6671
Epoch 17 Batch 200 Loss 1.5397 Accuracy 0.6658
Epoch 17 Batch 250 Loss 1.5395 Accuracy 0.6657
Epoch 17 Batch 300 Loss 1.5443 Accuracy 0.6648
Epoch 17 Batch 350 Loss 1.5483 Accuracy 0.6640
Epoch 17 Batch 400 Loss 1.5482 Accuracy 0.6640
Epoch 17 Batch 450 Loss 1.5500 Accuracy 0.6636
Epoch 17 Batch 500 Loss 1.5540 Accuracy 0.6628
Epoch 17 Batch 550 Loss 1.5565 Accuracy 0.6626
Epoch 17 Batch 600 Loss 1.5602 Accuracy 0.6622
Epoch 17 Batch 650 Loss 1.5627 Accuracy 0.6617
Epoch 17 Batch 700 Loss 1.5667 Accuracy 0.6611
Epoch 17 Batch 750 Loss 1.5706 Accuracy 0.6606
Epoch 17 Batch 800 Loss 1.5729 Accuracy 0.6603
Epoch 17 Loss 1.5732 Accuracy 0.6602
Time taken for 1 epoch: 50.85 secs

Epoch 18 Batch 0 Loss 1.5180 Accuracy 0.6517
Epoch 18 Batch 50 Loss 1.4869 Accuracy 0.6728
Epoch 18 Batch 100 Loss 1.4810 Accuracy 0.6755
Epoch 18 Batch 150 Loss 1.4822 Accuracy 0.6753
Epoch 18 Batch 200 Loss 1.4954 Accuracy 0.6726
Epoch 18 Batch 250 Loss 1.5004 Accuracy 0.6716
Epoch 18 Batch 300 Loss 1.5002 Accuracy 0.6715
Epoch 18 Batch 350 Loss 1.5054 Accuracy 0.6708
Epoch 18 Batch 400 Loss 1.5031 Accuracy 0.6712
Epoch 18 Batch 450 Loss 1.5069 Accuracy 0.6706
Epoch 18 Batch 500 Loss 1.5091 Accuracy 0.6703
Epoch 18 Batch 550 Loss 1.5129 Accuracy 0.6697
Epoch 18 Batch 600 Loss 1.5164 Accuracy 0.6691
Epoch 18 Batch 650 Loss 1.5187 Accuracy 0.6687
Epoch 18 Batch 700 Loss 1.5223 Accuracy 0.6680
Epoch 18 Batch 750 Loss 1.5250 Accuracy 0.6677
Epoch 18 Batch 800 Loss 1.5287 Accuracy 0.6671
Epoch 18 Loss 1.5287 Accuracy 0.6672
Time taken for 1 epoch: 50.13 secs

Epoch 19 Batch 0 Loss 1.3724 Accuracy 0.7021
Epoch 19 Batch 50 Loss 1.4231 Accuracy 0.6849
Epoch 19 Batch 100 Loss 1.4411 Accuracy 0.6820
Epoch 19 Batch 150 Loss 1.4435 Accuracy 0.6807
Epoch 19 Batch 200 Loss 1.4601 Accuracy 0.6778
Epoch 19 Batch 250 Loss 1.4603 Accuracy 0.6776
Epoch 19 Batch 300 Loss 1.4620 Accuracy 0.6775
Epoch 19 Batch 350 Loss 1.4651 Accuracy 0.6771
Epoch 19 Batch 400 Loss 1.4667 Accuracy 0.6768
Epoch 19 Batch 450 Loss 1.4695 Accuracy 0.6762
Epoch 19 Batch 500 Loss 1.4728 Accuracy 0.6758
Epoch 19 Batch 550 Loss 1.4749 Accuracy 0.6754
Epoch 19 Batch 600 Loss 1.4784 Accuracy 0.6749
Epoch 19 Batch 650 Loss 1.4813 Accuracy 0.6744
Epoch 19 Batch 700 Loss 1.4833 Accuracy 0.6741
Epoch 19 Batch 750 Loss 1.4868 Accuracy 0.6734
Epoch 19 Batch 800 Loss 1.4887 Accuracy 0.6733
Epoch 19 Loss 1.4898 Accuracy 0.6731
Time taken for 1 epoch: 49.26 secs

Epoch 20 Batch 0 Loss 1.4377 Accuracy 0.6692
Epoch 20 Batch 50 Loss 1.3902 Accuracy 0.6877
Epoch 20 Batch 100 Loss 1.3939 Accuracy 0.6892
Epoch 20 Batch 150 Loss 1.3993 Accuracy 0.6880
Epoch 20 Batch 200 Loss 1.4108 Accuracy 0.6860
Epoch 20 Batch 250 Loss 1.4194 Accuracy 0.6842
Epoch 20 Batch 300 Loss 1.4224 Accuracy 0.6836
Epoch 20 Batch 350 Loss 1.4269 Accuracy 0.6826
Epoch 20 Batch 400 Loss 1.4289 Accuracy 0.6820
Epoch 20 Batch 450 Loss 1.4347 Accuracy 0.6810
Epoch 20 Batch 500 Loss 1.4362 Accuracy 0.6809
Epoch 20 Batch 550 Loss 1.4402 Accuracy 0.6803
Epoch 20 Batch 600 Loss 1.4446 Accuracy 0.6797
Epoch 20 Batch 650 Loss 1.4458 Accuracy 0.6797
Epoch 20 Batch 700 Loss 1.4489 Accuracy 0.6793
Epoch 20 Batch 750 Loss 1.4511 Accuracy 0.6792
Epoch 20 Batch 800 Loss 1.4548 Accuracy 0.6786
Saving checkpoint for epoch 20 at ./checkpoints/train/ckpt-4
Epoch 20 Loss 1.4546 Accuracy 0.6787
Time taken for 1 epoch: 49.58 secs

Evaluate

The following steps are used for evaluation:

  • Encode the input sentence using the Portuguese tokenizer (tokenizers.pt). This is the encoder input.
  • The decoder input is initialized to the [START] token.
  • 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).
  • The model makes predictions of the next word for each word in the output. Most of these are redundant. Use the predictrions from the last word.
  • Concatentate the predicted word to the decoder input and pass it to the decoder.
  • In this approach, the decoder predicts the next word based on the previous words it predicted.
def evaluate(sentence, max_length=40):
  # inp sentence is portuguese, hence adding the start and end token
  sentence = tf.convert_to_tensor([sentence])
  sentence = tokenizers.pt.tokenize(sentence).to_tensor()

  encoder_input = sentence

  # as the target is english, the first word to the transformer should be the
  # english start token.
  start, end = tokenizers.en.tokenize([''])[0]
  output = tf.convert_to_tensor([start])
  output = tf.expand_dims(output, 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.argmax(predictions, axis=-1)

    # 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 the result if the predicted_id is equal to the end token
    if predicted_id == end:
      break

  # output.shape (1, tokens)
  text = tokenizers.en.detokenize(output)[0] # shape: ()

  tokens = tokenizers.en.lookup(output)[0]

  return text, tokens, attention_weights

def print_translation(sentence, tokens, ground_truth):
  print(f'{"Input:":15s}: {sentence}')
  print(f'{"Prediction":15s}: {tokens.numpy().decode("utf-8")}')
  print(f'{"Ground truth":15s}: {ground_truth}')

sentence = "este é um problema que temos que resolver."
ground_truth = "this is a problem we have to solve ."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

Input:         : este é um problema que temos que resolver.
Prediction     : this is a problem that we have to fix .
Ground truth   : this is a problem we have to solve .


sentence = "os meus vizinhos ouviram sobre esta ideia."
ground_truth = "and my neighboring homes heard about this idea ."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

Input:         : os meus vizinhos ouviram sobre esta ideia.
Prediction     : my neighbors heard about this idea .
Ground truth   : and my neighboring homes heard about this idea .


sentence = "vou então muito rapidamente partilhar convosco algumas histórias de algumas coisas mágicas que aconteceram."
ground_truth = "so i \'ll just share with you some stories very quickly of some magical things that have happened ."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

Input:         : vou então muito rapidamente partilhar convosco algumas histórias de algumas coisas mágicas que aconteceram.
Prediction     : so i ' m going to go very quickly share with you some of the magic things that happened .
Ground truth   : 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.

Attention plots

The evaluate function also returns a dictionary of attention maps you can use to visualize the internal working of the model:

sentence = "este é o primeiro livro que eu fiz."
ground_truth = "this is the first book i've ever done."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

Input:         : este é o primeiro livro que eu fiz.
Prediction     : this is the first book i did .
Ground truth   : this is the first book i've ever done.


def plot_attention_head(in_tokens, translated_tokens, attention):
  # The plot is of the attention when a token was generated.
  # The model didn't generate `<START>` in the output. Skip it. 
  translated_tokens = translated_tokens[1:]

  ax = plt.gca()
  ax.matshow(attention)
  ax.set_xticks(range(len(in_tokens)))
  ax.set_yticks(range(len(translated_tokens)))


  labels = [label.decode('utf-8') for label in in_tokens.numpy()]
  ax.set_xticklabels(
      labels, rotation=90)

  labels = [label.decode('utf-8') for label in translated_tokens.numpy()]
  ax.set_yticklabels(labels)

head = 0
# shape: (batch=1, num_heads, seq_len_q, seq_len_k)
attention_heads = tf.squeeze(
  attention_weights['decoder_layer4_block2'], 0)
attention = attention_heads[head]
attention.shape

TensorShape([9, 11])

in_tokens = tf.convert_to_tensor([sentence])
in_tokens = tokenizers.pt.tokenize(in_tokens).to_tensor()
in_tokens = tokenizers.pt.lookup(in_tokens)[0]
in_tokens

<tf.Tensor: shape=(11,), dtype=string, numpy=
array([b'[START]', b'este', b'e', b'o', b'primeiro', b'livro', b'que',
       b'eu', b'fiz', b'.', b'[END]'], dtype=object)>

translated_tokens

<tf.Tensor: shape=(10,), dtype=string, numpy=
array([b'[START]', b'this', b'is', b'the', b'first', b'book', b'i',
       b'did', b'.', b'[END]'], dtype=object)>

plot_attention_head(in_tokens, translated_tokens, attention)

png

def plot_attention_weights(sentence, translated_tokens, attention_heads):
  in_tokens = tf.convert_to_tensor([sentence])
  in_tokens = tokenizers.pt.tokenize(in_tokens).to_tensor()
  in_tokens = tokenizers.pt.lookup(in_tokens)[0]
  in_tokens

  fig = plt.figure(figsize=(16, 8))

  for h, head in enumerate(attention_heads):
    ax = fig.add_subplot(2, 4, h+1)

    plot_attention_head(in_tokens, translated_tokens, head)

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

  plt.tight_layout()
  plt.show()

plot_attention_weights(sentence, translated_tokens,
                       attention_weights['decoder_layer4_block2'][0])

png

The model does okay on unfamiliar words. Neither "triceratops" or "encyclopedia" are in the input dataset and the model almost learns to transliterare them, even withoput a shared vocabulary:

sentence = "Eu li sobre triceratops na enciclopédia."
ground_truth = "I read about triceratops in the encyclopedia."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

plot_attention_weights(sentence, translated_tokens,
                       attention_weights['decoder_layer4_block2'][0])

Input:         : Eu li sobre triceratops na enciclopédia.
Prediction     : i read about tariopates in eelevada .
Ground truth   : I read about triceratops in the encyclopedia.

png

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. Furthermore, you can implement beam search to get better predictions.