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.
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 ted_hrlr_translate/pt_to_en/1.0.0 (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... HBox(children=(FloatProgress(value=1.0, bar_style='info', description='Dl Completed...', max=1.0, style=Progre… HBox(children=(FloatProgress(value=1.0, bar_style='info', description='Dl Size...', max=1.0, style=ProgressSty… HBox(children=(FloatProgress(value=1.0, bar_style='info', description='Extraction completed...', max=1.0, styl… HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value=''))) Shuffling and writing examples to /home/kbuilder/tensorflow_datasets/ted_hrlr_translate/pt_to_en/1.0.0.incompleteGHXV0Y/ted_hrlr_translate-train.tfrecord HBox(children=(FloatProgress(value=0.0, max=51785.0), HTML(value=''))) HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value=''))) Shuffling and writing examples to /home/kbuilder/tensorflow_datasets/ted_hrlr_translate/pt_to_en/1.0.0.incompleteGHXV0Y/ted_hrlr_translate-validation.tfrecord HBox(children=(FloatProgress(value=0.0, max=1193.0), HTML(value=''))) HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value=''))) Shuffling and writing examples to /home/kbuilder/tensorflow_datasets/ted_hrlr_translate/pt_to_en/1.0.0.incompleteGHXV0Y/ted_hrlr_translate-test.tfrecord HBox(children=(FloatProgress(value=0.0, max=1803.0), HTML(value=''))) 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.features.text.SubwordTextEncoder.build_from_corpus(
(en.numpy() for pt, en in train_examples), target_vocab_size=2**13)
tokenizer_pt = tfds.features.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.experimental.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)
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
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 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
The transformer model follows the same general pattern as a standard sequence to sequence with attention model.
- The input sentence is passed through
Nencoder 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:
- Multi-head attention (with padding mask)
- 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:
- Masked multi-head attention (with look ahead mask and padding mask)
- 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.
- 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:
- Input Embedding
- Positional Encoding
- 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:
- Output Embedding
- Positional Encoding
- 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')
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)
train_loss = tf.keras.metrics.Mean(name='train_loss')
train_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(
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(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 9.0049 Accuracy 0.0000 Epoch 1 Batch 50 Loss 8.9567 Accuracy 0.0026 Epoch 1 Batch 100 Loss 8.8606 Accuracy 0.0128 Epoch 1 Batch 150 Loss 8.7559 Accuracy 0.0164 Epoch 1 Batch 200 Loss 8.6292 Accuracy 0.0182 Epoch 1 Batch 250 Loss 8.4742 Accuracy 0.0217 Epoch 1 Batch 300 Loss 8.2982 Accuracy 0.0261 Epoch 1 Batch 350 Loss 8.1130 Accuracy 0.0294 Epoch 1 Batch 400 Loss 7.9311 Accuracy 0.0321 Epoch 1 Batch 450 Loss 7.7655 Accuracy 0.0346 Epoch 1 Batch 500 Loss 7.6206 Accuracy 0.0378 Epoch 1 Batch 550 Loss 7.4850 Accuracy 0.0414 Epoch 1 Batch 600 Loss 7.3584 Accuracy 0.0450 Epoch 1 Batch 650 Loss 7.2369 Accuracy 0.0484 Epoch 1 Batch 700 Loss 7.1234 Accuracy 0.0517 Epoch 1 Loss 7.1193 Accuracy 0.0518 Time taken for 1 epoch: 58.32971215248108 secs Epoch 2 Batch 0 Loss 5.6753 Accuracy 0.0963 Epoch 2 Batch 50 Loss 5.5158 Accuracy 0.1016 Epoch 2 Batch 100 Loss 5.4521 Accuracy 0.1046 Epoch 2 Batch 150 Loss 5.3919 Accuracy 0.1072 Epoch 2 Batch 200 Loss 5.3447 Accuracy 0.1090 Epoch 2 Batch 250 Loss 5.3038 Accuracy 0.1107 Epoch 2 Batch 300 Loss 5.2630 Accuracy 0.1125 Epoch 2 Batch 350 Loss 5.2256 Accuracy 0.1141 Epoch 2 Batch 400 Loss 5.1901 Accuracy 0.1157 Epoch 2 Batch 450 Loss 5.1559 Accuracy 0.1175 Epoch 2 Batch 500 Loss 5.1251 Accuracy 0.1193 Epoch 2 Batch 550 Loss 5.0987 Accuracy 0.1207 Epoch 2 Batch 600 Loss 5.0707 Accuracy 0.1219 Epoch 2 Batch 650 Loss 5.0458 Accuracy 0.1231 Epoch 2 Batch 700 Loss 5.0200 Accuracy 0.1244 Epoch 2 Loss 5.0193 Accuracy 0.1245 Time taken for 1 epoch: 31.12289547920227 secs Epoch 3 Batch 0 Loss 4.6013 Accuracy 0.1310 Epoch 3 Batch 50 Loss 4.6188 Accuracy 0.1419 Epoch 3 Batch 100 Loss 4.6061 Accuracy 0.1443 Epoch 3 Batch 150 Loss 4.5925 Accuracy 0.1441 Epoch 3 Batch 200 Loss 4.5802 Accuracy 0.1446 Epoch 3 Batch 250 Loss 4.5636 Accuracy 0.1453 Epoch 3 Batch 300 Loss 4.5484 Accuracy 0.1462 Epoch 3 Batch 350 Loss 4.5344 Accuracy 0.1468 Epoch 3 Batch 400 Loss 4.5241 Accuracy 0.1476 Epoch 3 Batch 450 Loss 4.5147 Accuracy 0.1480 Epoch 3 Batch 500 Loss 4.5027 Accuracy 0.1490 Epoch 3 Batch 550 Loss 4.4878 Accuracy 0.1499 Epoch 3 Batch 600 Loss 4.4730 Accuracy 0.1505 Epoch 3 Batch 650 Loss 4.4613 Accuracy 0.1511 Epoch 3 Batch 700 Loss 4.4487 Accuracy 0.1517 Epoch 3 Loss 4.4482 Accuracy 0.1518 Time taken for 1 epoch: 31.507874250411987 secs Epoch 4 Batch 0 Loss 4.3252 Accuracy 0.1647 Epoch 4 Batch 50 Loss 4.1217 Accuracy 0.1680 Epoch 4 Batch 100 Loss 4.1180 Accuracy 0.1682 Epoch 4 Batch 150 Loss 4.1128 Accuracy 0.1686 Epoch 4 Batch 200 Loss 4.0968 Accuracy 0.1692 Epoch 4 Batch 250 Loss 4.0847 Accuracy 0.1701 Epoch 4 Batch 300 Loss 4.0731 Accuracy 0.1710 Epoch 4 Batch 350 Loss 4.0568 Accuracy 0.1718 Epoch 4 Batch 400 Loss 4.0430 Accuracy 0.1731 Epoch 4 Batch 450 Loss 4.0315 Accuracy 0.1738 Epoch 4 Batch 500 Loss 4.0195 Accuracy 0.1747 Epoch 4 Batch 550 Loss 4.0056 Accuracy 0.1755 Epoch 4 Batch 600 Loss 3.9901 Accuracy 0.1762 Epoch 4 Batch 650 Loss 3.9734 Accuracy 0.1774 Epoch 4 Batch 700 Loss 3.9594 Accuracy 0.1786 Epoch 4 Loss 3.9592 Accuracy 0.1786 Time taken for 1 epoch: 31.130823850631714 secs Epoch 5 Batch 0 Loss 3.6830 Accuracy 0.1795 Epoch 5 Batch 50 Loss 3.6181 Accuracy 0.1984 Epoch 5 Batch 100 Loss 3.6065 Accuracy 0.1990 Epoch 5 Batch 150 Loss 3.5962 Accuracy 0.1996 Epoch 5 Batch 200 Loss 3.5868 Accuracy 0.1995 Epoch 5 Batch 250 Loss 3.5774 Accuracy 0.2002 Epoch 5 Batch 300 Loss 3.5726 Accuracy 0.2004 Epoch 5 Batch 350 Loss 3.5633 Accuracy 0.2004 Epoch 5 Batch 400 Loss 3.5550 Accuracy 0.2012 Epoch 5 Batch 450 Loss 3.5452 Accuracy 0.2016 Epoch 5 Batch 500 Loss 3.5374 Accuracy 0.2023 Epoch 5 Batch 550 Loss 3.5278 Accuracy 0.2026 Epoch 5 Batch 600 Loss 3.5226 Accuracy 0.2032 Epoch 5 Batch 650 Loss 3.5132 Accuracy 0.2040 Epoch 5 Batch 700 Loss 3.5012 Accuracy 0.2045 Saving checkpoint for epoch 5 at ./checkpoints/train/ckpt-1 Epoch 5 Loss 3.5008 Accuracy 0.2045 Time taken for 1 epoch: 31.700332403182983 secs Epoch 6 Batch 0 Loss 3.1302 Accuracy 0.2366 Epoch 6 Batch 50 Loss 3.2096 Accuracy 0.2195 Epoch 6 Batch 100 Loss 3.2002 Accuracy 0.2196 Epoch 6 Batch 150 Loss 3.1916 Accuracy 0.2203 Epoch 6 Batch 200 Loss 3.1857 Accuracy 0.2205 Epoch 6 Batch 250 Loss 3.1802 Accuracy 0.2207 Epoch 6 Batch 300 Loss 3.1734 Accuracy 0.2213 Epoch 6 Batch 350 Loss 3.1619 Accuracy 0.2225 Epoch 6 Batch 400 Loss 3.1556 Accuracy 0.2231 Epoch 6 Batch 450 Loss 3.1496 Accuracy 0.2232 Epoch 6 Batch 500 Loss 3.1455 Accuracy 0.2235 Epoch 6 Batch 550 Loss 3.1386 Accuracy 0.2240 Epoch 6 Batch 600 Loss 3.1317 Accuracy 0.2242 Epoch 6 Batch 650 Loss 3.1265 Accuracy 0.2245 Epoch 6 Batch 700 Loss 3.1181 Accuracy 0.2247 Epoch 6 Loss 3.1179 Accuracy 0.2248 Time taken for 1 epoch: 31.49859118461609 secs Epoch 7 Batch 0 Loss 2.8325 Accuracy 0.2365 Epoch 7 Batch 50 Loss 2.7765 Accuracy 0.2396 Epoch 7 Batch 100 Loss 2.7797 Accuracy 0.2417 Epoch 7 Batch 150 Loss 2.7819 Accuracy 0.2401 Epoch 7 Batch 200 Loss 2.7811 Accuracy 0.2392 Epoch 7 Batch 250 Loss 2.7811 Accuracy 0.2399 Epoch 7 Batch 300 Loss 2.7807 Accuracy 0.2413 Epoch 7 Batch 350 Loss 2.7700 Accuracy 0.2420 Epoch 7 Batch 400 Loss 2.7640 Accuracy 0.2428 Epoch 7 Batch 450 Loss 2.7604 Accuracy 0.2436 Epoch 7 Batch 500 Loss 2.7552 Accuracy 0.2439 Epoch 7 Batch 550 Loss 2.7503 Accuracy 0.2442 Epoch 7 Batch 600 Loss 2.7413 Accuracy 0.2446 Epoch 7 Batch 650 Loss 2.7370 Accuracy 0.2452 Epoch 7 Batch 700 Loss 2.7331 Accuracy 0.2455 Epoch 7 Loss 2.7328 Accuracy 0.2455 Time taken for 1 epoch: 32.07778882980347 secs Epoch 8 Batch 0 Loss 2.4730 Accuracy 0.2611 Epoch 8 Batch 50 Loss 2.4004 Accuracy 0.2635 Epoch 8 Batch 100 Loss 2.4033 Accuracy 0.2650 Epoch 8 Batch 150 Loss 2.4061 Accuracy 0.2635 Epoch 8 Batch 200 Loss 2.4119 Accuracy 0.2624 Epoch 8 Batch 250 Loss 2.4131 Accuracy 0.2634 Epoch 8 Batch 300 Loss 2.4141 Accuracy 0.2632 Epoch 8 Batch 350 Loss 2.4072 Accuracy 0.2641 Epoch 8 Batch 400 Loss 2.4057 Accuracy 0.2643 Epoch 8 Batch 450 Loss 2.4036 Accuracy 0.2643 Epoch 8 Batch 500 Loss 2.4034 Accuracy 0.2644 Epoch 8 Batch 550 Loss 2.4026 Accuracy 0.2642 Epoch 8 Batch 600 Loss 2.4018 Accuracy 0.2642 Epoch 8 Batch 650 Loss 2.4023 Accuracy 0.2642 Epoch 8 Batch 700 Loss 2.4035 Accuracy 0.2647 Epoch 8 Loss 2.4028 Accuracy 0.2648 Time taken for 1 epoch: 30.80423927307129 secs Epoch 9 Batch 0 Loss 2.0968 Accuracy 0.2744 Epoch 9 Batch 50 Loss 2.1334 Accuracy 0.2744 Epoch 9 Batch 100 Loss 2.1473 Accuracy 0.2756 Epoch 9 Batch 150 Loss 2.1541 Accuracy 0.2774 Epoch 9 Batch 200 Loss 2.1594 Accuracy 0.2774 Epoch 9 Batch 250 Loss 2.1627 Accuracy 0.2774 Epoch 9 Batch 300 Loss 2.1658 Accuracy 0.2777 Epoch 9 Batch 350 Loss 2.1666 Accuracy 0.2781 Epoch 9 Batch 400 Loss 2.1599 Accuracy 0.2784 Epoch 9 Batch 450 Loss 2.1587 Accuracy 0.2788 Epoch 9 Batch 500 Loss 2.1618 Accuracy 0.2792 Epoch 9 Batch 550 Loss 2.1626 Accuracy 0.2793 Epoch 9 Batch 600 Loss 2.1612 Accuracy 0.2795 Epoch 9 Batch 650 Loss 2.1628 Accuracy 0.2790 Epoch 9 Batch 700 Loss 2.1646 Accuracy 0.2792 Epoch 9 Loss 2.1647 Accuracy 0.2792 Time taken for 1 epoch: 30.71023654937744 secs Epoch 10 Batch 0 Loss 1.9836 Accuracy 0.3049 Epoch 10 Batch 50 Loss 1.9642 Accuracy 0.2901 Epoch 10 Batch 100 Loss 1.9613 Accuracy 0.2905 Epoch 10 Batch 150 Loss 1.9555 Accuracy 0.2906 Epoch 10 Batch 200 Loss 1.9511 Accuracy 0.2909 Epoch 10 Batch 250 Loss 1.9554 Accuracy 0.2910 Epoch 10 Batch 300 Loss 1.9635 Accuracy 0.2907 Epoch 10 Batch 350 Loss 1.9666 Accuracy 0.2901 Epoch 10 Batch 400 Loss 1.9664 Accuracy 0.2897 Epoch 10 Batch 450 Loss 1.9694 Accuracy 0.2899 Epoch 10 Batch 500 Loss 1.9707 Accuracy 0.2898 Epoch 10 Batch 550 Loss 1.9743 Accuracy 0.2895 Epoch 10 Batch 600 Loss 1.9759 Accuracy 0.2893 Epoch 10 Batch 650 Loss 1.9784 Accuracy 0.2895 Epoch 10 Batch 700 Loss 1.9821 Accuracy 0.2893 Saving checkpoint for epoch 10 at ./checkpoints/train/ckpt-2 Epoch 10 Loss 1.9820 Accuracy 0.2893 Time taken for 1 epoch: 32.27570462226868 secs Epoch 11 Batch 0 Loss 1.5949 Accuracy 0.3141 Epoch 11 Batch 50 Loss 1.8009 Accuracy 0.3034 Epoch 11 Batch 100 Loss 1.8047 Accuracy 0.3008 Epoch 11 Batch 150 Loss 1.8053 Accuracy 0.2990 Epoch 11 Batch 200 Loss 1.8032 Accuracy 0.2988 Epoch 11 Batch 250 Loss 1.8088 Accuracy 0.2982 Epoch 11 Batch 300 Loss 1.8139 Accuracy 0.2991 Epoch 11 Batch 350 Loss 1.8164 Accuracy 0.2992 Epoch 11 Batch 400 Loss 1.8232 Accuracy 0.2994 Epoch 11 Batch 450 Loss 1.8250 Accuracy 0.2991 Epoch 11 Batch 500 Loss 1.8294 Accuracy 0.2989 Epoch 11 Batch 550 Loss 1.8307 Accuracy 0.2987 Epoch 11 Batch 600 Loss 1.8329 Accuracy 0.2985 Epoch 11 Batch 650 Loss 1.8366 Accuracy 0.2987 Epoch 11 Batch 700 Loss 1.8409 Accuracy 0.2985 Epoch 11 Loss 1.8409 Accuracy 0.2985 Time taken for 1 epoch: 31.283374547958374 secs Epoch 12 Batch 0 Loss 1.6747 Accuracy 0.3191 Epoch 12 Batch 50 Loss 1.6553 Accuracy 0.3079 Epoch 12 Batch 100 Loss 1.6594 Accuracy 0.3058 Epoch 12 Batch 150 Loss 1.6612 Accuracy 0.3071 Epoch 12 Batch 200 Loss 1.6711 Accuracy 0.3074 Epoch 12 Batch 250 Loss 1.6760 Accuracy 0.3076 Epoch 12 Batch 300 Loss 1.6845 Accuracy 0.3075 Epoch 12 Batch 350 Loss 1.6890 Accuracy 0.3080 Epoch 12 Batch 400 Loss 1.6916 Accuracy 0.3078 Epoch 12 Batch 450 Loss 1.6960 Accuracy 0.3073 Epoch 12 Batch 500 Loss 1.6997 Accuracy 0.3069 Epoch 12 Batch 550 Loss 1.7079 Accuracy 0.3064 Epoch 12 Batch 600 Loss 1.7122 Accuracy 0.3065 Epoch 12 Batch 650 Loss 1.7158 Accuracy 0.3065 Epoch 12 Batch 700 Loss 1.7206 Accuracy 0.3065 Epoch 12 Loss 1.7204 Accuracy 0.3065 Time taken for 1 epoch: 30.9318790435791 secs Epoch 13 Batch 0 Loss 1.3957 Accuracy 0.3540 Epoch 13 Batch 50 Loss 1.5373 Accuracy 0.3137 Epoch 13 Batch 100 Loss 1.5457 Accuracy 0.3145 Epoch 13 Batch 150 Loss 1.5546 Accuracy 0.3149 Epoch 13 Batch 200 Loss 1.5633 Accuracy 0.3143 Epoch 13 Batch 250 Loss 1.5724 Accuracy 0.3144 Epoch 13 Batch 300 Loss 1.5765 Accuracy 0.3141 Epoch 13 Batch 350 Loss 1.5818 Accuracy 0.3142 Epoch 13 Batch 400 Loss 1.5889 Accuracy 0.3137 Epoch 13 Batch 450 Loss 1.5910 Accuracy 0.3133 Epoch 13 Batch 500 Loss 1.5959 Accuracy 0.3134 Epoch 13 Batch 550 Loss 1.6030 Accuracy 0.3130 Epoch 13 Batch 600 Loss 1.6071 Accuracy 0.3129 Epoch 13 Batch 650 Loss 1.6129 Accuracy 0.3126 Epoch 13 Batch 700 Loss 1.6172 Accuracy 0.3126 Epoch 13 Loss 1.6171 Accuracy 0.3127 Time taken for 1 epoch: 30.743946313858032 secs Epoch 14 Batch 0 Loss 1.5949 Accuracy 0.3139 Epoch 14 Batch 50 Loss 1.4416 Accuracy 0.3248 Epoch 14 Batch 100 Loss 1.4612 Accuracy 0.3218 Epoch 14 Batch 150 Loss 1.4728 Accuracy 0.3213 Epoch 14 Batch 200 Loss 1.4790 Accuracy 0.3205 Epoch 14 Batch 250 Loss 1.4818 Accuracy 0.3207 Epoch 14 Batch 300 Loss 1.4895 Accuracy 0.3206 Epoch 14 Batch 350 Loss 1.4966 Accuracy 0.3205 Epoch 14 Batch 400 Loss 1.5002 Accuracy 0.3202 Epoch 14 Batch 450 Loss 1.5042 Accuracy 0.3201 Epoch 14 Batch 500 Loss 1.5094 Accuracy 0.3201 Epoch 14 Batch 550 Loss 1.5125 Accuracy 0.3198 Epoch 14 Batch 600 Loss 1.5195 Accuracy 0.3196 Epoch 14 Batch 650 Loss 1.5247 Accuracy 0.3190 Epoch 14 Batch 700 Loss 1.5309 Accuracy 0.3188 Epoch 14 Loss 1.5311 Accuracy 0.3189 Time taken for 1 epoch: 30.723429203033447 secs Epoch 15 Batch 0 Loss 1.3458 Accuracy 0.3471 Epoch 15 Batch 50 Loss 1.3494 Accuracy 0.3302 Epoch 15 Batch 100 Loss 1.3686 Accuracy 0.3321 Epoch 15 Batch 150 Loss 1.3868 Accuracy 0.3311 Epoch 15 Batch 200 Loss 1.4009 Accuracy 0.3297 Epoch 15 Batch 250 Loss 1.4054 Accuracy 0.3280 Epoch 15 Batch 300 Loss 1.4146 Accuracy 0.3272 Epoch 15 Batch 350 Loss 1.4191 Accuracy 0.3259 Epoch 15 Batch 400 Loss 1.4261 Accuracy 0.3251 Epoch 15 Batch 450 Loss 1.4307 Accuracy 0.3244 Epoch 15 Batch 500 Loss 1.4368 Accuracy 0.3241 Epoch 15 Batch 550 Loss 1.4421 Accuracy 0.3242 Epoch 15 Batch 600 Loss 1.4471 Accuracy 0.3235 Epoch 15 Batch 650 Loss 1.4507 Accuracy 0.3234 Epoch 15 Batch 700 Loss 1.4552 Accuracy 0.3234 Saving checkpoint for epoch 15 at ./checkpoints/train/ckpt-3 Epoch 15 Loss 1.4556 Accuracy 0.3233 Time taken for 1 epoch: 31.18160319328308 secs Epoch 16 Batch 0 Loss 1.2760 Accuracy 0.3277 Epoch 16 Batch 50 Loss 1.2938 Accuracy 0.3330 Epoch 16 Batch 100 Loss 1.3032 Accuracy 0.3326 Epoch 16 Batch 150 Loss 1.3188 Accuracy 0.3313 Epoch 16 Batch 200 Loss 1.3277 Accuracy 0.3314 Epoch 16 Batch 250 Loss 1.3352 Accuracy 0.3315 Epoch 16 Batch 300 Loss 1.3442 Accuracy 0.3314 Epoch 16 Batch 350 Loss 1.3498 Accuracy 0.3311 Epoch 16 Batch 400 Loss 1.3542 Accuracy 0.3309 Epoch 16 Batch 450 Loss 1.3592 Accuracy 0.3307 Epoch 16 Batch 500 Loss 1.3629 Accuracy 0.3308 Epoch 16 Batch 550 Loss 1.3688 Accuracy 0.3311 Epoch 16 Batch 600 Loss 1.3730 Accuracy 0.3307 Epoch 16 Batch 650 Loss 1.3797 Accuracy 0.3299 Epoch 16 Batch 700 Loss 1.3847 Accuracy 0.3294 Epoch 16 Loss 1.3848 Accuracy 0.3294 Time taken for 1 epoch: 30.866195917129517 secs Epoch 17 Batch 0 Loss 1.1798 Accuracy 0.3294 Epoch 17 Batch 50 Loss 1.2607 Accuracy 0.3333 Epoch 17 Batch 100 Loss 1.2647 Accuracy 0.3357 Epoch 17 Batch 150 Loss 1.2731 Accuracy 0.3341 Epoch 17 Batch 200 Loss 1.2743 Accuracy 0.3351 Epoch 17 Batch 250 Loss 1.2779 Accuracy 0.3348 Epoch 17 Batch 300 Loss 1.2827 Accuracy 0.3352 Epoch 17 Batch 350 Loss 1.2918 Accuracy 0.3341 Epoch 17 Batch 400 Loss 1.2941 Accuracy 0.3349 Epoch 17 Batch 450 Loss 1.2990 Accuracy 0.3340 Epoch 17 Batch 500 Loss 1.3035 Accuracy 0.3336 Epoch 17 Batch 550 Loss 1.3091 Accuracy 0.3338 Epoch 17 Batch 600 Loss 1.3152 Accuracy 0.3336 Epoch 17 Batch 650 Loss 1.3190 Accuracy 0.3329 Epoch 17 Batch 700 Loss 1.3243 Accuracy 0.3325 Epoch 17 Loss 1.3247 Accuracy 0.3324 Time taken for 1 epoch: 30.773754358291626 secs Epoch 18 Batch 0 Loss 1.1299 Accuracy 0.3285 Epoch 18 Batch 50 Loss 1.1815 Accuracy 0.3433 Epoch 18 Batch 100 Loss 1.1968 Accuracy 0.3419 Epoch 18 Batch 150 Loss 1.2051 Accuracy 0.3410 Epoch 18 Batch 200 Loss 1.2135 Accuracy 0.3398 Epoch 18 Batch 250 Loss 1.2218 Accuracy 0.3397 Epoch 18 Batch 300 Loss 1.2270 Accuracy 0.3394 Epoch 18 Batch 350 Loss 1.2354 Accuracy 0.3384 Epoch 18 Batch 400 Loss 1.2405 Accuracy 0.3388 Epoch 18 Batch 450 Loss 1.2453 Accuracy 0.3392 Epoch 18 Batch 500 Loss 1.2508 Accuracy 0.3385 Epoch 18 Batch 550 Loss 1.2560 Accuracy 0.3378 Epoch 18 Batch 600 Loss 1.2608 Accuracy 0.3375 Epoch 18 Batch 650 Loss 1.2654 Accuracy 0.3375 Epoch 18 Batch 700 Loss 1.2718 Accuracy 0.3370 Epoch 18 Loss 1.2721 Accuracy 0.3369 Time taken for 1 epoch: 30.90894341468811 secs Epoch 19 Batch 0 Loss 1.1891 Accuracy 0.3750 Epoch 19 Batch 50 Loss 1.1526 Accuracy 0.3441 Epoch 19 Batch 100 Loss 1.1434 Accuracy 0.3427 Epoch 19 Batch 150 Loss 1.1503 Accuracy 0.3430 Epoch 19 Batch 200 Loss 1.1565 Accuracy 0.3432 Epoch 19 Batch 250 Loss 1.1654 Accuracy 0.3430 Epoch 19 Batch 300 Loss 1.1775 Accuracy 0.3422 Epoch 19 Batch 350 Loss 1.1866 Accuracy 0.3424 Epoch 19 Batch 400 Loss 1.1908 Accuracy 0.3422 Epoch 19 Batch 450 Loss 1.1935 Accuracy 0.3419 Epoch 19 Batch 500 Loss 1.1990 Accuracy 0.3416 Epoch 19 Batch 550 Loss 1.2045 Accuracy 0.3417 Epoch 19 Batch 600 Loss 1.2105 Accuracy 0.3413 Epoch 19 Batch 650 Loss 1.2165 Accuracy 0.3407 Epoch 19 Batch 700 Loss 1.2228 Accuracy 0.3405 Epoch 19 Loss 1.2230 Accuracy 0.3405 Time taken for 1 epoch: 30.853646993637085 secs Epoch 20 Batch 0 Loss 1.0539 Accuracy 0.4047 Epoch 20 Batch 50 Loss 1.0953 Accuracy 0.3540 Epoch 20 Batch 100 Loss 1.0994 Accuracy 0.3487 Epoch 20 Batch 150 Loss 1.1061 Accuracy 0.3480 Epoch 20 Batch 200 Loss 1.1146 Accuracy 0.3470 Epoch 20 Batch 250 Loss 1.1211 Accuracy 0.3477 Epoch 20 Batch 300 Loss 1.1244 Accuracy 0.3478 Epoch 20 Batch 350 Loss 1.1362 Accuracy 0.3472 Epoch 20 Batch 400 Loss 1.1404 Accuracy 0.3472 Epoch 20 Batch 450 Loss 1.1451 Accuracy 0.3467 Epoch 20 Batch 500 Loss 1.1528 Accuracy 0.3458 Epoch 20 Batch 550 Loss 1.1592 Accuracy 0.3452 Epoch 20 Batch 600 Loss 1.1657 Accuracy 0.3443 Epoch 20 Batch 650 Loss 1.1719 Accuracy 0.3437 Epoch 20 Batch 700 Loss 1.1766 Accuracy 0.3440 Saving checkpoint for epoch 20 at ./checkpoints/train/ckpt-4 Epoch 20 Loss 1.1768 Accuracy 0.3440 Time taken for 1 epoch: 32.28810262680054 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
decoderthen outputs the predictions by looking at theencoder outputand 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 the top problem . 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 be very quickly with you some very magical stories that happened . 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 that i did .
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.