Help protect the Great Barrier Reef with TensorFlow on Kaggle Join Challenge

Text generation with an RNN

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

This tutorial demonstrates how to generate text using a character-based RNN. You will work with a dataset of Shakespeare's writing from Andrej Karpathy's The Unreasonable Effectiveness of Recurrent Neural Networks. Given a sequence of characters from this data ("Shakespear"), train a model to predict the next character in the sequence ("e"). Longer sequences of text can be generated by calling the model repeatedly.

This tutorial includes runnable code implemented using tf.keras and eager execution. The following is the sample output when the model in this tutorial trained for 30 epochs, and started with the prompt "Q":

QUEENE:
I had thought thou hadst a Roman; for the oracle,
Thus by All bids the man against the word,
Which are so weak of care, by old care done;
Your children were in your holy love,
And the precipitation through the bleeding throne.

BISHOP OF ELY:
Marry, and will, my lord, to weep in such a one were prettiest;
Yet now I was adopted heir
Of the world's lamentable day,
To watch the next way with his father with his face?

ESCALUS:
The cause why then we are all resolved more sons.

VOLUMNIA:
O, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, it is no sin it should be dead,
And love and pale as any will to that word.

QUEEN ELIZABETH:
But how long have I heard the soul for this world,
And show his hands of life be proved to stand.

PETRUCHIO:
I say he look'd on, if I must be content
To stay him from the fatal of our country's bliss.
His lordship pluck'd from this sentence then for prey,
And then let us twain, being the moon,
were she such a case as fills m

While some of the sentences are grammatical, most do not make sense. The model has not learned the meaning of words, but consider:

  • The model is character-based. When training started, the model did not know how to spell an English word, or that words were even a unit of text.

  • The structure of the output resembles a play—blocks of text generally begin with a speaker name, in all capital letters similar to the dataset.

  • As demonstrated below, the model is trained on small batches of text (100 characters each), and is still able to generate a longer sequence of text with coherent structure.

Setup

Import TensorFlow and other libraries

import tensorflow as tf

import numpy as np
import os
import time

Download the Shakespeare dataset

Change the following line to run this code on your own data.

path_to_file = tf.keras.utils.get_file('shakespeare.txt', 'https://storage.googleapis.com/download.tensorflow.org/data/shakespeare.txt')
Downloading data from https://storage.googleapis.com/download.tensorflow.org/data/shakespeare.txt
1122304/1115394 [==============================] - 0s 0us/step
1130496/1115394 [==============================] - 0s 0us/step

Read the data

First, look in the text:

# Read, then decode for py2 compat.
text = open(path_to_file, 'rb').read().decode(encoding='utf-8')
# length of text is the number of characters in it
print(f'Length of text: {len(text)} characters')
Length of text: 1115394 characters
# Take a look at the first 250 characters in text
print(text[:250])
First Citizen:
Before we proceed any further, hear me speak.

All:
Speak, speak.

First Citizen:
You are all resolved rather to die than to famish?

All:
Resolved. resolved.

First Citizen:
First, you know Caius Marcius is chief enemy to the people.
# The unique characters in the file
vocab = sorted(set(text))
print(f'{len(vocab)} unique characters')
65 unique characters

Process the text

Vectorize the text

Before training, you need to convert the strings to a numerical representation.

The tf.keras.layers.StringLookup layer can convert each character into a numeric ID. It just needs the text to be split into tokens first.

example_texts = ['abcdefg', 'xyz']

chars = tf.strings.unicode_split(example_texts, input_encoding='UTF-8')
chars
<tf.RaggedTensor [[b'a', b'b', b'c', b'd', b'e', b'f', b'g'], [b'x', b'y', b'z']]>

Now create the tf.keras.layers.StringLookup layer:

ids_from_chars = tf.keras.layers.StringLookup(
    vocabulary=list(vocab), mask_token=None)

It converts from tokens to character IDs:

ids = ids_from_chars(chars)
ids
<tf.RaggedTensor [[40, 41, 42, 43, 44, 45, 46], [63, 64, 65]]>

Since the goal of this tutorial is to generate text, it will also be important to invert this representation and recover human-readable strings from it. For this you can use tf.keras.layers.StringLookup(..., invert=True).

chars_from_ids = tf.keras.layers.StringLookup(
    vocabulary=ids_from_chars.get_vocabulary(), invert=True, mask_token=None)

This layer recovers the characters from the vectors of IDs, and returns them as a tf.RaggedTensor of characters:

chars = chars_from_ids(ids)
chars
<tf.RaggedTensor [[b'a', b'b', b'c', b'd', b'e', b'f', b'g'], [b'x', b'y', b'z']]>

You can tf.strings.reduce_join to join the characters back into strings.

tf.strings.reduce_join(chars, axis=-1).numpy()
array([b'abcdefg', b'xyz'], dtype=object)
def text_from_ids(ids):
  return tf.strings.reduce_join(chars_from_ids(ids), axis=-1)

The prediction task

Given a character, or a sequence of characters, what is the most probable next character? This is the task you're training the model to perform. The input to the model will be a sequence of characters, and you train the model to predict the output—the following character at each time step.

Since RNNs maintain an internal state that depends on the previously seen elements, given all the characters computed until this moment, what is the next character?

Create training examples and targets

Next divide the text into example sequences. Each input sequence will contain seq_length characters from the text.

For each input sequence, the corresponding targets contain the same length of text, except shifted one character to the right.

So break the text into chunks of seq_length+1. For example, say seq_length is 4 and our text is "Hello". The input sequence would be "Hell", and the target sequence "ello".

To do this first use the tf.data.Dataset.from_tensor_slices function to convert the text vector into a stream of character indices.

all_ids = ids_from_chars(tf.strings.unicode_split(text, 'UTF-8'))
all_ids
<tf.Tensor: shape=(1115394,), dtype=int64, numpy=array([19, 48, 57, ..., 46,  9,  1])>
ids_dataset = tf.data.Dataset.from_tensor_slices(all_ids)
for ids in ids_dataset.take(10):
    print(chars_from_ids(ids).numpy().decode('utf-8'))
F
i
r
s
t
 
C
i
t
i
seq_length = 100
examples_per_epoch = len(text)//(seq_length+1)

The batch method lets you easily convert these individual characters to sequences of the desired size.

sequences = ids_dataset.batch(seq_length+1, drop_remainder=True)

for seq in sequences.take(1):
  print(chars_from_ids(seq))
tf.Tensor(
[b'F' b'i' b'r' b's' b't' b' ' b'C' b'i' b't' b'i' b'z' b'e' b'n' b':'
 b'\n' b'B' b'e' b'f' b'o' b'r' b'e' b' ' b'w' b'e' b' ' b'p' b'r' b'o'
 b'c' b'e' b'e' b'd' b' ' b'a' b'n' b'y' b' ' b'f' b'u' b'r' b't' b'h'
 b'e' b'r' b',' b' ' b'h' b'e' b'a' b'r' b' ' b'm' b'e' b' ' b's' b'p'
 b'e' b'a' b'k' b'.' b'\n' b'\n' b'A' b'l' b'l' b':' b'\n' b'S' b'p' b'e'
 b'a' b'k' b',' b' ' b's' b'p' b'e' b'a' b'k' b'.' b'\n' b'\n' b'F' b'i'
 b'r' b's' b't' b' ' b'C' b'i' b't' b'i' b'z' b'e' b'n' b':' b'\n' b'Y'
 b'o' b'u' b' '], shape=(101,), dtype=string)
2022-01-14 12:12:45.336250: W tensorflow/core/data/root_dataset.cc:200] Optimization loop failed: CANCELLED: Operation was cancelled

It's easier to see what this is doing if you join the tokens back into strings:

for seq in sequences.take(5):
  print(text_from_ids(seq).numpy())
b'First Citizen:\nBefore we proceed any further, hear me speak.\n\nAll:\nSpeak, speak.\n\nFirst Citizen:\nYou '
b'are all resolved rather to die than to famish?\n\nAll:\nResolved. resolved.\n\nFirst Citizen:\nFirst, you k'
b"now Caius Marcius is chief enemy to the people.\n\nAll:\nWe know't, we know't.\n\nFirst Citizen:\nLet us ki"
b"ll him, and we'll have corn at our own price.\nIs't a verdict?\n\nAll:\nNo more talking on't; let it be d"
b'one: away, away!\n\nSecond Citizen:\nOne word, good citizens.\n\nFirst Citizen:\nWe are accounted poor citi'

For training you'll need a dataset of (input, label) pairs. Where input and label are sequences. At each time step the input is the current character and the label is the next character.

Here's a function that takes a sequence as input, duplicates, and shifts it to align the input and label for each timestep:

def split_input_target(sequence):
    input_text = sequence[:-1]
    target_text = sequence[1:]
    return input_text, target_text
split_input_target(list("Tensorflow"))
(['T', 'e', 'n', 's', 'o', 'r', 'f', 'l', 'o'],
 ['e', 'n', 's', 'o', 'r', 'f', 'l', 'o', 'w'])
dataset = sequences.map(split_input_target)
for input_example, target_example in dataset.take(1):
    print("Input :", text_from_ids(input_example).numpy())
    print("Target:", text_from_ids(target_example).numpy())
Input : b'First Citizen:\nBefore we proceed any further, hear me speak.\n\nAll:\nSpeak, speak.\n\nFirst Citizen:\nYou'
Target: b'irst Citizen:\nBefore we proceed any further, hear me speak.\n\nAll:\nSpeak, speak.\n\nFirst Citizen:\nYou '

Create training batches

You used tf.data to split the text into manageable sequences. But before feeding this data into the model, you need to shuffle the data and pack it into batches.

# Batch size
BATCH_SIZE = 64

# Buffer size to shuffle the dataset
# (TF data is designed to work with possibly infinite sequences,
# so it doesn't attempt to shuffle the entire sequence in memory. Instead,
# it maintains a buffer in which it shuffles elements).
BUFFER_SIZE = 10000

dataset = (
    dataset
    .shuffle(BUFFER_SIZE)
    .batch(BATCH_SIZE, drop_remainder=True)
    .prefetch(tf.data.experimental.AUTOTUNE))

dataset
<PrefetchDataset element_spec=(TensorSpec(shape=(64, 100), dtype=tf.int64, name=None), TensorSpec(shape=(64, 100), dtype=tf.int64, name=None))>

Build The Model

This section defines the model as a keras.Model subclass (For details see Making new Layers and Models via subclassing).

This model has three layers:

  • tf.keras.layers.Embedding: The input layer. A trainable lookup table that will map each character-ID to a vector with embedding_dim dimensions;
  • tf.keras.layers.GRU: A type of RNN with size units=rnn_units (You can also use an LSTM layer here.)
  • tf.keras.layers.Dense: The output layer, with vocab_size outputs. It outputs one logit for each character in the vocabulary. These are the log-likelihood of each character according to the model.
# Length of the vocabulary in chars
vocab_size = len(vocab)

# The embedding dimension
embedding_dim = 256

# Number of RNN units
rnn_units = 1024
class MyModel(tf.keras.Model):
  def __init__(self, vocab_size, embedding_dim, rnn_units):
    super().__init__(self)
    self.embedding = tf.keras.layers.Embedding(vocab_size, embedding_dim)
    self.gru = tf.keras.layers.GRU(rnn_units,
                                   return_sequences=True,
                                   return_state=True)
    self.dense = tf.keras.layers.Dense(vocab_size)

  def call(self, inputs, states=None, return_state=False, training=False):
    x = inputs
    x = self.embedding(x, training=training)
    if states is None:
      states = self.gru.get_initial_state(x)
    x, states = self.gru(x, initial_state=states, training=training)
    x = self.dense(x, training=training)

    if return_state:
      return x, states
    else:
      return x
model = MyModel(
    # Be sure the vocabulary size matches the `StringLookup` layers.
    vocab_size=len(ids_from_chars.get_vocabulary()),
    embedding_dim=embedding_dim,
    rnn_units=rnn_units)

For each character the model looks up the embedding, runs the GRU one timestep with the embedding as input, and applies the dense layer to generate logits predicting the log-likelihood of the next character:

A drawing of the data passing through the model

Try the model

Now run the model to see that it behaves as expected.

First check the shape of the output:

for input_example_batch, target_example_batch in dataset.take(1):
    example_batch_predictions = model(input_example_batch)
    print(example_batch_predictions.shape, "# (batch_size, sequence_length, vocab_size)")
(64, 100, 66) # (batch_size, sequence_length, vocab_size)

In the above example the sequence length of the input is 100 but the model can be run on inputs of any length:

model.summary()
Model: "my_model"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 embedding (Embedding)       multiple                  16896     
                                                                 
 gru (GRU)                   multiple                  3938304   
                                                                 
 dense (Dense)               multiple                  67650     
                                                                 
=================================================================
Total params: 4,022,850
Trainable params: 4,022,850
Non-trainable params: 0
_________________________________________________________________

To get actual predictions from the model you need to sample from the output distribution, to get actual character indices. This distribution is defined by the logits over the character vocabulary.

Try it for the first example in the batch:

sampled_indices = tf.random.categorical(example_batch_predictions[0], num_samples=1)
sampled_indices = tf.squeeze(sampled_indices, axis=-1).numpy()

This gives us, at each timestep, a prediction of the next character index:

sampled_indices
array([ 3, 64, 58, 50, 52, 54,  5, 65, 44, 52, 11, 17, 16, 22, 62, 14, 19,
       26, 63,  2, 39, 19, 21, 14, 39, 46, 26, 53,  4, 16, 49, 43, 15, 53,
       31, 45,  2,  9, 43, 53,  6,  0, 43, 30, 24,  2, 35, 22, 23,  3, 17,
       15, 10, 30, 42, 48, 22,  1, 27, 52, 36, 51, 51, 43, 65, 39, 43, 38,
       39,  0,  2, 56, 49, 48, 31, 42, 51, 48,  7, 24, 33, 56, 58, 31, 51,
       61, 59, 61,  3, 65, 44, 14, 34, 30, 44, 27, 47, 18, 42, 15])

Decode these to see the text predicted by this untrained model:

print("Input:\n", text_from_ids(input_example_batch[0]).numpy())
print()
print("Next Char Predictions:\n", text_from_ids(sampled_indices).numpy())
Input:
 b'genet.\nEdward for Edward pays a dying debt.\n\nQUEEN ELIZABETH:\nWilt thou, O God, fly from such gentle'

Next Char Predictions:
 b"!yskmo&zem:DCIwAFMx ZFHAZgMn$CjdBnRf .dn'[UNK]dQK VIJ!DB3QciI\nNmWlldzZdYZ[UNK] qjiRcli,KTqsRlvtv!zeAUQeNhEcB"

Train the model

At this point the problem can be treated as a standard classification problem. Given the previous RNN state, and the input this time step, predict the class of the next character.

Attach an optimizer, and a loss function

The standard tf.keras.losses.sparse_categorical_crossentropy loss function works in this case because it is applied across the last dimension of the predictions.

Because your model returns logits, you need to set the from_logits flag.

loss = tf.losses.SparseCategoricalCrossentropy(from_logits=True)
example_batch_loss = loss(target_example_batch, example_batch_predictions)
mean_loss = example_batch_loss.numpy().mean()
print("Prediction shape: ", example_batch_predictions.shape, " # (batch_size, sequence_length, vocab_size)")
print("Mean loss:        ", mean_loss)
Prediction shape:  (64, 100, 66)  # (batch_size, sequence_length, vocab_size)
Mean loss:         4.1900826

A newly initialized model shouldn't be too sure of itself, the output logits should all have similar magnitudes. To confirm this you can check that the exponential of the mean loss is approximately equal to the vocabulary size. A much higher loss means the model is sure of its wrong answers, and is badly initialized:

tf.exp(mean_loss).numpy()
66.028244

Configure the training procedure using the tf.keras.Model.compile method. Use tf.keras.optimizers.Adam with default arguments and the loss function.

model.compile(optimizer='adam', loss=loss)

Configure checkpoints

Use a tf.keras.callbacks.ModelCheckpoint to ensure that checkpoints are saved during training:

# Directory where the checkpoints will be saved
checkpoint_dir = './training_checkpoints'
# Name of the checkpoint files
checkpoint_prefix = os.path.join(checkpoint_dir, "ckpt_{epoch}")

checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
    filepath=checkpoint_prefix,
    save_weights_only=True)

Execute the training

To keep training time reasonable, use 10 epochs to train the model. In Colab, set the runtime to GPU for faster training.

EPOCHS = 20
history = model.fit(dataset, epochs=EPOCHS, callbacks=[checkpoint_callback])
Epoch 1/20
172/172 [==============================] - 7s 25ms/step - loss: 2.7056
Epoch 2/20
172/172 [==============================] - 5s 24ms/step - loss: 1.9838
Epoch 3/20
172/172 [==============================] - 5s 24ms/step - loss: 1.7074
Epoch 4/20
172/172 [==============================] - 5s 24ms/step - loss: 1.5489
Epoch 5/20
172/172 [==============================] - 5s 24ms/step - loss: 1.4498
Epoch 6/20
172/172 [==============================] - 5s 24ms/step - loss: 1.3825
Epoch 7/20
172/172 [==============================] - 5s 24ms/step - loss: 1.3300
Epoch 8/20
172/172 [==============================] - 5s 24ms/step - loss: 1.2846
Epoch 9/20
172/172 [==============================] - 5s 24ms/step - loss: 1.2441
Epoch 10/20
172/172 [==============================] - 5s 24ms/step - loss: 1.2030
Epoch 11/20
172/172 [==============================] - 5s 24ms/step - loss: 1.1641
Epoch 12/20
172/172 [==============================] - 5s 24ms/step - loss: 1.1223
Epoch 13/20
172/172 [==============================] - 6s 24ms/step - loss: 1.0786
Epoch 14/20
172/172 [==============================] - 5s 24ms/step - loss: 1.0316
Epoch 15/20
172/172 [==============================] - 6s 24ms/step - loss: 0.9826
Epoch 16/20
172/172 [==============================] - 5s 24ms/step - loss: 0.9326
Epoch 17/20
172/172 [==============================] - 5s 24ms/step - loss: 0.8790
Epoch 18/20
172/172 [==============================] - 5s 24ms/step - loss: 0.8260
Epoch 19/20
172/172 [==============================] - 5s 24ms/step - loss: 0.7759
Epoch 20/20
172/172 [==============================] - 5s 24ms/step - loss: 0.7256

Generate text

The simplest way to generate text with this model is to run it in a loop, and keep track of the model's internal state as you execute it.

To generate text the model's output is fed back to the input

Each time you call the model you pass in some text and an internal state. The model returns a prediction for the next character and its new state. Pass the prediction and state back in to continue generating text.

The following makes a single step prediction:

class OneStep(tf.keras.Model):
  def __init__(self, model, chars_from_ids, ids_from_chars, temperature=1.0):
    super().__init__()
    self.temperature = temperature
    self.model = model
    self.chars_from_ids = chars_from_ids
    self.ids_from_chars = ids_from_chars

    # Create a mask to prevent "[UNK]" from being generated.
    skip_ids = self.ids_from_chars(['[UNK]'])[:, None]
    sparse_mask = tf.SparseTensor(
        # Put a -inf at each bad index.
        values=[-float('inf')]*len(skip_ids),
        indices=skip_ids,
        # Match the shape to the vocabulary
        dense_shape=[len(ids_from_chars.get_vocabulary())])
    self.prediction_mask = tf.sparse.to_dense(sparse_mask)

  @tf.function
  def generate_one_step(self, inputs, states=None):
    # Convert strings to token IDs.
    input_chars = tf.strings.unicode_split(inputs, 'UTF-8')
    input_ids = self.ids_from_chars(input_chars).to_tensor()

    # Run the model.
    # predicted_logits.shape is [batch, char, next_char_logits]
    predicted_logits, states = self.model(inputs=input_ids, states=states,
                                          return_state=True)
    # Only use the last prediction.
    predicted_logits = predicted_logits[:, -1, :]
    predicted_logits = predicted_logits/self.temperature
    # Apply the prediction mask: prevent "[UNK]" from being generated.
    predicted_logits = predicted_logits + self.prediction_mask

    # Sample the output logits to generate token IDs.
    predicted_ids = tf.random.categorical(predicted_logits, num_samples=1)
    predicted_ids = tf.squeeze(predicted_ids, axis=-1)

    # Convert from token ids to characters
    predicted_chars = self.chars_from_ids(predicted_ids)

    # Return the characters and model state.
    return predicted_chars, states
one_step_model = OneStep(model, chars_from_ids, ids_from_chars)

Run it in a loop to generate some text. Looking at the generated text, you'll see the model knows when to capitalize, make paragraphs and imitates a Shakespeare-like writing vocabulary. With the small number of training epochs, it has not yet learned to form coherent sentences.

start = time.time()
states = None
next_char = tf.constant(['ROMEO:'])
result = [next_char]

for n in range(1000):
  next_char, states = one_step_model.generate_one_step(next_char, states=states)
  result.append(next_char)

result = tf.strings.join(result)
end = time.time()
print(result[0].numpy().decode('utf-8'), '\n\n' + '_'*80)
print('\nRun time:', end - start)
ROMEO:
But I do lose his land, and get a summer ruin.

SICINIUS:
That's your beave.

BIONDELLO:
Ha? Rucher that dost thou art again to provide,
In thy right injurious princely number.
Dear resembrance at this Antium, York as a man
That 'leven your steel pedlars mounted.
Sad, he hath privilege with the causer.

NORTHUMBERLAND:
The lords o'clock my sovereign slay;
You say do so young master. Come, let's go
Avoid with brittle fifteen years; therefore I wail, for he
is your before, sir, I will deep thing and rich my steal father.
Then, till he be that must be done: Henry, such a peast,
Our doings were free,' quoth he that Clifford say so himself.
If I be here to make thee jardon, 'milis,
And cannot but better wed this withded; but,
I'll pawn the platts, to France Henry's forward come?

BENVOLIO:
Twice one defend,
Or wail not be; this last of endless creatures
We may title out above to due, a feast.

Tailor:
Such a more mortal to your death
Taking this forward injury, that did, but hears
Bearing  

________________________________________________________________________________

Run time: 2.630192518234253

The easiest thing you can do to improve the results is to train it for longer (try EPOCHS = 30).

You can also experiment with a different start string, try adding another RNN layer to improve the model's accuracy, or adjust the temperature parameter to generate more or less random predictions.

If you want the model to generate text faster the easiest thing you can do is batch the text generation. In the example below the model generates 5 outputs in about the same time it took to generate 1 above.

start = time.time()
states = None
next_char = tf.constant(['ROMEO:', 'ROMEO:', 'ROMEO:', 'ROMEO:', 'ROMEO:'])
result = [next_char]

for n in range(1000):
  next_char, states = one_step_model.generate_one_step(next_char, states=states)
  result.append(next_char)

result = tf.strings.join(result)
end = time.time()
print(result, '\n\n' + '_'*80)
print('\nRun time:', end - start)
tf.Tensor(
[b"ROMEO:\nThe Duke of Norfolk, when ang low-hingded is a\ndangerous tongue; thou hast never in our tale\n'side, and talk of wine and me. Leave me, and\nI wish thy breathing: he hath had no poison\nMust call Edward's faces, and that sends the roan drowning tears.\nThe sum our lovers bride, or bid me speak.\n\nBUCKINGHAM:\nWelcome, great bring our contents.\n\nCAMILLO:\nThe kings of Coriolanus!\n\nVOLUMNIA:\nAnd, by a dish, sort words in person.\n\nBRUTUS:\nGo ask.\n\nMARIANA:\nThen with a grains what thine eyes say\nWill you done free speech lies?\n\nLADY CAPULET:\nI know not what to be proud?\nFow, I do bit speak a worthy feast; but tweit me\nForbid-hearken more deeds do that presently to't in no\nmore mad off, what a subjected\nregion and the devil serve my name world's dead,\nAnd till the portier rare my brows, not with this fable,\nnurse, for he is come to swear and spee himself and France:\nClarence hath all postery of the marriage\nBoth have but a rather.\n\nDUKE VINCENTIO:\nYou have lats in the charm? the two brittle gale\n"
 b"ROMEO:\nThey are but foolish tox, that you rogues!\n\nARCHIDAMUS:\nWell, why lie yon ege you for that name; for forward of the father\nWoeed out me in the invited suffer: if\nThe Volsces have by me with twenty thousand men?\nBut he, my lord, with honourable forfeit,\nWhich ebbles your partiest, will do you\nTo make the measure of a weal Lord Hastings;\nI will not judger in Bohemia.\n\nCOMINIUS:\nGo, be gone, let's he for her success.\n\nMIRANDA:\nYou have told something only poored?\nWhere then Ramentione hath wrong'd himself.\n\nKING RICHARD III:\nHe was not may be, do not bellay. Here comest thou not:\nWhat a medler sticks your daughter and my slave,\nAs if thy words may talk for love, thy knee\nCitsand in't, for fair lady's husband,\nTo entail yield Lohe Becomes your bosom's side;\ncannot hold out alone. Farewell: my swift is several tone,\nThat rests more cause to go walk by to her hearing a trabe,\nOr ever eished him that walk'd you, whiles he yields\nSmiles as easy ansien and be true.\n\nDUCHESS OF YORK:\nAll men, I"
 b"ROMEO:\nBut rather we have eved the ground\nbelieve thee, horse! what has left in the city butchers.\n\nDUKE VINCENTIO:\nA goddess helprous Derby, and yet no parley, complet\nThe valiant part Campish. Tell me, mark I commine\nYour changet over-little.\n\nPOMPEY:\nI would fail sir, and here I leave you, sir.\n\nVINCENTIO:\nLet a mark to-morrow to be ta'en true; be gone.\n\nShepherd:\nNo, he doth no daughter to part.\n\nPAULINA:\nWhere is my lord Ged breed!\nO go, no visage, and fear:\nYou are school'd, methinks, and finds enough.\nWhat, are you heep? when valiant Bilond, owe hour,\nSince presently the lording jewels each part\nThe honour's emilia. A scurcely, which are\nthey are, which Hasby Buckingham deposed?\n\nRome:\nAy, to thy child is frame of death,\nBut death to part of satisffiting.\n\nNORTHUMBERLAND:\nNay, but sad, I'ld not be so troop;\nShow me them speak. This way, let's hand thy pride.\n\nHENRY BOLINGBROKE:\nHe did, my ladying and dig friend,\nNor to thy heart were god, that I have seen in\ntwenty humours live. For j"
 b"ROMEO:\nThe mayor alieas! no senses of spleen,\nWhile I am come to Padua and gallows Claudio; we is there\nA shimble witness to this father's death\n'seep lusty to Covents, alive,\nSend by the father bear themselves; for I am law,\nAnd there not mine is no other beauteous days!\n\nJOHN OF GAUNT:\nO, but O, the account I would content\na boain-to bed; there as good news will it purge;\nFor then the unstable's accept of teet, and aged\nThan one but follow'd that he did the house\nto make a papse tread on to dispatch his power to bid\nOf the tribunes of ut my sweet revolting.\n\nVOLUMNIA:\nMy husband come I heard you as from an\nangry and a father before him\nTo be so boundly, as affairs\nShall rest by back from grief and fortune shall be talk.\n\nGLOUCESTER:\nLook, horse! Viseria feet us on the child,\nVerily, and tell him heap of brief.\nFor I woo'll so it is.\n\nPOMPEY:\nBut can you not forgot your hends are pack'd?\n\nBIONDELLO:\nO, what a spenat lay be born ta'll fim,\nHave a burthen and a second Kate of France,\nAnd call"
 b"ROMEO:\nBut I do defend my lotgern any in the county.\n\nJULIET:\nI would the duke way dog! This vullain and all his\nfriends, nature bids thy pinish whom, by, boshing rewards.\nCome, my Lord Northumberland.\n\nPRINCE EDWARD:\nWatch this follows many more than this! O ho!\n\nNORFOLK:\nSpurs, praise I seal of trial. Hadst thou wilt amender:\nGood Poase and thy voice indeeds as year.\n\nKING RICHARD III:\nAnd do, my lord. Lord Angelo, have I pronound\nMy swinter saust and jegla.\n\nPRINCE:\nLet's fit how thy peace hath been bring me to my life.\nBut now I am a lover and that grace to her tears.\nAh, for like deep dead, and I then live.\nAlack, he'll prove you the king's side,\nI poison, think you, on what a fed afford a hearted faults\nAs I else hand of secret.\n\nLODSE:\nStopp'd in such a slanderous coward!\n\nDUCHESS OF YORK:\nGive me my boon, I'll rust, but that I did for the\nnoble priest let mall in Coriol whipping news,\nSwarn hurts in every purged further;\nLooks wink,--for shame, this lark doth among these wars,\nHis ac"], shape=(5,), dtype=string) 

________________________________________________________________________________

Run time: 2.452502489089966

Export the generator

This single-step model can easily be saved and restored, allowing you to use it anywhere a tf.saved_model is accepted.

tf.saved_model.save(one_step_model, 'one_step')
one_step_reloaded = tf.saved_model.load('one_step')
WARNING:tensorflow:Skipping full serialization of Keras layer <__main__.OneStep object at 0x7f2dec122490>, because it is not built.
2022-01-14 12:14:46.156693: W tensorflow/python/util/util.cc:368] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them.
WARNING:absl:Found untraced functions such as gru_cell_layer_call_fn, gru_cell_layer_call_and_return_conditional_losses while saving (showing 2 of 2). These functions will not be directly callable after loading.
INFO:tensorflow:Assets written to: one_step/assets
INFO:tensorflow:Assets written to: one_step/assets
states = None
next_char = tf.constant(['ROMEO:'])
result = [next_char]

for n in range(100):
  next_char, states = one_step_reloaded.generate_one_step(next_char, states=states)
  result.append(next_char)

print(tf.strings.join(result)[0].numpy().decode("utf-8"))
ROMEO:
Why dost thou dissemble.

BRUTUS:
But, O, prite, hath becall'd it;
And so, have patience:
'Tis like

Advanced: Customized Training

The above training procedure is simple, but does not give you much control. It uses teacher-forcing which prevents bad predictions from being fed back to the model, so the model never learns to recover from mistakes.

So now that you've seen how to run the model manually next you'll implement the training loop. This gives a starting point if, for example, you want to implement curriculum learning to help stabilize the model's open-loop output.

The most important part of a custom training loop is the train step function.

Use tf.GradientTape to track the gradients. You can learn more about this approach by reading the eager execution guide.

The basic procedure is:

  1. Execute the model and calculate the loss under a tf.GradientTape.
  2. Calculate the updates and apply them to the model using the optimizer.
class CustomTraining(MyModel):
  @tf.function
  def train_step(self, inputs):
      inputs, labels = inputs
      with tf.GradientTape() as tape:
          predictions = self(inputs, training=True)
          loss = self.loss(labels, predictions)
      grads = tape.gradient(loss, model.trainable_variables)
      self.optimizer.apply_gradients(zip(grads, model.trainable_variables))

      return {'loss': loss}

The above implementation of the train_step method follows Keras' train_step conventions. This is optional, but it allows you to change the behavior of the train step and still use keras' Model.compile and Model.fit methods.

model = CustomTraining(
    vocab_size=len(ids_from_chars.get_vocabulary()),
    embedding_dim=embedding_dim,
    rnn_units=rnn_units)
model.compile(optimizer = tf.keras.optimizers.Adam(),
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True))
model.fit(dataset, epochs=1)
172/172 [==============================] - 7s 24ms/step - loss: 2.6896
<keras.callbacks.History at 0x7f2d98313ed0>

Or if you need more control, you can write your own complete custom training loop:

EPOCHS = 10

mean = tf.metrics.Mean()

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

    mean.reset_states()
    for (batch_n, (inp, target)) in enumerate(dataset):
        logs = model.train_step([inp, target])
        mean.update_state(logs['loss'])

        if batch_n % 50 == 0:
            template = f"Epoch {epoch+1} Batch {batch_n} Loss {logs['loss']:.4f}"
            print(template)

    # saving (checkpoint) the model every 5 epochs
    if (epoch + 1) % 5 == 0:
        model.save_weights(checkpoint_prefix.format(epoch=epoch))

    print()
    print(f'Epoch {epoch+1} Loss: {mean.result().numpy():.4f}')
    print(f'Time taken for 1 epoch {time.time() - start:.2f} sec')
    print("_"*80)

model.save_weights(checkpoint_prefix.format(epoch=epoch))
Epoch 1 Batch 0 Loss 2.1657
Epoch 1 Batch 50 Loss 2.0466
Epoch 1 Batch 100 Loss 1.9395
Epoch 1 Batch 150 Loss 1.8729

Epoch 1 Loss: 1.9635
Time taken for 1 epoch 6.03 sec
________________________________________________________________________________
Epoch 2 Batch 0 Loss 1.7759
Epoch 2 Batch 50 Loss 1.7126
Epoch 2 Batch 100 Loss 1.6657
Epoch 2 Batch 150 Loss 1.6516

Epoch 2 Loss: 1.6907
Time taken for 1 epoch 5.30 sec
________________________________________________________________________________
Epoch 3 Batch 0 Loss 1.5436
Epoch 3 Batch 50 Loss 1.5582
Epoch 3 Batch 100 Loss 1.5763
Epoch 3 Batch 150 Loss 1.5273

Epoch 3 Loss: 1.5359
Time taken for 1 epoch 5.21 sec
________________________________________________________________________________
Epoch 4 Batch 0 Loss 1.4566
Epoch 4 Batch 50 Loss 1.4341
Epoch 4 Batch 100 Loss 1.4065
Epoch 4 Batch 150 Loss 1.4215

Epoch 4 Loss: 1.4411
Time taken for 1 epoch 5.17 sec
________________________________________________________________________________
Epoch 5 Batch 0 Loss 1.3676
Epoch 5 Batch 50 Loss 1.3668
Epoch 5 Batch 100 Loss 1.3656
Epoch 5 Batch 150 Loss 1.3548

Epoch 5 Loss: 1.3761
Time taken for 1 epoch 5.42 sec
________________________________________________________________________________
Epoch 6 Batch 0 Loss 1.2972
Epoch 6 Batch 50 Loss 1.3501
Epoch 6 Batch 100 Loss 1.3261
Epoch 6 Batch 150 Loss 1.3215

Epoch 6 Loss: 1.3257
Time taken for 1 epoch 5.31 sec
________________________________________________________________________________
Epoch 7 Batch 0 Loss 1.2677
Epoch 7 Batch 50 Loss 1.2358
Epoch 7 Batch 100 Loss 1.2528
Epoch 7 Batch 150 Loss 1.2749

Epoch 7 Loss: 1.2818
Time taken for 1 epoch 5.46 sec
________________________________________________________________________________
Epoch 8 Batch 0 Loss 1.2007
Epoch 8 Batch 50 Loss 1.2336
Epoch 8 Batch 100 Loss 1.2466
Epoch 8 Batch 150 Loss 1.2282

Epoch 8 Loss: 1.2404
Time taken for 1 epoch 5.28 sec
________________________________________________________________________________
Epoch 9 Batch 0 Loss 1.1911
Epoch 9 Batch 50 Loss 1.1818
Epoch 9 Batch 100 Loss 1.1954
Epoch 9 Batch 150 Loss 1.2156

Epoch 9 Loss: 1.2019
Time taken for 1 epoch 5.27 sec
________________________________________________________________________________
Epoch 10 Batch 0 Loss 1.1061
Epoch 10 Batch 50 Loss 1.1386
Epoch 10 Batch 100 Loss 1.1444
Epoch 10 Batch 150 Loss 1.1587

Epoch 10 Loss: 1.1627
Time taken for 1 epoch 5.58 sec
________________________________________________________________________________