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":

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.

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?

The cause why then we are all resolved more sons.

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.

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

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.


Import TensorFlow and other libraries

import tensorflow as tf

import numpy as np
import os
import time
2023-11-16 12:28:52.207051: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
2023-11-16 12:28:52.207090: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
2023-11-16 12:28:52.208630: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered

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
1115394/1115394 [==============================] - 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
First Citizen:
Before we proceed any further, hear me speak.

Speak, speak.

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

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')
<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)
<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)
<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'))
<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):
seq_length = 100

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):
[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)

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

for seq in sequences.take(5):
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
(['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

# 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).

dataset = (
    .batch(BATCH_SIZE, drop_remainder=True)

<_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 StringLookup Layer
vocab_size = len(ids_from_chars.get_vocabulary())

# 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):
    self.embedding = tf.keras.layers.Embedding(vocab_size, embedding_dim)
    self.gru = tf.keras.layers.GRU(rnn_units,
    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
      return x
model = MyModel(

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: "my_model"
 Layer (type)                Output Shape              Param #   
 embedding (Embedding)       multiple                  16896     
 gru (GRU)                   multiple                  3938304   
 dense (Dense)               multiple                  67650     
Total params: 4022850 (15.35 MB)
Trainable params: 4022850 (15.35 MB)
Non-trainable params: 0 (0.00 Byte)

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:

array([15, 52, 19, 34,  6, 39, 41, 62, 50, 61, 42, 26, 29, 57, 34, 46, 12,
       61, 53, 14, 26, 50,  5,  8, 29, 44,  2, 65, 62, 52, 53, 26, 25, 39,
       64, 36, 53, 21, 34, 30, 12, 58, 61, 43, 38, 29,  1, 26, 47, 35, 52,
       30, 10, 20, 59,  9, 11, 34, 59, 45, 56, 20, 39, 29, 46, 10, 54, 56,
       57, 17, 19, 19, 14, 40, 12, 12,  4, 54, 22, 17, 31,  7, 61, 44, 56,
       36,  5, 38, 30, 32, 23, 21, 52, 39, 42, 30, 42,  8, 17, 53])

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

print("Input:\n", text_from_ids(input_example_batch[0]).numpy())
print("Next Char Predictions:\n", text_from_ids(sampled_indices).numpy())
 b" of woman in the world,\nAy, every dram of woman's flesh is false, If she be.\n\nLEONTES:\nHold your pea"

Next Char Predictions:
 b"BmFU'ZbwkvcMPrUg;vnAMk&-Pe zwmnMLZyWnHUQ;svdYP\nMhVmQ3Gt.:UtfqGZPg3oqrDFFAa;;$oIDR,veqW&YQSJHmZcQc-Dn"

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_mean_loss = loss(target_example_batch, example_batch_predictions)
print("Prediction shape: ", example_batch_predictions.shape, " # (batch_size, sequence_length, vocab_size)")
print("Mean loss:        ", example_batch_mean_loss)
Prediction shape:  (64, 100, 66)  # (batch_size, sequence_length, vocab_size)
Mean loss:         tf.Tensor(4.1884556, shape=(), dtype=float32)

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:


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(

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.

history = model.fit(dataset, epochs=EPOCHS, callbacks=[checkpoint_callback])
Epoch 1/20
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1700137742.036116   34050 device_compiler.h:186] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.
172/172 [==============================] - 12s 53ms/step - loss: 2.7219
Epoch 2/20
172/172 [==============================] - 10s 54ms/step - loss: 1.9972
Epoch 3/20
172/172 [==============================] - 11s 55ms/step - loss: 1.7187
Epoch 4/20
172/172 [==============================] - 11s 57ms/step - loss: 1.5568
Epoch 5/20
172/172 [==============================] - 11s 59ms/step - loss: 1.4579
Epoch 6/20
172/172 [==============================] - 11s 61ms/step - loss: 1.3891
Epoch 7/20
172/172 [==============================] - 12s 61ms/step - loss: 1.3356
Epoch 8/20
172/172 [==============================] - 12s 62ms/step - loss: 1.2913
Epoch 9/20
172/172 [==============================] - 11s 60ms/step - loss: 1.2501
Epoch 10/20
172/172 [==============================] - 11s 59ms/step - loss: 1.2090
Epoch 11/20
172/172 [==============================] - 11s 59ms/step - loss: 1.1693
Epoch 12/20
172/172 [==============================] - 11s 59ms/step - loss: 1.1283
Epoch 13/20
172/172 [==============================] - 11s 60ms/step - loss: 1.0859
Epoch 14/20
172/172 [==============================] - 11s 61ms/step - loss: 1.0391
Epoch 15/20
172/172 [==============================] - 11s 61ms/step - loss: 0.9928
Epoch 16/20
172/172 [==============================] - 11s 61ms/step - loss: 0.9408
Epoch 17/20
172/172 [==============================] - 11s 60ms/step - loss: 0.8888
Epoch 18/20
172/172 [==============================] - 11s 60ms/step - loss: 0.8361
Epoch 19/20
172/172 [==============================] - 11s 60ms/step - loss: 0.7840
Epoch 20/20
172/172 [==============================] - 11s 60ms/step - loss: 0.7337

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):
    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.
        # Match the shape to the vocabulary
    self.prediction_mask = tf.sparse.to_dense(sparse_mask)

  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,
    # 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 = tf.strings.join(result)
end = time.time()
print(result[0].numpy().decode('utf-8'), '\n\n' + '_'*80)
print('\nRun time:', end - start)
The dayning use your brodous parchemn a memaver
But to my shrow against it.

Do you think of't, leaving too?

It is, sir, let's see: to put my freedy-sorrow
In the resolution of the fell as of;
And she shall to the fish, whose parts-have
drey'd the process or owe we eptits the abtent,
That they have often been men asiel, as now
I lay stoly to none of your adversary title.
Nay, stay, what, nurse, shall I respect by him. Assisted with,
Hadst thou depart; we should have seen some name of meat:
Might from this coast was though his purpose, and they can ffor me to
And lack upon your royal king.

It is now pale; but not a man with winds,
And breathed sunshine way: in give of less, the nurse
In thy extreme budden and his wives.
One more, most noble friend.

Third Servingman:
I have no more of it.

O, she is found withal. Hark ye
and given me thence at we? or die among thes?
Evermother, Thomas, Duke old York and him,
So fast aspected: s 


Run time: 2.878746271133423

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 = tf.strings.join(result)
end = time.time()
print(result, '\n\n' + '_'*80)
print('\nRun time:', end - start)
[b"ROMEO:\nFirst, heaven leave us: O, rest thy wild,\nI will my father dead?\n\nTRANIO:\nBut did you send us run; lay, fool!\nI would the world say no; what brought to? hear?\n\nKATHARINA:\nFear you the heavess to take away the time\nof a kiss of sold murder, to keep this luck\nWhose uncle branches order, I cry you\ndo, and the moting father of the like days\nThey shall seem from reason, whilst I live or else\nTo raise his writing fallent to speak.\nBut, Clifford, he is gone unto these wents were faults.\n\nLUCENTIO:\nAh, Warwick, art thou hear, the worst of in my false\nMoth thy meaning brave through the seas.\nIn what occasion not thyself?\n\nMONTAGUE:\nGood queen; Antigons, and brave amazed at the\ngreaty will out-plane. Calm those that valiant quench,\nthough us't! it is your baseness. Come, sit down;\nFor, madam: son, away! Now, by the worst.\n\nAll:\nTeems as it hath mest.\n\nDORSET:\nAnd love as you; I could heed home.\nAlas! I needs must out, as thou art a\nfraitors hate than for the year, and the conquer'd bowes\nIn gro"
 b"ROMEO:\n\nJULIET:\nAs well return.\n\nShepherd:\nBut, good my lord,\nThe Duke of Clarence hath so\n't born: but all, within mine own absence\nI married my face, moved nor boy;\nAnd I from Deepany, advangal keeps and cheer\nOf heaven nor power, at some thread.\nThat English jedwats shall not stuck a dream;\nFairer that proud thoughts about the abonty of the crown.\n\nJULIET:\nThen, I may have now, she you seal the story\nWhere nowing an oath with lips.\n\nKING RICHARD II:\nDiscourse of any good conceit? Abas!\nMy gracious sovereitn, ladies that vaunts in\nhis wonder age; but yet we should hear\nMy words shed by, and thought on sue the word,\nFor, lords, to-marry my greatness: if my trotubous lies?\n\nLADY GREY:\nTo his submission. Lady and Darb.\n\nDUCHESS OF YORK:\nWouldst thou go? then I'll judge my tongue,\nAnd graced trembling of the widdows, or\nshe's the cause? a coucle's hand thither? O!\nBelance oph?\n\nESCALUS:\nAy; for thou wert kill'd\nwhen the wall's power I could's high a peal: be corn\nAt o'cropph the instrument of "
 b"ROMEO:\nOr else be swellest Edward: if this long-impellow:\nAll unacounted said Lecious honour have many sortons\nThat first we all go men.\n\nDUKE OF YORK:\nWelcome, my lord, widow!\n\nALONS:\nIf it be not; she was not made a queen,\n'This people will still have mine eyes to heaven from his wife.\n\nTHOMAS MOWBRAY:\nAy, ay.\n\nCitizens:\nDown, ladies,--which is my name is Edward.\nWhen yet she finds alone.\n\nBRAKENBURIO:\nBoth, young and older,, marry; nor my countenance, I did see\nForbidst unreasonably of our queen,--\n\nSTAN:\nI know, I thank thee, knew whose heads the rest,\nBut that the rodut of this speech,\nIs this of noble things you have.\n\nDUKE OF YORK:\nAy, so: you are hereafter, it may make you an\na-bow-forth. But O, pity!\nMake your affections are up, Signior Lucentio.\n\nLUCENTIO:\nIt may not pass:\nThen we shall be most well. but drawly pay\nMe all the world's shore: Alack the dire dream of your sword,\nOne that our kingdom to the head of monsminy.\n\nPRINCE EDWARD:\nAn is her fawling! here's your done,\nHe Canst"
 b"ROMEO:\nLet me see: there were heard.\n\nGLOUCESTER:\n\nKING EDWARD IV:\nAn beast my wife's will obey: how meaning in him, sin\nJest, and thou shalt smile at the fashion;\nFor inholence my most courtiesy, Warwick, steble,\nThou mayst not, sawn, Kate, and then Oxford's vast\nIs sworn to lay at honours on my head\nOf her travel the curses; but we shall bear it.\n\nROMEO:\nNot one would please, as I brother!\n\nCLAUDIO:\nPett up, I,\nWe cannot white of this.\n\nRIVERS:\nAy, for once whither that's no remembrance?\nThou ammonest not most partly keeps, new,\nShould we were cross-wounded to sadver a lord,\nThe horn and holy thank your soldiers arm,\nYou advice for his head to Margar to resign his cloace:\nSure thy wife indeed are womanish.\n\nLUCENTIO:\nThen, good my lord, what talk I possess;\nAnd for the county will he did bear\nOf free speech; and will assect thee,\nThat stands the thrawhilderful: whom I do, an either painted stain\nThe spirits of their power. If you requires a mettain,\nYour mess arrormmeth in either. But we m"
 b"ROMEO:\nWhy, so; farewell: one in, another death or dread\nWhen he were leons.\nSo first less thou the time 'twere my badied known,\nChequering to you.\n\nShepherd:\nGo heaven and false, she's highbell'd.\n\nGLOUCESTER:\nCome, go with you, be with him, if the condaction tears\nHe thither was post to revenge,\nAnd thither shall we could grace to the white:\nThou art to-more oraple! Rise of it!\n\nCOMINIUS:\nKeere yes; if we all shall fork\nAs every cousin, drawn.\n\nPRINCE EDWARD:\nLet me unkiss! my master is safe, Furils, take your power\nTo embraceous have at the goor humour and as yours,\nAnd pluck my babes against your tears, and there before the\ncut; or if she were hered betwixt his ease.'\n\nGLOUCESTER:\nHe reverend guilt return.\n\nPARIS:\nIn vain for your name\nIs not denied and drooping too.\n\nQUEEN MARGARET:\nWhat, thou? whence could not, how?\nSail note or two!\nWho stands the matter:--Now, payied and recensit, anst\nthen to fear every end on the matom, I\nam proud to meet your gate upon her mastray's.\n\nFLORIZEL:\nDo"], shape=(5,), dtype=string) 


Run time: 2.8853743076324463

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 0x7f1a9c2e6880>, because it is not built.
WARNING:tensorflow:Model's `__init__()` arguments contain non-serializable objects. Please implement a `get_config()` method in the subclassed Model for proper saving and loading. Defaulting to empty config.
WARNING:tensorflow:Model's `__init__()` arguments contain non-serializable objects. Please implement a `get_config()` method in the subclassed Model for proper saving and loading. Defaulting to empty config.
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)

While shall we toward them, gaunt and married. Urped me say I
doubt not, for this world is gentle,

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):
  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(
model.compile(optimizer = tf.keras.optimizers.Adam(),
model.fit(dataset, epochs=1)
172/172 [==============================] - 13s 58ms/step - loss: 2.7075
<keras.src.callbacks.History at 0x7f1a9c1b8190>

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


mean = tf.metrics.Mean()

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

    for (batch_n, (inp, target)) in enumerate(dataset):
        logs = model.train_step([inp, target])

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

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

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

Epoch 1 Batch 0 Loss 2.1913
Epoch 1 Batch 50 Loss 2.0591
Epoch 1 Batch 100 Loss 1.9363
Epoch 1 Batch 150 Loss 1.8937

Epoch 1 Loss: 1.9814
Time taken for 1 epoch 12.44 sec
Epoch 2 Batch 0 Loss 1.8354
Epoch 2 Batch 50 Loss 1.7515
Epoch 2 Batch 100 Loss 1.6990
Epoch 2 Batch 150 Loss 1.6749

Epoch 2 Loss: 1.7090
Time taken for 1 epoch 11.45 sec
Epoch 3 Batch 0 Loss 1.5859
Epoch 3 Batch 50 Loss 1.5785
Epoch 3 Batch 100 Loss 1.5548
Epoch 3 Batch 150 Loss 1.5089

Epoch 3 Loss: 1.5524
Time taken for 1 epoch 11.49 sec
Epoch 4 Batch 0 Loss 1.4963
Epoch 4 Batch 50 Loss 1.4674
Epoch 4 Batch 100 Loss 1.4629
Epoch 4 Batch 150 Loss 1.4254

Epoch 4 Loss: 1.4550
Time taken for 1 epoch 11.26 sec
Epoch 5 Batch 0 Loss 1.3884
Epoch 5 Batch 50 Loss 1.4480
Epoch 5 Batch 100 Loss 1.3669
Epoch 5 Batch 150 Loss 1.3619

Epoch 5 Loss: 1.3870
Time taken for 1 epoch 11.19 sec
Epoch 6 Batch 0 Loss 1.3157
Epoch 6 Batch 50 Loss 1.3346
Epoch 6 Batch 100 Loss 1.3065
Epoch 6 Batch 150 Loss 1.2660

Epoch 6 Loss: 1.3341
Time taken for 1 epoch 11.25 sec
Epoch 7 Batch 0 Loss 1.3223
Epoch 7 Batch 50 Loss 1.2794
Epoch 7 Batch 100 Loss 1.2886
Epoch 7 Batch 150 Loss 1.3036

Epoch 7 Loss: 1.2888
Time taken for 1 epoch 11.10 sec
Epoch 8 Batch 0 Loss 1.2318
Epoch 8 Batch 50 Loss 1.2245
Epoch 8 Batch 100 Loss 1.2677
Epoch 8 Batch 150 Loss 1.2397

Epoch 8 Loss: 1.2480
Time taken for 1 epoch 11.13 sec
Epoch 9 Batch 0 Loss 1.2021
Epoch 9 Batch 50 Loss 1.2654
Epoch 9 Batch 100 Loss 1.2190
Epoch 9 Batch 150 Loss 1.1929

Epoch 9 Loss: 1.2083
Time taken for 1 epoch 11.31 sec
Epoch 10 Batch 0 Loss 1.1429
Epoch 10 Batch 50 Loss 1.1642
Epoch 10 Batch 100 Loss 1.1455
Epoch 10 Batch 150 Loss 1.1687

Epoch 10 Loss: 1.1684
Time taken for 1 epoch 11.55 sec