ML Community Day is November 9! Join us for updates from TensorFlow, JAX, and more Learn more

Text generation with an RNN

View on 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
from tensorflow.keras.layers.experimental import preprocessing

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', '')
Downloading data from
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
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 preprocessing.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')
2021-08-11 18:24:53.295402: I tensorflow/stream_executor/cuda/] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-11 18:24:53.303654: I tensorflow/stream_executor/cuda/] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-11 18:24:53.304580: I tensorflow/stream_executor/cuda/] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-11 18:24:53.306209: I tensorflow/core/platform/] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 AVX512F FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2021-08-11 18:24:53.306828: I tensorflow/stream_executor/cuda/] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-11 18:24:53.307802: I tensorflow/stream_executor/cuda/] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-11 18:24:53.308798: I tensorflow/stream_executor/cuda/] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-11 18:24:53.896425: I tensorflow/stream_executor/cuda/] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-11 18:24:53.897329: I tensorflow/stream_executor/cuda/] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-11 18:24:53.898198: I tensorflow/stream_executor/cuda/] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-11 18:24:53.899171: I tensorflow/core/common_runtime/gpu/] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 14648 MB memory:  -> device: 0, name: Tesla V100-SXM2-16GB, pci bus id: 0000:00:05.0, compute capability: 7.0
<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 preprocessing.StringLookup layer:

ids_from_chars = preprocessing.StringLookup(
    vocabulary=list(vocab), mask_token=None)

It converts form 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 preprocessing.StringLookup(..., invert=True).

chars_from_ids = tf.keras.layers.experimental.preprocessing.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 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 =
for ids in ids_dataset.take(10):
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):
[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 =
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 '
2021-08-11 18:24:54.893532: I tensorflow/compiler/mlir/] None of the MLIR Optimization Passes are enabled (registered 2)

Create training batches

You used 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 shapes: ((64, 100), (64, 100)), types: (tf.int64, tf.int64)>

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):
    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(
    # Be sure the vocabulary size matches the `StringLookup` layers.

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)")
2021-08-11 18:24:57.345541: I tensorflow/stream_executor/cuda/] Loaded cuDNN version 8100
(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: 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:

array([41, 38,  9, 28,  6, 50, 20, 59, 44,  5, 51, 19, 40, 61, 13, 18, 32,
        0, 13,  0, 27, 37, 10, 46, 38, 40, 28, 22, 14, 44, 35, 22, 44, 16,
       17,  8, 55, 17, 39, 47, 47, 23,  3, 32, 30, 15, 10, 32,  8,  8,  3,
       47, 40, 38, 13,  5, 57, 12, 39,  5,  6, 14, 30, 12, 63, 51, 10, 14,
       52,  1, 47, 15, 48, 28, 38, 16, 22,  7, 59, 45, 44, 62, 23, 32, 36,
       40, 28, 65, 60,  7,  8,  0, 19, 28, 32, 62, 61, 20, 64,  6])

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'ous, and not valiant, you have shamed me\nIn your condemned seconds.\n\nCOMINIUS:\nIf I should tell thee'

Next Char Predictions:

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:         4.191435

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 =, epochs=EPOCHS, callbacks=[checkpoint_callback])
Epoch 1/20
172/172 [==============================] - 6s 23ms/step - loss: 2.7361
Epoch 2/20
172/172 [==============================] - 5s 23ms/step - loss: 2.0067
Epoch 3/20
172/172 [==============================] - 5s 23ms/step - loss: 1.7364
Epoch 4/20
172/172 [==============================] - 5s 23ms/step - loss: 1.5729
Epoch 5/20
172/172 [==============================] - 5s 23ms/step - loss: 1.4700
Epoch 6/20
172/172 [==============================] - 5s 23ms/step - loss: 1.4000
Epoch 7/20
172/172 [==============================] - 5s 23ms/step - loss: 1.3465
Epoch 8/20
172/172 [==============================] - 5s 23ms/step - loss: 1.3007
Epoch 9/20
172/172 [==============================] - 5s 23ms/step - loss: 1.2610
Epoch 10/20
172/172 [==============================] - 5s 23ms/step - loss: 1.2223
Epoch 11/20
172/172 [==============================] - 5s 23ms/step - loss: 1.1842
Epoch 12/20
172/172 [==============================] - 5s 23ms/step - loss: 1.1460
Epoch 13/20
172/172 [==============================] - 5s 23ms/step - loss: 1.1055
Epoch 14/20
172/172 [==============================] - 5s 23ms/step - loss: 1.0626
Epoch 15/20
172/172 [==============================] - 5s 24ms/step - loss: 1.0170
Epoch 16/20
172/172 [==============================] - 5s 23ms/step - loss: 0.9692
Epoch 17/20
172/172 [==============================] - 5s 23ms/step - loss: 0.9181
Epoch 18/20
172/172 [==============================] - 5s 23ms/step - loss: 0.8670
Epoch 19/20
172/172 [==============================] - 5s 23ms/step - loss: 0.8143
Epoch 20/20
172/172 [==============================] - 5s 23ms/step - loss: 0.7647

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)
It is a very example
Here done to Elcompash of her griefs, wherein Choise,
Without my enemy; you are o'er this scene
Thoughts that sown'd off to have a sufficient mon
hath made it on the people, break our case:
Who inciddst the hour, think you be gone?

For what I see, I doubt there was more periol to their friends?

Have you not hear? the senate pass down forth,
Countenance, prefermants, devised in courtezage,
Of it at punishes, and cry batter King Henry's use!

If they did I but last; I say to thir,
And fly: my vooking in those thing, it brings;
After an act, may stand in my foe instant?

So much upon the serving-creature.

Second Katharinan,
Save you this young father, news, will kiss
your honour to a covert fance to Farcius' blaze is expiled
till choose and call the foem of cheer himself.
Not so deliver, for this night shall be a cut-out
Yourselfs; as the flowers cannot no: what he pleg-son,
As the pay to her heavy, marches?



Run time: 2.3087921142578125

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:\nIt is my daughter, whom thou hast, no, no, what many which ho\ncaused for fear. Then?\n\nFirst Citizen:\nCousin of Buckingham, and therefore wast thou thin,\nBy Jove her thunder, not on him.\n\nFLORIZEL:\nMy lord,\nYou never spow him so perform her life;\nBut had thought the wanted counsel on the world,\nThe baid of old tale from him by foes,\nLike all forms, he doth not the duke well for herself.\nThe sons and fam is strucken murder;\nAnd bless he shall not be long.\nWhereto he better nothing, by the east,\nWas factionary against Exeter!\n\nHERMION:\nWhere is your pain? hings in a soldier.\n\nShepherd:\n'Tis south; I will not go by this; he loves' me\nThough noble Contro's shump.\n\nAEdile:\nHe's sudden; tood my friends are too sun\nPat on him an embastiest York by day, my liege,\nProfesses to follow Marcius.\n\nCOMINIUS:\nIt was come to us!\nBut, our queen, those weeping pay the formers any other;\nAnon even he should seem to dry.\n\nHESS OF YORK:\nMy lord, he both be so farther,\nBut 'tis as banish'd from the mind of "
 b"ROMEO:\nIt is spoke for triumphant garly, fis\nFresh out my daughter and the deed-joy\njeasons that I was lost innation and eyes from the\nthy glims.\n\nFROTH:\nHere comes this way, and sellow'd for and\nspeechange; cry 'D; inchance his down and with the or-house,\nWhere indeed the sedicing scholarging disdains\nDrows you.\n\nAlipan:\nWhere's Clifford; we will confess too,\nOr, by this song, nor pray now what I did\nHer uncle Rivers stands you to take away;\nBut in the like known thereof discresed at his\nheart wept humble as a pitch'd any right.\nWhereto I, 'Hill Henry, and you, my lord,\nKnow't again by Angelo, the head maid\nFalse to another scorns thus daring for\nAn angry ay angry. Veriling you\nThan which you are heart, gave war nor none within;\nTell he that first wretched to her dower, though it begin.\n\nDUKE VINCENTIO:\nWhere is Aufidius sister? how much factos loath\nto pride: King Richard in Bianco's singing.\n\nMARIANA:\nWhy art thou harst: for, to retire yourself\nTo County many thousand humble stains.\nSawnt"
 b"ROMEO:\nSatisfy!\nThink'st thou hast thou out of true applace: throw away\nThe rather for incapab-torment.\n\nGLoUCESTER:\nSo Gaunt in Eye wrong'd, belike.\n\nQUEEN:\n'Tis little friend, thou couldst know; mencle, Clifford.\nDid ut up the flesh; the sons and blubter\nTannot countervail the conquest of thyself.\nBut how must be a king, as hideous ass\nShould you go's assural trembling adjer!\nWhy shall deserve you but assuar their\ncoats of such persons to be your castle.\nCondemning soul to him and heir more than\nHer sups, moresely three women\none and a hongy: you have like his curediar,\nAnd chase him in the infirmine breachs.\n\nKING EDWARD IV:\nCansault thou son? She's a word.\n\nSICINIUS:\nThis shows assurance how the house of love\nLidst both our subjects as the senate's death;\nSoce thou consent to bitter, by the way to life\nBut my entity to give I agree:\nHield!\n\nBUCKINGHAM:\nMy lord, this last out with our complexions\nCherish rooted distapsups and call folls.\n\nLADY ANNE:\nWere he that wonders to us all the chan"
 b"ROMEO:\nI pray you, gentlemen.\n\nJULIET:\nMy lord, gath nothing in Padua for a\npiece of cut as a horseman I please;\nI'll follow what we speak again of love,\nIs broke an oath from false for me.\n\nGLOUCESTER:\nWell, jost ignorant of despite of my grief;\nAnd thus I pity three thou wast born.\n\nQUEEN ELIZABETH:\nWhy have you not done, Henry's coming smiles,\n'Tis like one inferious vengeance condemn'd\nBy Heavens and noblence foldying\nto her honour. what he comes long eate?\n\nHASTINGS:\nGo, get thee even to thus, that flies;\nI would adont the royally out of dist;\nAnd thus I turn and much since that make fair\nSun with such finger in quiet wnat, and Sariant\nShould have been either queen.\n\nISABELLA:\nPetruchio! Who is is the supper venge.\n\nSecond Murderer:\nO looken soul!\n\nA Forders, Earl of Clarence,--here is coming him.\n\nHORTENSIO:\nSay, when you saw you shall bectwary.\n\nCOMINIUS:\nYou have fought it the elder, the\nson: xishonour here the soretire passing slaves.\nAnd in his tidly I brought my good deed,\nAre nev"
 b"ROMEO:\nVillanted the blood reign purpose\nnot more and she would quench it. Should Such a\npentinus lipt from worth of charity.\nHow can we fing it, like a drum of me?\nSpeak, tending, O, how can I have seen your\nsaids, lest the hirs weeping earth, one shall\nIn such as you to bitter, but we east for King of\nThe pretties of his officer: yet your bey,\nThe curn'd deputy nexty. Tybalt, that's\nunfortunage, take this poor delivers to a friend,\nAnd grief hath kept in sign of knotking note.\nWelcome! Saint yet Murderer: to this scoldif cares\nThat I have not in my desire.\nNay, what will you such things prevent it, hands.\n\nKING RICHARD II:\nHow now, by thee!\n\nCLAUDIO:\nNo, good father.\n\nDUKE VINCENTIO:\nHow now, is gone to Raptatur, add, took fortune between\nmy life for time put forth parture most straitle queen's.\n\nHENRY BOLINGBROKE:\nUrge in any, unhappy by this news,\nWhilst thou lies She not remain, as if\nher fortune is not so rise report the queen?\n\nGLOUCESTER:\nStand up, Oncring me?\n\nLADYARAN:\n\nHERMIONE:\nN"], shape=(5,), dtype=string) 


Run time: 2.1990060806274414

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., 'one_step')
one_step_reloaded = tf.saved_model.load('one_step')
WARNING:tensorflow:Skipping full serialization of Keras layer <__main__.OneStep object at 0x7fdfad429d90>, because it is not built.
2021-08-11 18:26:53.785069: W tensorflow/python/util/] 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, gru_cell_layer_call_fn, gru_cell_layer_call_and_return_conditional_losses, gru_cell_layer_call_and_return_conditional_losses while saving (showing 5 of 5). 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)

Be a booqued banish'd: sly us or old
Yeed Margaret: and therefore follow'd there?


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 methods.

model = CustomTraining(
model.compile(optimizer = tf.keras.optimizers.Adam(),
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)), epochs=1)
172/172 [==============================] - 7s 23ms/step - loss: 2.7296
<keras.callbacks.History at 0x7fdfad7bf090>

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 2.1729
Epoch 1 Batch 50 Loss 2.0531
Epoch 1 Batch 100 Loss 1.9573
Epoch 1 Batch 150 Loss 1.8028

Epoch 1 Loss: 1.9959
Time taken for 1 epoch 5.83 sec
Epoch 2 Batch 0 Loss 1.8247
Epoch 2 Batch 50 Loss 1.7950
Epoch 2 Batch 100 Loss 1.7317
Epoch 2 Batch 150 Loss 1.6410

Epoch 2 Loss: 1.7202
Time taken for 1 epoch 5.28 sec
Epoch 3 Batch 0 Loss 1.6101
Epoch 3 Batch 50 Loss 1.5863
Epoch 3 Batch 100 Loss 1.5252
Epoch 3 Batch 150 Loss 1.5194

Epoch 3 Loss: 1.5582
Time taken for 1 epoch 5.23 sec
Epoch 4 Batch 0 Loss 1.4622
Epoch 4 Batch 50 Loss 1.4623
Epoch 4 Batch 100 Loss 1.4729
Epoch 4 Batch 150 Loss 1.4334

Epoch 4 Loss: 1.4580
Time taken for 1 epoch 5.30 sec
Epoch 5 Batch 0 Loss 1.4144
Epoch 5 Batch 50 Loss 1.4157
Epoch 5 Batch 100 Loss 1.3952
Epoch 5 Batch 150 Loss 1.3634

Epoch 5 Loss: 1.3902
Time taken for 1 epoch 5.48 sec
Epoch 6 Batch 0 Loss 1.3419
Epoch 6 Batch 50 Loss 1.3228
Epoch 6 Batch 100 Loss 1.3308
Epoch 6 Batch 150 Loss 1.3092

Epoch 6 Loss: 1.3365
Time taken for 1 epoch 5.22 sec
Epoch 7 Batch 0 Loss 1.3353
Epoch 7 Batch 50 Loss 1.2958
Epoch 7 Batch 100 Loss 1.2993
Epoch 7 Batch 150 Loss 1.3049

Epoch 7 Loss: 1.2915
Time taken for 1 epoch 5.33 sec
Epoch 8 Batch 0 Loss 1.2323
Epoch 8 Batch 50 Loss 1.2712
Epoch 8 Batch 100 Loss 1.2089
Epoch 8 Batch 150 Loss 1.2661

Epoch 8 Loss: 1.2513
Time taken for 1 epoch 5.21 sec
Epoch 9 Batch 0 Loss 1.2154
Epoch 9 Batch 50 Loss 1.2268
Epoch 9 Batch 100 Loss 1.2334
Epoch 9 Batch 150 Loss 1.2292

Epoch 9 Loss: 1.2124
Time taken for 1 epoch 5.24 sec
Epoch 10 Batch 0 Loss 1.1712
Epoch 10 Batch 50 Loss 1.1542
Epoch 10 Batch 100 Loss 1.1887
Epoch 10 Batch 150 Loss 1.2040

Epoch 10 Loss: 1.1734
Time taken for 1 epoch 5.56 sec