In this tutorial we are going to learn to generate text using a character-based RNN. We will use a toy dataset, which consists dinosaur names, to generate new and unique dinosaur names. This toy example may not have many practical applications, but it can be used to learn fundamentals, which can then be applied to a more challenging real world problem. So, let us get started by reviewing the goals and outline of this tutorial:

After this tutorial you will understand:

  • Preparing dataset for the RNN
  • Building a RNN in TensorFlow
  • Generating text using the trained RNN
  • Customizing training loop in TensorFlow

This tutorial is Jupyter notebook styled, so that you can follow the code or even run it in a browser. You can find all the necessary things needed to learn from this tutorial here.

# Let us import some libraries

import tensorflow as tf
import os
import numpy as np
import matplotlib.pyplot as plt
import time

Let us get started by reading the data file.

Exploring the dataset

# Read in the text
text = open('dinos.txt', 'rb').read().decode(encoding='utf-8')
print ('Length of text: {} characters'.format(len(text)))


    Length of text: 19913 characters

Let us see what the text looks like, by printing first 100 characters of the text.




Let us now see the unique characters in the file.

vocab = sorted(set(text))
print ('{} unique characters'.format(len(vocab)))


    53 unique characters

Let us plot the word length distribution in the text.

words = text.split('\n')
lengths = []
for word in words:

plt.title('Word length distribution')
plt.xlabel('Word Length')


Word length distribution

We see that most of the words are of length 12 or 13. Let us find the length of the longest name.

max_length = max(lengths)
print(f'max length: {max_length}')


    max length: 23

Process the text

As the first preprocessing step we will pad the words to make them of the same length. We choose the new line character for the same.

padding_char = '\n'    # declaring the padding character
text = ''              # stores the processed text

for word in words:
    # For each word, add it to the text, and add appropriate padding
    text = text + word + padding_char * (1 + max_length - len(word))




Before training, we need to map strings to a numerical representation. Create two lookup tables: one mapping characters to numbers, and another for numbers to characters.

# Creating a mapping from unique characters to indices
char2idx = {u:i for i, u in enumerate(vocab)}
idx2char = np.array(vocab)

text_as_int = np.array([char2idx[c] for c in text])

Now we have an integer representation for each character. Notice that we mapped the character as indexes from 0 to len(unique).

for char,_ in zip(char2idx, range(20)):
    print('  {:4s}: {:3d},'.format(repr(char), char2idx[char]))
print('  ...\n}')


      '\n':   0,
      'A' :   1,
      'B' :   2,
      'C' :   3,
      'D' :   4,
      'E' :   5,
      'F' :   6,
      'G' :   7,
      'H' :   8,
      'I' :   9,
      'J' :  10,
      'K' :  11,
      'L' :  12,
      'M' :  13,
      'N' :  14,
      'O' :  15,
      'P' :  16,
      'Q' :  17,
      'R' :  18,
      'S' :  19,

Let us see first 100 characters of text as sequence of numbers.



    array([ 1, 27, 29, 34, 31, 40, 41, 45, 27, 47, 44, 47, 45,  0,  0,  0,  0,
            0,  0,  0,  0,  0,  0,  0,  1, 27, 44, 30, 41, 40, 51, 50,  0,  0,
            0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  1, 28, 30,
           27, 38, 38, 27, 34, 45, 27, 47, 44, 47, 45,  0,  0,  0,  0,  0,  0,
            0,  0,  0,  0,  1, 28, 31, 38, 35, 45, 27, 47, 44, 47, 45,  0,  0,
            0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  1, 28, 44, 35])
# Show how the first 13 characters from the text are mapped to integers
print ('{} ---- characters mapped to int ---- > {}'.format(repr(text[:13]), text_as_int[:13]))


    'Aachenosaurus' ---- characters mapped to int ---- > [ 1 27 29 34 31 40 41 45 27 47 44 47 45]

The prediction task

Given a character, or a sequence of characters, what is the most probable next character? This is the task we’re training the model to perform. The input to the model will be a sequence of characters, and we 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.

# The maximum length sentence we want for a single input in characters
seq_length = max_length + 1
examples_per_epoch = len(text)//(seq_length+1)

# Create training examples / targets
char_dataset =

for i in char_dataset.take(5):



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

sequences = char_dataset.batch(seq_length, drop_remainder=True)

for item in sequences.take(5):



For each sequence, duplicate and shift it to form the input and target text by using the map method to apply a simple function to each batch:

def split_input_target(chunk):
    input_text = chunk[:-1]
    target_text = chunk[1:]
    return input_text, target_text

dataset =

Print the first examples input and target values:

for input_example, target_example in  dataset.take(1):
    print ('Input data: ', repr(''.join(idx2char[input_example.numpy()])), ':', len(input_example.numpy()))
    print ('Target data:', repr(''.join(idx2char[target_example.numpy()])), ':', len(target_example.numpy()))


    Input data:  'Aachenosaurus\n\n\n\n\n\n\n\n\n\n' : 23
    Target data: 'achenosaurus\n\n\n\n\n\n\n\n\n\n\n' : 23

Each index of these vectors are processed as one time step. For the input at time step 0, the model receives the index for “F” and trys to predict the index for “i” as the next character. At the next timestep, it does the same thing but the RNN considers the previous step context in addition to the current input character.

for i, (input_idx, target_idx) in enumerate(zip(input_example[:5], target_example[:5])):
    print("Step {:4d}".format(i))
    print("  input: {} ({:s})".format(input_idx, repr(idx2char[input_idx])))
    print("  expected output: {} ({:s})".format(target_idx, repr(idx2char[target_idx])))


    Step    0
      input: 1 ('A')
      expected output: 27 ('a')
    Step    1
      input: 27 ('a')
      expected output: 29 ('c')
    Step    2
      input: 29 ('c')
      expected output: 34 ('h')
    Step    3
      input: 34 ('h')
      expected output: 31 ('e')
    Step    4
      input: 31 ('e')
      expected output: 40 ('n')

Create training batches

We used to split the text into manageable sequences. But before feeding this data into the model, we 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 = dataset.shuffle(BUFFER_SIZE).batch(BATCH_SIZE, drop_remainder=True)



    <BatchDataset shapes: ((64, 23), (64, 23)), types: (tf.int64, tf.int64)>

Build the model

Use tf.keras.Sequential to define the model. For this simple example three layers are used to define our model:

Use tf.keras.Sequential to define the model. For this simple example three layers are used to define our model:

# Length of the vocabulary in chars
vocab_size = len(vocab)

# The embedding dimension
embedding_dim = 256

# Number of RNN units
rnn_units = 1024
def build_model(vocab_size, embedding_dim, rnn_units, batch_size):
    model = tf.keras.Sequential([
        tf.keras.layers.Embedding(vocab_size, embedding_dim,
                              batch_input_shape=[batch_size, None]),
    return model
model = build_model(
  vocab_size = len(vocab),

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 [Credits - Tensorflow]

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, 23, 53) # (batch_size, sequence_length, vocab_size)

In the above example the sequence length of the input is 22 but the model can be run on inputs of any length:



    Model: "sequential"
    Layer (type)                 Output Shape              Param #   
    embedding (Embedding)        (64, None, 256)           13568     
    gru (GRU)                    (64, None, 1024)          3938304   
    dense (Dense)                (64, None, 53)            54325     
    Total params: 4,006,197
    Trainable params: 4,006,197
    Non-trainable params: 0

To get actual predictions from the model we 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([13,  7, 12, 40, 33,  6, 49,  8, 26, 37, 46, 35, 25,  9, 51, 47, 25,
           21, 27, 31,  0,  6, 35])

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

print("Input: \n", repr("".join(idx2char[input_example_batch[0].numpy()])))
print("Next Char Predictions: \n", repr("".join(idx2char[sampled_indices])))


    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 our model returns logits, we need to set the from_logits flag.

def loss(labels, logits):
    return tf.keras.losses.sparse_categorical_crossentropy(labels, logits, from_logits=True)

example_batch_loss  = loss(target_example_batch, example_batch_predictions)
print("Prediction shape: ", example_batch_predictions.shape, " # (batch_size, sequence_length, vocab_size)")
print("scalar_loss:      ", example_batch_loss.numpy().mean())


    Prediction shape:  (64, 23, 53)  # (batch_size, sequence_length, vocab_size)
    scalar_loss:       3.9535964

Configure the training procedure using the tf.keras.Model.compile method. We’ll 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}")


Execute the training

To keep training time reasonable, use 20 epochs to train the model.

history =, epochs=EPOCHS, callbacks=[checkpoint_callback])


    Epoch 1/20
    24/24 [==============================] - 15s 625ms/step - loss: 3.0601
    Epoch 20/20
    24/24 [==============================] - 21s 868ms/step - loss: 0.6993

Generate text

Restore the latest checkpoint

To keep this prediction step simple, use a batch size of 1.

Because of the way the RNN state is passed from timestep to timestep, the model only accepts a fixed batch size once built.

To run the model with a different batch_size, we need to rebuild the model and restore the weights from the checkpoint.



model = build_model(vocab_size, embedding_dim, rnn_units, batch_size=1)

model.load_weights(tf.train.latest_checkpoint(checkpoint_dir))[1, None]))


    Model: "sequential_1"
    Layer (type)                 Output Shape              Param #   
    embedding_1 (Embedding)      (1, None, 256)            13568     
    gru_1 (GRU)                  (1, None, 1024)           3938304   
    dense_1 (Dense)              (1, None, 53)             54325     
    Total params: 4,006,197
    Trainable params: 4,006,197
    Non-trainable params: 0

The prediction loop

The following code block generates the text:

  • It Starts by choosing a start string, initializing the RNN state and setting the number of characters to generate.
  • Get the prediction distribution of the next character using the start string and the RNN state.
  • Then, use a categorical distribution to calculate the index of the predicted character. Use this predicted character as our next input to the model.
  • The RNN state returned by the model is fed back into the model so that it now has more context, instead than only one character. After predicting the next character, the modified RNN states are again fed back into the model, which is how it learns as it gets more context from the previously predicted characters.

To generate text the model's output is fed back to the input [Credit - TensorFlow]

Looking at the generated text, you’ll see the model knows when to capitalize and when to end the generated text. With the small number of epochs it generates believable dinosaur names.

def generate_text(model, start_string):
    # Evaluation step (generating text using the learned model)
    # Number of characters to generate
    num_generate = max_length
    # Converting our start string to numbers (vectorizing)
    input_eval = [char2idx[s] for s in start_string]
    input_eval = tf.expand_dims(input_eval, 0)
    # Empty string to store our results
    text_generated = []
    # Low temperatures results in more predictable text.
    # Higher temperatures results in more surprising text.
    # Experiment to find the best setting.
    temperature = 0.2
    # Here batch size == 1
    for i in range(num_generate):
        predictions = model(input_eval)
        # remove the batch dimension
        predictions = tf.squeeze(predictions, 0)
        # using a categorical distribution to predict the character returned by the model
        predictions = predictions / temperature
        predicted_id = tf.random.categorical(predictions, num_samples=1)[-1,0].numpy()
        # We pass the predicted character as the next input to the model
        # along with the previous hidden state
        input_eval = tf.expand_dims([predicted_id], 0)
    return (start_string + ''.join(text_generated))

Let us generate a new dinosaur name, that starts with Ram.

generated_text = generate_text(model, start_string=u"Ram")
print(generated_text.strip()) # strip new characters and print



Try to generate dinosaur names yourself and while doing so play around with the temperature parameter above. The easiest thing you can do to improve the results it to train it for longer (try EPOCHS=30).

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

Customized Training

The above training procedure is simple, but does not give you much control.

So now that you’ve seen how to run the model manually let’s unpack the training loop, and implement it ourselves. This gives a starting point if, for example, to implement curriculum learning to help stabilize the model’s open-loop output.

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

The procedure works as follows:

  • First, initialize the RNN state. We do this by calling the tf.keras.Model.reset_states method.
  • Next, iterate over the dataset (batch by batch) and calculate the predictions associated with each.
  • Open a tf.GradientTape, and calculate the predictions and loss in that context.
  • Calculate the gradients of the loss with respect to the model variables using the tf.GradientTape.grads method.
  • Finally, take a step downwards by using the optimizer’s tf.train.Optimizer.apply_gradients method.
model = build_model(
  vocab_size = len(vocab),
optimizer = tf.keras.optimizers.Adam()
def train_step(inp, target):
    with tf.GradientTape() as tape:
        predictions = model(inp)
        loss = tf.reduce_mean(
                target, predictions, from_logits=True))
    grads = tape.gradient(loss, model.trainable_variables)
    optimizer.apply_gradients(zip(grads, model.trainable_variables))
    return loss
# Training step

for epoch in range(EPOCHS):
    start = time.time()
    # initializing the hidden state at the start of every epoch
    # initally hidden is None
    hidden = model.reset_states()
    for (batch_n, (inp, target)) in enumerate(dataset):
        loss = train_step(inp, target)
        if batch_n % 100 == 0:
            template = 'Epoch {} Batch {} Loss {}'
            print(template.format(epoch+1, batch_n, loss))
    # saving (checkpoint) the model every 5 epochs
    if (epoch + 1) % 5 == 0:
    print ('Epoch {} Loss {:.4f}'.format(epoch+1, loss))
    print ('Time taken for 1 epoch {} sec\n'.format(time.time() - start))



    Epoch 1 Batch 0 Loss 3.9669971466064453
    Epoch 1 Loss 1.5980
    Time taken for 1 epoch 21.90713667869568 sec
    Epoch 10 Batch 0 Loss 0.9955548644065857
    Epoch 10 Loss 0.8829
    Time taken for 1 epoch 15.176138639450073 sec