Keras 3 API documentation / KerasNLP / Models / Roberta / RobertaMaskedLMPreprocessor layer

RobertaMaskedLMPreprocessor layer

[source]

RobertaMaskedLMPreprocessor class

keras_nlp.models.RobertaMaskedLMPreprocessor(
    tokenizer,
    sequence_length=512,
    truncate="round_robin",
    mask_selection_rate=0.15,
    mask_selection_length=96,
    mask_token_rate=0.8,
    random_token_rate=0.1,
    **kwargs
)

RoBERTa preprocessing for the masked language modeling task.

This preprocessing layer will prepare inputs for a masked language modeling task. It is primarily intended for use with the keras_nlp.models.RobertaMaskedLM task model. Preprocessing will occur in multiple steps.

  1. Tokenize any number of input segments using the tokenizer.
  2. Pack the inputs together with the appropriate "<s>", "</s>" and "<pad>" tokens, i.e., adding a single "<s>" at the start of the entire sequence, "</s></s>" between each segment, and a "</s>" at the end of the entire sequence.
  3. Randomly select non-special tokens to mask, controlled by mask_selection_rate.
  4. Construct a (x, y, sample_weight) tuple suitable for training with a keras_nlp.models.RobertaMaskedLM task model.

Arguments

  • tokenizer: A keras_nlp.models.RobertaTokenizer instance.
  • sequence_length: int. The length of the packed inputs.
  • truncate: string. The algorithm to truncate a list of batched segments to fit within sequence_length. The value can be either round_robin or waterfall: - "round_robin": Available space is assigned one token at a time in a round-robin fashion to the inputs that still need some, until the limit is reached. - "waterfall": The allocation of the budget is done using a "waterfall" algorithm that allocates quota in a left-to-right manner and fills up the buckets until we run out of budget. It supports an arbitrary number of segments.
  • mask_selection_rate: float. The probability an input token will be dynamically masked.
  • mask_selection_length: int. The maximum number of masked tokens in a given sample.
  • mask_token_rate: float. The probability the a selected token will be replaced with the mask token.
  • random_token_rate: float. The probability the a selected token will be replaced with a random token from the vocabulary. A selected token will be left as is with probability 1 - mask_token_rate - random_token_rate.

Call arguments

  • x: A tensor of single string sequences, or a tuple of multiple tensor sequences to be packed together. Inputs may be batched or unbatched. For single sequences, raw python inputs will be converted to tensors. For multiple sequences, pass tensors directly.
  • y: Label data. Should always be None as the layer generates labels.
  • sample_weight: Label weights. Should always be None as the layer generates label weights.

Examples

Directly calling the layer on data.

# Load the preprocessor from a preset.
preprocessor = keras_nlp.models.RobertaMaskedLMPreprocessor.from_preset(
    "roberta_base_en"
)

# Tokenize and mask a single sentence.
preprocessor("The quick brown fox jumped.")

# Tokenize and mask a batch of single sentences.
preprocessor(["The quick brown fox jumped.", "Call me Ishmael."])

# Tokenize and mask sentence pairs.
# In this case, always convert input to tensors before calling the layer.
first = tf.constant(["The quick brown fox jumped.", "Call me Ishmael."])
second = tf.constant(["The fox tripped.", "Oh look, a whale."])
preprocessor((first, second))

Mapping with tf.data.Dataset.

preprocessor = keras_nlp.models.RobertaMaskedLMPreprocessor.from_preset(
    "roberta_base_en"
)

first = tf.constant(["The quick brown fox jumped.", "Call me Ishmael."])
second = tf.constant(["The fox tripped.", "Oh look, a whale."])

# Map single sentences.
ds = tf.data.Dataset.from_tensor_slices(first)
ds = ds.map(preprocessor, num_parallel_calls=tf.data.AUTOTUNE)

# Map sentence pairs.
ds = tf.data.Dataset.from_tensor_slices((first, second))
# Watch out for tf.data's default unpacking of tuples here!
# Best to invoke the `preprocessor` directly in this case.
ds = ds.map(
    lambda first, second: preprocessor(x=(first, second)),
    num_parallel_calls=tf.data.AUTOTUNE,
)

[source]

from_preset method

RobertaMaskedLMPreprocessor.from_preset()

Instantiate RobertaMaskedLMPreprocessor from preset architecture.

Arguments

  • preset: string. Must be one of "roberta_base_en", "roberta_large_en".

Examples

# Load a preprocessor layer from a preset.
preprocessor = keras_nlp.models.RobertaMaskedLMPreprocessor.from_preset(
    "roberta_base_en",
)
Preset name Parameters Description
roberta_base_en 124.05M 12-layer RoBERTa model where case is maintained.Trained on English Wikipedia, BooksCorpus, CommonCraw, and OpenWebText.
roberta_large_en 354.31M 24-layer RoBERTa model where case is maintained.Trained on English Wikipedia, BooksCorpus, CommonCraw, and OpenWebText.

tokenizer property

keras_nlp.models.RobertaMaskedLMPreprocessor.tokenizer

The tokenizer used to tokenize strings.