keras-hub-nightly 0.15.0.dev20240823171555__py3-none-any.whl

keras_hub/src/models/albert/albert_masked_lm_preprocessor.py

@@ -0,0 +1,194 @@
+# Copyright 2024 The KerasHub Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import keras
+from absl import logging
+
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.layers.preprocessing.masked_lm_mask_generator import (
+    MaskedLMMaskGenerator,
+)
+from keras_hub.src.models.albert.albert_preprocessor import AlbertPreprocessor
+
+
+@keras_hub_export("keras_hub.models.AlbertMaskedLMPreprocessor")
+class AlbertMaskedLMPreprocessor(AlbertPreprocessor):
+    """ALBERT 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_hub.models.AlbertMaskedLM` task model. Preprocessing will occur in
+    multiple steps.
+
+    - Tokenize any number of input segments using the `tokenizer`.
+    - 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.
+    - Randomly select non-special tokens to mask, controlled by
+      `mask_selection_rate`.
+    - Construct a `(x, y, sample_weight)` tuple suitable for training with a
+      `keras_hub.models.AlbertMaskedLM` task model.
+
+    Args:
+        tokenizer: A `keras_hub.models.AlbertTokenizer` instance.
+        sequence_length: The length of the packed inputs.
+        mask_selection_rate: The probability an input token will be dynamically
+            masked.
+        mask_selection_length: The maximum number of masked tokens supported
+            by the layer.
+        mask_token_rate: float. `mask_token_rate` must be
+            between 0 and 1 which indicates how often the mask_token is
+            substituted for tokens selected for masking.
+            Defaults to `0.8`.
+        random_token_rate: float. `random_token_rate` must be
+            between 0 and 1 which indicates how often a random token is
+            substituted for tokens selected for masking. Default is 0.1.
+            Note: mask_token_rate + random_token_rate <= 1,  and for
+            (1 - mask_token_rate - random_token_rate), the token will not be
+            changed. Defaults to `0.1`.
+        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.
+
+    Examples:
+
+    Directly calling the layer on data.
+    ```python
+    preprocessor = keras_hub.models.AlbertMaskedLMPreprocessor.from_preset(
+        "albert_base_en_uncased"
+    )
+
+    # 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`.
+    ```python
+    preprocessor = keras_hub.models.AlbertMaskedLMPreprocessor.from_preset(
+        "albert_base_en_uncased"
+    )
+
+    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,
+    )
+    ```
+    """
+
+    def __init__(
+        self,
+        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,
+    ):
+        super().__init__(
+            tokenizer,
+            sequence_length=sequence_length,
+            truncate=truncate,
+            **kwargs,
+        )
+        self.mask_selection_rate = mask_selection_rate
+        self.mask_selection_length = mask_selection_length
+        self.mask_token_rate = mask_token_rate
+        self.random_token_rate = random_token_rate
+        self.masker = None
+
+    def build(self, input_shape):
+        super().build(input_shape)
+        # Defer masker creation to `build()` so that we can be sure tokenizer
+        # assets have loaded when restoring a saved model.
+        self.masker = MaskedLMMaskGenerator(
+            mask_selection_rate=self.mask_selection_rate,
+            mask_selection_length=self.mask_selection_length,
+            mask_token_rate=self.mask_token_rate,
+            random_token_rate=self.random_token_rate,
+            vocabulary_size=self.tokenizer.vocabulary_size(),
+            mask_token_id=self.tokenizer.mask_token_id,
+            unselectable_token_ids=[
+                self.tokenizer.cls_token_id,
+                self.tokenizer.sep_token_id,
+                self.tokenizer.pad_token_id,
+            ],
+        )
+
+    def get_config(self):
+        config = super().get_config()
+        config.update(
+            {
+                "mask_selection_rate": self.mask_selection_rate,
+                "mask_selection_length": self.mask_selection_length,
+                "mask_token_rate": self.mask_token_rate,
+                "random_token_rate": self.random_token_rate,
+            }
+        )
+        return config
+
+    def call(self, x, y=None, sample_weight=None):
+        if y is not None or sample_weight is not None:
+            logging.warning(
+                f"{self.__class__.__name__} generates `y` and `sample_weight` "
+                "based on your input data, but your data already contains `y` "
+                "or `sample_weight`. Your `y` and `sample_weight` will be "
+                "ignored."
+            )
+
+        x = super().call(x)
+        token_ids, segment_ids, padding_mask = (
+            x["token_ids"],
+            x["segment_ids"],
+            x["padding_mask"],
+        )
+        masker_outputs = self.masker(token_ids)
+        x = {
+            "token_ids": masker_outputs["token_ids"],
+            "segment_ids": segment_ids,
+            "padding_mask": padding_mask,
+            "mask_positions": masker_outputs["mask_positions"],
+        }
+        y = masker_outputs["mask_ids"]
+        sample_weight = masker_outputs["mask_weights"]
+        return keras.utils.pack_x_y_sample_weight(x, y, sample_weight)

keras_hub/src/models/albert/albert_preprocessor.py

@@ -0,0 +1,206 @@
+# Copyright 2024 The KerasHub Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import keras
+
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.layers.preprocessing.multi_segment_packer import (
+    MultiSegmentPacker,
+)
+from keras_hub.src.models.albert.albert_tokenizer import AlbertTokenizer
+from keras_hub.src.models.preprocessor import Preprocessor
+from keras_hub.src.utils.keras_utils import (
+    convert_inputs_to_list_of_tensor_segments,
+)
+
+
+@keras_hub_export("keras_hub.models.AlbertPreprocessor")
+class AlbertPreprocessor(Preprocessor):
+    """An ALBERT preprocessing layer which tokenizes and packs inputs.
+
+    This preprocessing layer will do three things:
+
+     - Tokenize any number of input segments using the `tokenizer`.
+     - Pack the inputs together using a `keras_hub.layers.MultiSegmentPacker`.
+       with the appropriate `"[CLS]"`, `"[SEP]"` and `"<pad>"` tokens.
+     - Construct a dictionary with keys `"token_ids"`, `"segment_ids"` and
+       `"padding_mask"`, that can be passed directly to
+       `keras_hub.models.AlbertBackbone`.
+
+    This layer can be used directly with `tf.data.Dataset.map` to preprocess
+    string data in the `(x, y, sample_weight)` format used by
+    `keras.Model.fit`.
+
+    The call method of this layer accepts three arguments, `x`, `y`, and
+    `sample_weight`. `x` can be a python string or tensor representing a single
+    segment, a list of python strings representing a batch of single segments,
+    or a list of tensors representing multiple segments to be packed together.
+    `y` and `sample_weight` are both optional, can have any format, and will be
+    passed through unaltered.
+
+    Special care should be taken when using `tf.data` to map over an unlabeled
+    tuple of string segments. `tf.data.Dataset.map` will unpack this tuple
+    directly into the call arguments of this layer, rather than forward all
+    argument to `x`. To handle this case, it is recommended to  explicitly call
+    the layer, e.g. `ds.map(lambda seg1, seg2: preprocessor(x=(seg1, seg2)))`.
+
+    Args:
+        tokenizer: A `keras_hub.models.AlbertTokenizer` instance.
+        sequence_length: 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.
+
+    Examples:
+    Directly calling the layer on data.
+    ```python
+    preprocessor = keras_hub.models.AlbertPreprocessor.from_preset(
+        "albert_base_en_uncased"
+    )
+
+    # Tokenize and pack a single sentence.
+    preprocessor("The quick brown fox jumped.")
+
+    # Tokenize a batch of single sentences.
+    preprocessor(["The quick brown fox jumped.", "Call me Ishmael."])
+
+    # Preprocess a batch of sentence pairs.
+    # When handling multiple sequences, always convert to tensors first!
+    first = tf.constant(["The quick brown fox jumped.", "Call me Ishmael."])
+    second = tf.constant(["The fox tripped.", "Oh look, a whale."])
+    preprocessor((first, second))
+
+    # Custom vocabulary.
+    bytes_io = io.BytesIO()
+    ds = tf.data.Dataset.from_tensor_slices(["The quick brown fox jumped."])
+    sentencepiece.SentencePieceTrainer.train(
+        sentence_iterator=ds.as_numpy_iterator(),
+        model_writer=bytes_io,
+        vocab_size=10,
+        model_type="WORD",
+        pad_id=0,
+        unk_id=1,
+        bos_id=2,
+        eos_id=3,
+        pad_piece="<pad>",
+        unk_piece="<unk>",
+        bos_piece="[CLS]",
+        eos_piece="[SEP]",
+        user_defined_symbols="[MASK]",
+    )
+    tokenizer = keras_hub.models.AlbertTokenizer(
+        proto=bytes_io.getvalue(),
+    )
+    preprocessor = keras_hub.models.AlbertPreprocessor(tokenizer)
+    preprocessor("The quick brown fox jumped.")
+    ```
+
+    Mapping with `tf.data.Dataset`.
+    ```python
+    preprocessor = keras_hub.models.AlbertPreprocessor.from_preset(
+        "albert_base_en_uncased"
+    )
+
+    first = tf.constant(["The quick brown fox jumped.", "Call me Ishmael."])
+    second = tf.constant(["The fox tripped.", "Oh look, a whale."])
+    label = tf.constant([1, 1])
+
+    # Map labeled single sentences.
+    ds = tf.data.Dataset.from_tensor_slices((first, label))
+    ds = ds.map(preprocessor, num_parallel_calls=tf.data.AUTOTUNE)
+
+    # Map unlabeled single sentences.
+    ds = tf.data.Dataset.from_tensor_slices(first)
+    ds = ds.map(preprocessor, num_parallel_calls=tf.data.AUTOTUNE)
+
+    # Map labeled sentence pairs.
+    ds = tf.data.Dataset.from_tensor_slices(((first, second), label))
+    ds = ds.map(preprocessor, num_parallel_calls=tf.data.AUTOTUNE)
+
+    # Map unlabeled 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,
+    )
+    ```
+    """
+
+    tokenizer_cls = AlbertTokenizer
+
+    def __init__(
+        self,
+        tokenizer,
+        sequence_length=512,
+        truncate="round_robin",
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.tokenizer = tokenizer
+        self.packer = None
+        self.truncate = truncate
+        self.sequence_length = sequence_length
+
+    def build(self, input_shape):
+        # Defer packer creation to `build()` so that we can be sure tokenizer
+        # assets have loaded when restoring a saved model.
+        self.packer = MultiSegmentPacker(
+            start_value=self.tokenizer.cls_token_id,
+            end_value=self.tokenizer.sep_token_id,
+            pad_value=self.tokenizer.pad_token_id,
+            truncate=self.truncate,
+            sequence_length=self.sequence_length,
+        )
+        self.built = True
+
+    def get_config(self):
+        config = super().get_config()
+        config.update(
+            {
+                "sequence_length": self.sequence_length,
+                "truncate": self.truncate,
+            }
+        )
+        return config
+
+    def call(self, x, y=None, sample_weight=None):
+        x = convert_inputs_to_list_of_tensor_segments(x)
+        x = [self.tokenizer(segment) for segment in x]
+        token_ids, segment_ids = self.packer(x)
+        x = {
+            "token_ids": token_ids,
+            "segment_ids": segment_ids,
+            "padding_mask": token_ids != self.tokenizer.pad_token_id,
+        }
+        return keras.utils.pack_x_y_sample_weight(x, y, sample_weight)
+
+    @property
+    def sequence_length(self):
+        """The padded length of model input sequences."""
+        return self._sequence_length
+
+    @sequence_length.setter
+    def sequence_length(self, value):
+        self._sequence_length = value
+        if self.packer is not None:
+            self.packer.sequence_length = value

keras_hub/src/models/albert/albert_presets.py

@@ -0,0 +1,70 @@
+# Copyright 2024 The KerasHub Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""ALBERT model preset configurations."""
+
+
+backbone_presets = {
+    "albert_base_en_uncased": {
+        "metadata": {
+            "description": (
+                "12-layer ALBERT model where all input is lowercased. "
+                "Trained on English Wikipedia + BooksCorpus."
+            ),
+            "params": 11683584,
+            "official_name": "ALBERT",
+            "path": "albert",
+            "model_card": "https://github.com/google-research/albert/blob/master/README.md",
+        },
+        "kaggle_handle": "kaggle://keras/albert/keras/albert_base_en_uncased/2",
+    },
+    "albert_large_en_uncased": {
+        "metadata": {
+            "description": (
+                "24-layer ALBERT model where all input is lowercased. "
+                "Trained on English Wikipedia + BooksCorpus."
+            ),
+            "params": 17683968,
+            "official_name": "ALBERT",
+            "path": "albert",
+            "model_card": "https://github.com/google-research/albert/blob/master/README.md",
+        },
+        "kaggle_handle": "kaggle://keras/albert/keras/albert_large_en_uncased/2",
+    },
+    "albert_extra_large_en_uncased": {
+        "metadata": {
+            "description": (
+                "24-layer ALBERT model where all input is lowercased. "
+                "Trained on English Wikipedia + BooksCorpus."
+            ),
+            "params": 58724864,
+            "official_name": "ALBERT",
+            "path": "albert",
+            "model_card": "https://github.com/google-research/albert/blob/master/README.md",
+        },
+        "kaggle_handle": "kaggle://keras/albert/keras/albert_extra_large_en_uncased/2",
+    },
+    "albert_extra_extra_large_en_uncased": {
+        "metadata": {
+            "description": (
+                "12-layer ALBERT model where all input is lowercased. "
+                "Trained on English Wikipedia + BooksCorpus."
+            ),
+            "params": 222595584,
+            "official_name": "ALBERT",
+            "path": "albert",
+            "model_card": "https://github.com/google-research/albert/blob/master/README.md",
+        },
+        "kaggle_handle": "kaggle://keras/albert/keras/albert_extra_extra_large_en_uncased/2",
+    },
+}

keras_hub/src/models/albert/albert_tokenizer.py

@@ -0,0 +1,119 @@
+# Copyright 2024 The KerasHub Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.tokenizers.sentence_piece_tokenizer import (
+    SentencePieceTokenizer,
+)
+
+
+@keras_hub_export("keras_hub.models.AlbertTokenizer")
+class AlbertTokenizer(SentencePieceTokenizer):
+    """ALBERT tokenizer layer based on SentencePiece.
+
+    This tokenizer class will tokenize raw strings into integer sequences and
+    is based on `keras_hub.tokenizers.SentencePieceTokenizer`. Unlike the
+    underlying tokenizer, it will check for all special tokens needed by
+    ALBERT models and provides a `from_preset()` method to automatically
+    download a matching vocabulary for a ALBERT preset.
+
+    This tokenizer does not provide truncation or padding of inputs. It can be
+    combined with a `keras_hub.models.AlbertPreprocessor` layer for input
+    packing.
+
+    If input is a batch of strings (rank > 0), the layer will output a
+    `tf.RaggedTensor` where the last dimension of the output is ragged.
+
+    If input is a scalar string (rank == 0), the layer will output a dense
+    `tf.Tensor` with static shape `[None]`.
+
+    Args:
+        proto: Either a `string` path to a SentencePiece proto file, or a
+            `bytes` object with a serialized SentencePiece proto. See the
+            [SentencePiece repository](https://github.com/google/sentencepiece)
+            for more details on the format.
+
+    Examples:
+
+    ```python
+    # Unbatched input.
+    tokenizer = keras_hub.models.AlbertTokenizer.from_preset(
+        "albert_base_en_uncased",
+    )
+    tokenizer("The quick brown fox jumped.")
+
+    # Batched input.
+    tokenizer(["The quick brown fox jumped.", "The fox slept."])
+
+    # Detokenization.
+    tokenizer.detokenize(tokenizer("The quick brown fox jumped."))
+
+    # Custom vocabulary.
+    bytes_io = io.BytesIO()
+    ds = tf.data.Dataset.from_tensor_slices(["The quick brown fox jumped."])
+    sentencepiece.SentencePieceTrainer.train(
+        sentence_iterator=ds.as_numpy_iterator(),
+        model_writer=bytes_io,
+        vocab_size=10,
+        model_type="WORD",
+        pad_id=0,
+        unk_id=1,
+        bos_id=2,
+        eos_id=3,
+        pad_piece="<pad>",
+        unk_piece="<unk>",
+        bos_piece="[CLS]",
+        eos_piece="[SEP]",
+        user_defined_symbols="[MASK]",
+    )
+    tokenizer = keras_hub.models.AlbertTokenizer(
+        proto=bytes_io.getvalue(),
+    )
+    tokenizer("The quick brown fox jumped.")
+    ```
+    """
+
+    def __init__(self, proto, **kwargs):
+        self.cls_token = "[CLS]"
+        self.sep_token = "[SEP]"
+        self.pad_token = "<pad>"
+        self.mask_token = "[MASK]"
+
+        super().__init__(proto=proto, **kwargs)
+
+    def set_proto(self, proto):
+        super().set_proto(proto)
+        if proto is not None:
+            for token in [
+                self.cls_token,
+                self.sep_token,
+                self.pad_token,
+                self.mask_token,
+            ]:
+                if token not in self.get_vocabulary():
+                    raise ValueError(
+                        f"Cannot find token `'{token}'` in the provided "
+                        f"`vocabulary`. Please provide `'{token}'` in your "
+                        "`vocabulary` or use a pretrained `vocabulary` name."
+                    )
+
+            self.cls_token_id = self.token_to_id(self.cls_token)
+            self.sep_token_id = self.token_to_id(self.sep_token)
+            self.pad_token_id = self.token_to_id(self.pad_token)
+            self.mask_token_id = self.token_to_id(self.mask_token)
+        else:
+            self.cls_token_id = None
+            self.sep_token_id = None
+            self.pad_token_id = None
+            self.mask_token_id = None