Skip to content

GitLab

Commit 1abd53b1 authored by Lysandre's avatar Lysandre
Browse files

Patch ALBERT with heads in TensorFlow

parent e6767642
Hide whitespace changes
Inline Side-by-side
Showing with 146 additions and 136 deletions
src/transformers/modeling_tf_albert.py
View file @ 1abd53b1
...@@ -29,14 +29,14 @@ from .modeling_tf_utils import TFPreTrainedModel, get_initializer, shape_list ...@@ -29,14 +29,14 @@ from .modeling_tf_utils import TFPreTrainedModel, get_initializer, shape_list
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
TF_ALBERT_PRETRAINED_MODEL_ARCHIVE_MAP = { TF_ALBERT_PRETRAINED_MODEL_ARCHIVE_MAP = {
"albert-base-v1": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-base-v1-tf_model.h5", "albert-base-v1": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-base-v1-with-prefix-tf_model.h5",
"albert-large-v1": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-large-v1-tf_model.h5", "albert-large-v1": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-large-v1-with-prefix-tf_model.h5",
"albert-xlarge-v1": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xlarge-v1-tf_model.h5", "albert-xlarge-v1": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xlarge-v1-with-prefix-tf_model.h5",
"albert-xxlarge-v1": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xxlarge-v1-tf_model.h5", "albert-xxlarge-v1": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xxlarge-v1-with-prefix-tf_model.h5",
"albert-base-v2": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-base-v2-tf_model.h5", "albert-base-v2": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-base-v2-with-prefix-tf_model.h5",
"albert-large-v2": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-large-v2-tf_model.h5", "albert-large-v2": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-large-v2-with-prefix-tf_model.h5",
"albert-xlarge-v2": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xlarge-v2-tf_model.h5", "albert-xlarge-v2": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xlarge-v2-with-prefix-tf_model.h5",
"albert-xxlarge-v2": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xxlarge-v2-tf_model.h5", "albert-xxlarge-v2": "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xxlarge-v2-with-prefix-tf_model.h5",
} }
...@@ -478,88 +478,7 @@ class TFAlbertMLMHead(tf.keras.layers.Layer): ...@@ -478,88 +478,7 @@ class TFAlbertMLMHead(tf.keras.layers.Layer):
return hidden_states return hidden_states
ALBERT_START_DOCSTRING = r""" class TFAlbertMainLayer(tf.keras.layers.Layer):
This model is a `tf.keras.Model <https://www.tensorflow.org/api_docs/python/tf/keras/Model>`__ sub-class.
Use it as a regular TF 2.0 Keras Model and
refer to the TF 2.0 documentation for all matter related to general usage and behavior.
.. _`ALBERT: A Lite BERT for Self-supervised Learning of Language Representations`:
https://arxiv.org/abs/1909.11942
.. _`tf.keras.Model`:
https://www.tensorflow.org/versions/r2.0/api_docs/python/tf/keras/Model
.. note::
TF 2.0 models accepts two formats as inputs:
- having all inputs as keyword arguments (like PyTorch models), or
- having all inputs as a list, tuple or dict in the first positional arguments.
This second option is useful when using :obj:`tf.keras.Model.fit()` method which currently requires having
all the tensors in the first argument of the model call function: :obj:`model(inputs)`.
If you choose this second option, there are three possibilities you can use to gather all the input Tensors
in the first positional argument :
- a single Tensor with input_ids only and nothing else: :obj:`model(inputs_ids)`
- a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
:obj:`model([input_ids, attention_mask])` or :obj:`model([input_ids, attention_mask, token_type_ids])`
- a dictionary with one or several input Tensors associated to the input names given in the docstring:
:obj:`model({'input_ids': input_ids, 'token_type_ids': token_type_ids})`
Args:
config (:class:`~transformers.AlbertConfig`): Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the configuration.
Check out the :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model weights.
"""
ALBERT_INPUTS_DOCSTRING = r"""
Args:
input_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`):
Indices of input sequence tokens in the vocabulary.
Indices can be obtained using :class:`transformers.AlbertTokenizer`.
See :func:`transformers.PreTrainedTokenizer.encode` and
:func:`transformers.PreTrainedTokenizer.encode_plus` for details.
`What are input IDs? <../glossary.html#input-ids>`__
attention_mask (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional, defaults to :obj:`None`):
Mask to avoid performing attention on padding token indices.
Mask values selected in ``[0, 1]``:
``1`` for tokens that are NOT MASKED, ``0`` for MASKED tokens.
`What are attention masks? <../glossary.html#attention-mask>`__
token_type_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
Segment token indices to indicate first and second portions of the inputs.
Indices are selected in ``[0, 1]``: ``0`` corresponds to a `sentence A` token, ``1``
corresponds to a `sentence B` token
`What are token type IDs? <../glossary.html#token-type-ids>`_
position_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
Indices of positions of each input sequence tokens in the position embeddings.
Selected in the range ``[0, config.max_position_embeddings - 1]``.
`What are position IDs? <../glossary.html#position-ids>`_
head_mask (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`, defaults to :obj:`None`):
Mask to nullify selected heads of the self-attention modules.
Mask values selected in ``[0, 1]``:
``1`` indicates the head is **not masked**, ``0`` indicates the head is **masked**.
input_embeds (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`, defaults to :obj:`None`):
Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
This is useful if you want more control over how to convert `input_ids` indices into associated vectors
than the model's internal embedding lookup matrix.
training (:obj:`boolean`, `optional`, defaults to :obj:`False`):
Whether to activate dropout modules (if set to :obj:`True`) during training or to de-activate them
(if set to :obj:`False`) for evaluation.
"""
@add_start_docstrings(
"The bare Albert Model transformer outputing raw hidden-states without any specific head on top.",
ALBERT_START_DOCSTRING,
)
class TFAlbertModel(TFAlbertPreTrainedModel):
def __init__(self, config, **kwargs): def __init__(self, config, **kwargs):
super().__init__(config, **kwargs) super().__init__(config, **kwargs)
self.num_hidden_layers = config.num_hidden_layers self.num_hidden_layers = config.num_hidden_layers
...@@ -586,52 +505,16 @@ class TFAlbertModel(TFAlbertPreTrainedModel): ...@@ -586,52 +505,16 @@ class TFAlbertModel(TFAlbertPreTrainedModel):
""" """
raise NotImplementedError raise NotImplementedError
@add_start_docstrings_to_callable(ALBERT_INPUTS_DOCSTRING)
def call( def call(
self, self,
inputs, inputs,
attention_mask=None, attention_mask=None,
token_type_ids=None, token_type_ids=None,
position_ids=None, position_ids=None,
head_mask=None, head_mask=None,
inputs_embeds=None, inputs_embeds=None,
training=False, training=False,
): ):
r"""
Returns:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.AlbertConfig`) and inputs:
last_hidden_state (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`):
Sequence of hidden-states at the output of the last layer of the model.
pooler_output (:obj:`tf.Tensor` of shape :obj:`(batch_size, hidden_size)`):
Last layer hidden-state of the first token of the sequence (classification token)
further processed by a Linear layer and a Tanh activation function. The Linear
layer weights are trained from the next sentence prediction (classification)
objective during Albert pretraining. This output is usually *not* a good summary
of the semantic content of the input, you're often better with averaging or pooling
the sequence of hidden-states for the whole input sequence.
hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
of shape :obj:`(batch_size, sequence_length, hidden_size)`.
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
tuple of :obj:`tf.Tensor` (one for each layer) of shape
:obj:`(batch_size, num_heads, sequence_length, sequence_length)`:
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
Examples::
import tensorflow as tf
from transformers import AlbertTokenizer, TFAlbertModel
tokenizer = AlbertTokenizer.from_pretrained('albert-base-v2')
model = TFAlbertModel.from_pretrained('albert-base-v2')
input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :] # Batch size 1
outputs = model(input_ids)
last_hidden_states = outputs[0] # The last hidden-state is the first element of the output tuple
"""
if isinstance(inputs, (tuple, list)): if isinstance(inputs, (tuple, list)):
input_ids = inputs[0] input_ids = inputs[0]
attention_mask = inputs[1] if len(inputs) > 1 else attention_mask attention_mask = inputs[1] if len(inputs) > 1 else attention_mask
...@@ -704,12 +587,139 @@ class TFAlbertModel(TFAlbertPreTrainedModel): ...@@ -704,12 +587,139 @@ class TFAlbertModel(TFAlbertPreTrainedModel):
return outputs return outputs
ALBERT_START_DOCSTRING = r"""
This model is a `tf.keras.Model <https://www.tensorflow.org/api_docs/python/tf/keras/Model>`__ sub-class.
Use it as a regular TF 2.0 Keras Model and
refer to the TF 2.0 documentation for all matter related to general usage and behavior.
.. _`ALBERT: A Lite BERT for Self-supervised Learning of Language Representations`:
https://arxiv.org/abs/1909.11942
.. _`tf.keras.Model`:
https://www.tensorflow.org/versions/r2.0/api_docs/python/tf/keras/Model
.. note::
TF 2.0 models accepts two formats as inputs:
- having all inputs as keyword arguments (like PyTorch models), or
- having all inputs as a list, tuple or dict in the first positional arguments.
This second option is useful when using :obj:`tf.keras.Model.fit()` method which currently requires having
all the tensors in the first argument of the model call function: :obj:`model(inputs)`.
If you choose this second option, there are three possibilities you can use to gather all the input Tensors
in the first positional argument :
- a single Tensor with input_ids only and nothing else: :obj:`model(inputs_ids)`
- a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
:obj:`model([input_ids, attention_mask])` or :obj:`model([input_ids, attention_mask, token_type_ids])`
- a dictionary with one or several input Tensors associated to the input names given in the docstring:
:obj:`model({'input_ids': input_ids, 'token_type_ids': token_type_ids})`
Args:
config (:class:`~transformers.AlbertConfig`): Model configuration class with all the parameters of the model.
Initializing with a config file does not load the weights associated with the model, only the configuration.
Check out the :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model weights.
"""
ALBERT_INPUTS_DOCSTRING = r"""
Args:
input_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`):
Indices of input sequence tokens in the vocabulary.
Indices can be obtained using :class:`transformers.AlbertTokenizer`.
See :func:`transformers.PreTrainedTokenizer.encode` and
:func:`transformers.PreTrainedTokenizer.encode_plus` for details.
`What are input IDs? <../glossary.html#input-ids>`__
attention_mask (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional, defaults to :obj:`None`):
Mask to avoid performing attention on padding token indices.
Mask values selected in ``[0, 1]``:
``1`` for tokens that are NOT MASKED, ``0`` for MASKED tokens.
`What are attention masks? <../glossary.html#attention-mask>`__
token_type_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
Segment token indices to indicate first and second portions of the inputs.
Indices are selected in ``[0, 1]``: ``0`` corresponds to a `sentence A` token, ``1``
corresponds to a `sentence B` token
`What are token type IDs? <../glossary.html#token-type-ids>`_
position_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
Indices of positions of each input sequence tokens in the position embeddings.
Selected in the range ``[0, config.max_position_embeddings - 1]``.
`What are position IDs? <../glossary.html#position-ids>`_
head_mask (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`, defaults to :obj:`None`):
Mask to nullify selected heads of the self-attention modules.
Mask values selected in ``[0, 1]``:
``1`` indicates the head is **not masked**, ``0`` indicates the head is **masked**.
input_embeds (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`, defaults to :obj:`None`):
Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
This is useful if you want more control over how to convert `input_ids` indices into associated vectors
than the model's internal embedding lookup matrix.
training (:obj:`boolean`, `optional`, defaults to :obj:`False`):
Whether to activate dropout modules (if set to :obj:`True`) during training or to de-activate them
(if set to :obj:`False`) for evaluation.
"""
@add_start_docstrings(
"The bare Albert Model transformer outputing raw hidden-states without any specific head on top.",
ALBERT_START_DOCSTRING,
)
class TFAlbertModel(TFAlbertPreTrainedModel):
def __init__(self, config, *inputs, **kwargs):
super().__init__(config, *inputs, **kwargs)
self.albert = TFAlbertMainLayer(config, name="albert")
@add_start_docstrings_to_callable(ALBERT_INPUTS_DOCSTRING)
def call(self, inputs, **kwargs):
r"""
Returns:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.AlbertConfig`) and inputs:
last_hidden_state (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`):
Sequence of hidden-states at the output of the last layer of the model.
pooler_output (:obj:`tf.Tensor` of shape :obj:`(batch_size, hidden_size)`):
Last layer hidden-state of the first token of the sequence (classification token)
further processed by a Linear layer and a Tanh activation function. The Linear
layer weights are trained from the next sentence prediction (classification)
objective during Albert pretraining. This output is usually *not* a good summary
of the semantic content of the input, you're often better with averaging or pooling
the sequence of hidden-states for the whole input sequence.
hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
of shape :obj:`(batch_size, sequence_length, hidden_size)`.
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
tuple of :obj:`tf.Tensor` (one for each layer) of shape
:obj:`(batch_size, num_heads, sequence_length, sequence_length)`:
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
Examples::
import tensorflow as tf
from transformers import AlbertTokenizer, TFAlbertModel
tokenizer = AlbertTokenizer.from_pretrained('albert-base-v2')
model = TFAlbertModel.from_pretrained('albert-base-v2')
input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :] # Batch size 1
outputs = model(input_ids)
last_hidden_states = outputs[0] # The last hidden-state is the first element of the output tuple
"""
outputs = self.albert(inputs, **kwargs)
return outputs
@add_start_docstrings("""Albert Model with a `language modeling` head on top. """, ALBERT_START_DOCSTRING) @add_start_docstrings("""Albert Model with a `language modeling` head on top. """, ALBERT_START_DOCSTRING)
class TFAlbertForMaskedLM(TFAlbertPreTrainedModel): class TFAlbertForMaskedLM(TFAlbertPreTrainedModel):
def __init__(self, config, *inputs, **kwargs): def __init__(self, config, *inputs, **kwargs):
super(TFAlbertForMaskedLM, self).__init__(config, *inputs, **kwargs) super(TFAlbertForMaskedLM, self).__init__(config, *inputs, **kwargs)
self.albert = TFAlbertModel(config, name="albert") self.albert = TFAlbertMainLayer(config, name="albert")
self.predictions = TFAlbertMLMHead(config, self.albert.embeddings, name="predictions") self.predictions = TFAlbertMLMHead(config, self.albert.embeddings, name="predictions")
def get_output_embeddings(self): def get_output_embeddings(self):
...@@ -766,7 +776,7 @@ class TFAlbertForSequenceClassification(TFAlbertPreTrainedModel): ...@@ -766,7 +776,7 @@ class TFAlbertForSequenceClassification(TFAlbertPreTrainedModel):
super(TFAlbertForSequenceClassification, self).__init__(config, *inputs, **kwargs) super(TFAlbertForSequenceClassification, self).__init__(config, *inputs, **kwargs)
self.num_labels = config.num_labels self.num_labels = config.num_labels
self.albert = TFAlbertModel(config, name="albert") self.albert = TFAlbertMainLayer(config, name="albert")
self.dropout = tf.keras.layers.Dropout(config.hidden_dropout_prob) self.dropout = tf.keras.layers.Dropout(config.hidden_dropout_prob)
self.classifier = tf.keras.layers.Dense( self.classifier = tf.keras.layers.Dense(
config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="classifier" config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="classifier"
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment