microsoft
/
huggingface-transformers
зеркало из https://github.com/microsoft/huggingface-transformers.git
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность

Update the new model template (#6019)

This commit is contained in:
Sylvain Gugger 2020-07-24 14:15:37 -04:00 коммит произвёл GitHub
Родитель 295466aae6
Коммит a884b7fa38
Не найден ключ, соответствующий данной подписи
Идентификатор ключа GPG: 4AEE18F83AFDEB23
5 изменённых файлов: 998 добавлений и 580 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

96
templates/adding_a_new_model/README.md
Просмотреть файл

@ -1,64 +1,90 @@
# How to add a new model in 🤗Transformers
# How to add a new model in 🤗 Transformers
This folder describes the process to add a new model in 🤗Transformers and provide templates for the required files.
This folder describes the process to add a new model in 🤗 Transformers and provide templates for the required files.
The library is designed to incorporate a variety of models and code bases. As such the process for adding a new model usually mostly consists in copy-pasting to relevant original code in the various sections of the templates included in the present repository.
The library is designed to incorporate a variety of models and code bases. As such the process for adding a new model
usually mostly consists in copy-pasting to relevant original code in the various sections of the templates included in
the present repository.
One important point though is that the library has the following goals impacting the way models are incorporated:
- one specific feature of the API is the capability to run the model and tokenizer inline. The tokenization code thus often have to be slightly adapted to allow for running in the python interpreter.
- the package is also designed to be as self-consistent and with a small and reliable set of packages dependencies. In consequence, additional dependencies are usually not allowed when adding a model but can be allowed for the inclusion of a new tokenizer (recent examples of dependencies added for tokenizer specificities include `sentencepiece` and `sacremoses`). Please make sure to check the existing dependencies when possible before adding a new one.
- One specific feature of the API is the capability to run the model and tokenizer inline. The tokenization code thus
often have to be slightly adapted to allow for running in the python interpreter.
- the package is also designed to be as self-consistent and with a small and reliable set of packages dependencies. In
consequence, additional dependencies are usually not allowed when adding a model but can be allowed for the
inclusion of a new tokenizer (recent examples of dependencies added for tokenizer specificities include
`sentencepiece` and `sacremoses`). Please make sure to check the existing dependencies when possible before adding a
new one.
For a quick overview of the library organization, please check the [QuickStart section of the documentation](https://huggingface.co/transformers/quickstart.html).
For a quick overview of the general philosphy of the library and its organization, please check the
[QuickStart section of the documentation](https://huggingface.co/transformers/philosophy.html).
# Typical workflow for including a model
Here an overview of the general workflow:
- [ ] add model/configuration/tokenization classes
- [ ] add conversion scripts
- [ ] add tests
- [ ] add @slow integration test
- [ ] finalize
- [ ] Add model/configuration/tokenization classes.
- [ ] Add conversion scripts.
- [ ] Add tests and a @slow integration test.
- [ ] Document your model.
- [ ] Finalize.
Let's detail what should be done at each step
Let's detail what should be done at each step.
## Adding model/configuration/tokenization classes
Here is the workflow for adding model/configuration/tokenization classes:
- [ ] copy the python files from the present folder to the main folder and rename them, replacing `xxx` with your model name,
- [ ] edit the files to replace `XXX` (with various casing) with your model name
- [ ] copy-paste or create a simple configuration class for your model in the `configuration_...` file
- [ ] copy-paste or create the code for your model in the `modeling_...` files (PyTorch and TF 2.0)
- [ ] copy-paste or create a tokenizer class for your model in the `tokenization_...` file
- [ ] Copy the python files from the present folder to the main folder and rename them, replacing `xxx` with your model
name.
- [ ] Edit the files to replace `XXX` (with various casing) with your model name.
- [ ] Copy-paste or create a simple configuration class for your model in the `configuration_...` file.
- [ ] Copy-paste or create the code for your model in the `modeling_...` files (PyTorch and TF 2.0).
- [ ] Copy-paste or create a tokenizer class for your model in the `tokenization_...` file.
# Adding conversion scripts
## Adding conversion scripts
Here is the workflow for the conversion scripts:
- [ ] copy the conversion script (`convert_...`) from the present folder to the main folder.
- [ ] edit this script to convert your original checkpoint weights to the current pytorch ones.
- [ ] Copy the conversion script (`convert_...`) from the present folder to the main folder.
- [ ] Edit this script to convert your original checkpoint weights to the current pytorch ones.
# Adding tests:
## Adding tests:
Here is the workflow for the adding tests:
- [ ] copy the python files from the `tests` sub-folder of the present folder to the `tests` subfolder of the main folder and rename them, replacing `xxx` with your model name,
- [ ] edit the tests files to replace `XXX` (with various casing) with your model name
- [ ] edit the tests code as needed
- [ ] Copy the python files from the `tests` sub-folder of the present folder to the `tests` subfolder of the main
folder and rename them, replacing `xxx` with your model name.
- [ ] Edit the tests files to replace `XXX` (with various casing) with your model name.
- [ ] Edit the tests code as needed.
# Final steps
## Documenting your model:
Here is the workflow for documentation:
- [ ] Make sure all your arguments are properly documened in your configuration and tokenizer.
- [ ] Most of the documentation of the models is automatically generated, you just ahve to male sure that
`XXX_START_DOCSTRING` contains an introduction to the model you're adding and a link to the original
article and that `XXX_INPUTS_DOCSTRING` contains all the inputs of your model.
- [ ] Create a new page `xxx.rst` in the folder `docs/source/model_doc` and add this file in `docs/source/index.rst`.
Make sure to check you have no sphinx warnings when building the documentation locally and follow our
[documentaiton guide](https://github.com/huggingface/transformers/tree/master/docs#writing-documentation---specification).
## Final steps
You can then finish the addition step by adding imports for your classes in the common files:
- [ ] add import for all the relevant classes in `__init__.py`
- [ ] add your configuration in `configuration_auto.py`
- [ ] add your PyTorch and TF 2.0 model respectively in `modeling_auto.py` and `modeling_tf_auto.py`
- [ ] add your tokenizer in `tokenization_auto.py`
- [ ] add your models and tokenizer to `pipeline.py`
- [ ] add a link to your conversion script in the main conversion utility (in `commands/convert.py`)
- [ ] edit the PyTorch to TF 2.0 conversion script to add your model in the `convert_pytorch_checkpoint_to_tf2.py` file
- [ ] add a mention of your model in the doc: `README.md` and the documentation itself at `docs/source/pretrained_models.rst`.
- [ ] upload the pretrained weights, configurations and vocabulary files.
- [ ] create model card(s) for your models on huggingface.co. For those last two steps, check the [model sharing documentation](https://github.com/huggingface/transformers#quick-tour-of-model-sharing).
- [ ] Add import for all the relevant classes in `__init__.py`.
- [ ] Add your configuration in `configuration_auto.py`.
- [ ] Add your PyTorch and TF 2.0 model respectively in `modeling_auto.py` and `modeling_tf_auto.py`.
- [ ] Add your tokenizer in `tokenization_auto.py`.
- [ ] Add your models and tokenizer to `pipeline.py`.
- [ ] Add a link to your conversion script in the main conversion utility (in `commands/convert.py`)
- [ ] Edit the PyTorch to TF 2.0 conversion script to add your model in the `convert_pytorch_checkpoint_to_tf2.py`
file.
- [ ] Add a mention of your model in the doc: `README.md` and the documentation itself
in `docs/source/index.rst` and `docs/source/pretrained_models.rst`.
- [ ] Upload the pretrained weights, configurations and vocabulary files.
- [ ] Create model card(s) for your models on huggingface.co. For those last two steps, check the
[model sharing documentation](https://huggingface.co/transformers/model_sharing.html).

132
templates/adding_a_new_model/configuration_xxx.py
Просмотреть файл

@ -16,6 +16,7 @@
import logging
from typing import Callable, Union
from .configuration_utils import PretrainedConfig
@ -30,85 +31,76 @@ XXX_PRETRAINED_CONFIG_ARCHIVE_MAP = {
class XxxConfig(PretrainedConfig):
r"""
:class:`~transformers.XxxConfig` is the configuration class to store the configuration of a
`XxxModel`.
This is the configuration class to store the configuration of a :class:`~transformers.XXXModel`.
It is used to instantiate a XXX model according to the specified arguments, defining the model
architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of
the XXX `xxx-base-uncased <https://huggingface.co/xxx/xxx-base-uncased>`__ architecture.
Configuration objects inherit from :class:`~transformers.PretrainedConfig` and can be used
to control the model outputs. Read the documentation from :class:`~transformers.PretrainedConfig`
for more information.
Arguments:
vocab_size: Vocabulary size of `inputs_ids` in `XxxModel`.
hidden_size: Size of the encoder layers and the pooler layer.
num_hidden_layers: Number of hidden layers in the Transformer encoder.
num_attention_heads: Number of attention heads for each attention layer in
the Transformer encoder.
intermediate_size: The size of the "intermediate" (i.e., feed-forward)
layer in the Transformer encoder.
hidden_act: The non-linear activation function (function or string) in the
encoder and pooler. If string, "gelu", "relu", "swish" and "gelu_new" are supported.
hidden_dropout_prob: The dropout probabilitiy for all fully connected
layers in the embeddings, encoder, and pooler.
attention_probs_dropout_prob: The dropout ratio for the attention
probabilities.
max_position_embeddings: The maximum sequence length that this model might
ever be used with. Typically set this to something large just in case
(e.g., 512 or 1024 or 2048).
type_vocab_size: The vocabulary size of the `token_type_ids` passed into
`XxxModel`.
initializer_range: The sttdev of the truncated_normal_initializer for
initializing all weight matrices.
layer_norm_eps: The epsilon used by LayerNorm.
Args:
vocab_size (:obj:`int`, optional, defaults to 30522):
Vocabulary size of the XXX model. Defines the different tokens that
can be represented by the `inputs_ids` passed to the forward method of :class:`~transformers.XXXModel`.
hidden_size (:obj:`int`, optional, defaults to 768):
Dimensionality of the encoder layers and the pooler layer.
num_hidden_layers (:obj:`int`, optional, defaults to 12):
Number of hidden layers in the Transformer encoder.
num_attention_heads (:obj:`int`, optional, defaults to 12):
Number of attention heads for each attention layer in the Transformer encoder.
hidden_act (:obj:`str` or :obj:`function`, optional, defaults to :obj:`"gelu"`):
The non-linear activation function (function or string) in the encoder and pooler.
If string, :obj:`"gelu"`, :obj:`"relu"`, :obj:`"swish"` and :obj:`"gelu_new"` are supported.
hidden_dropout_prob (:obj:`float`, optional, defaults to 0.1):
The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
attention_probs_dropout_prob (:obj:`float`, optional, defaults to 0.1):
The dropout ratio for the attention probabilities.
max_position_embeddings (:obj:`int`, optional, defaults to 512):
The maximum sequence length that this model might ever be used with.
Typically set this to something large just in case (e.g., 512 or 1024 or 2048).
type_vocab_size (:obj:`int`, optional, defaults to 2):
The vocabulary size of the `token_type_ids` passed into :class:`~transformers.BertModel`.
initializer_range (:obj:`float`, optional, defaults to 0.02):
The standard deviation of the :obj:`truncated_normal_initializer` for initializing all weight matrices.
layer_norm_eps (:obj:`float`, optional, defaults to 1e-5):
The epsilon used by the layer normalization layers.
gradient_checkpointing (:obj:`bool`, optional, defaults to :obj:`False`):
If :obj:`True`, use gradient checkpointing to save memory at the expense of slower backward pass.
kwargs:
Additional arguments for common configurations, passed to :class:`~transformers.PretrainedConfig`.
"""
model_type = "xxx"
def __init__(
self,
vocab_size=50257,
n_positions=1024,
n_ctx=1024,
n_embd=768,
n_layer=12,
n_head=12,
resid_pdrop=0.1,
embd_pdrop=0.1,
attn_pdrop=0.1,
layer_norm_epsilon=1e-5,
initializer_range=0.02,
summary_type="cls_index",
summary_use_proj=True,
summary_activation=None,
summary_proj_to_labels=True,
summary_first_dropout=0.1,
vocab_size: int = 50257,
hidden_size: int = 1024,
num_hidden_layers: int = 12,
num_attention_heads: int = 12,
hidden_act: Union[str, Callable] = "gelu",
hidden_dropout_prob: float = 0.1,
attention_probs_dropout_prob: float = 0.1,
max_position_embeddings: int = 512,
type_vocab_size: int = 2,
initializer_range: float = 0.02,
layer_norm_epsilon: float = 1e-5,
gradient_checkpointing: bool = False,
**kwargs
):
super().__init__(**kwargs)
self.vocab_size = vocab_size
self.n_ctx = n_ctx
self.n_positions = n_positions
self.n_embd = n_embd
self.n_layer = n_layer
self.n_head = n_head
self.resid_pdrop = resid_pdrop
self.embd_pdrop = embd_pdrop
self.attn_pdrop = attn_pdrop
self.layer_norm_epsilon = layer_norm_epsilon
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.initializer_range = initializer_range
self.summary_type = summary_type
self.summary_use_proj = summary_use_proj
self.summary_activation = summary_activation
self.summary_first_dropout = summary_first_dropout
self.summary_proj_to_labels = summary_proj_to_labels
@property
def max_position_embeddings(self):
return self.n_positions
@property
def hidden_size(self):
return self.n_embd
@property
def num_attention_heads(self):
return self.n_head
@property
def num_hidden_layers(self):
return self.n_layer
self.layer_norm_epsilon = layer_norm_epsilon
self.gradient_checkpointing = gradient_checkpointing

691
templates/adding_a_new_model/modeling_tf_xxx.py
Просмотреть файл

@ -25,12 +25,29 @@ import logging
import tensorflow as tf
from .configuration_xxx import XxxConfig
from .file_utils import add_start_docstrings
from .modeling_tf_utils import TFPreTrainedModel, get_initializer, shape_list
from .file_utils import (
MULTIPLE_CHOICE_DUMMY_INPUTS,
add_code_sample_docstrings,
add_start_docstrings,
add_start_docstrings_to_callable,
)
from .modeling_tf_utils import (
TFMaskedLanguageModelingLoss,
TFMultipleChoiceLoss,
TFPreTrainedModel,
TFQuestionAnsweringLoss,
TFSequenceClassificationLoss,
TFTokenClassificationLoss,
get_initializer,
shape_list,
)
from .tokenization_utils import BatchEncoding
logger = logging.getLogger(__name__)
_TOKENIZER_FOR_DOC = "XxxTokenizer"
####################################################
# This list contrains shortcut names for some of
# the pretrained weights provided with the models
@ -183,36 +200,33 @@ class TFXxxPreTrainedModel(TFPreTrainedModel):
base_model_prefix = "transformer"
XXX_START_DOCSTRING = r""" The XXX model was proposed in
`XXX: Pre-training of Deep Bidirectional Transformers for Language Understanding`_
by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova. It's a bidirectional transformer
pre-trained using a combination of masked language modeling objective and next sentence prediction
on a large corpus comprising the Toronto Book Corpus and Wikipedia.
XXX_START_DOCSTRING = r"""
The XXX model was proposed in
`XXX: Pre-training of Deep Bidirectional Transformers for Language Understanding
<https://arxiv.org/abs/1810.04805>`__ by....
This model is a tf.keras.Model `tf.keras.Model`_ sub-class. Use it as a regular TF 2.0 Keras Model and
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.
.. _`XXX: Pre-training of Deep Bidirectional Transformers for Language Understanding`:
https://arxiv.org/abs/1810.04805
.. note::
.. _`tf.keras.Model`:
https://www.tensorflow.org/versions/r2.0/api_docs/python/tf/keras/Model
Note on the model inputs:
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 usefull when using `tf.keras.Model.fit()` method which currently requires having all the tensors in the first argument of the model call function: `model(inputs)`.
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 :
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: `model(inputs_ids)
- 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:
`model([input_ids, attention_mask])` or `model([input_ids, attention_mask, token_type_ids])`
- a dictionary with one or several input Tensors associaed to the input names given in the docstring:
`model({'input_ids': input_ids, 'token_type_ids': token_type_ids})`
: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})`
Parameters:
config (:class:`~transformers.XxxConfig`): Model configuration class with all the parameters of the model.
@ -221,95 +235,84 @@ XXX_START_DOCSTRING = r""" The XXX model was proposed in
"""
XXX_INPUTS_DOCSTRING = r"""
Inputs:
**input_ids**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
Args:
input_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`{0}`):
Indices of input sequence tokens in the vocabulary.
To match pre-training, XXX input sequence should be formatted with [CLS] and [SEP] tokens as follows:
(a) For sequence pairs:
``tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]``
``token_type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1``
(b) For single sequences:
``tokens: [CLS] the dog is hairy . [SEP]``
``token_type_ids: 0 0 0 0 0 0 0``
Xxx is a model with absolute position embeddings so it's usually advised to pad the inputs on
the right rather than the left.
Indices can be obtained using :class:`transformers.XxxTokenizer`.
See :func:`transformers.PreTrainedTokenizer.encode` and
:func:`transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.
**attention_mask**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
:func:`transformers.PreTrainedTokenizer.__call__` for details.
`What are input IDs? <../glossary.html#input-ids>`__
attention_mask (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`{0}`, `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.
**token_type_ids**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
`What are attention masks? <../glossary.html#attention-mask>`__
token_type_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`{0}`, `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
(see `XXX: Pre-training of Deep Bidirectional Transformers for Language Understanding`_ for more details).
**position_ids**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length)``:
`What are token type IDs? <../glossary.html#token-type-ids>`__
position_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`{0}`, `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]``.
**head_mask**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(num_heads,)`` or ``(num_layers, num_heads)``:
`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**.
**inputs_embeds**: (`optional`) ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length, embedding_dim)``:
Optionally, instead of passing ``input_ids`` you can choose to directly pass an embedded representation.
:obj:`1` indicates the head is **not masked**, :obj:`0` indicates the head is **masked**.
inputs_embeds (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, embedding_dim)`, `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.
output_attentions (:obj:`bool`, `optional`, defaults to :obj:`None`):
If set to ``True``, the attentions tensors of all attention layers are returned. See ``attentions`` under returned tensors for more detail.
"""
@add_start_docstrings(
"The bare Xxx Model transformer outputing raw hidden-states without any specific head on top.",
"The bare XXX Model transformer outputing raw hidden-states without any specific head on top.",
XXX_START_DOCSTRING,
XXX_INPUTS_DOCSTRING,
)
class TFXxxModel(TFXxxPreTrainedModel):
r"""
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**last_hidden_state**: ``tf.Tensor`` of shape ``(batch_size, sequence_length, hidden_size)``
Sequence of hidden-states at the output of the last layer of the model.
**pooler_output**: ``tf.Tensor`` of shape ``(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 Xxx 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**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``tf.Tensor`` (one for each layer) of shape ``(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 XxxTokenizer, TFXxxModel
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = TFXxxModel.from_pretrained('xxx-base-uncased')
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
"""
def __init__(self, config, *inputs, **kwargs):
super().__init__(config, *inputs, **kwargs)
self.transformer = TFXxxMainLayer(config, name="transformer")
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
@add_code_sample_docstrings(tokenizer_class=_TOKENIZER_FOR_DOC, checkpoint="xxx-base-cased")
def call(self, inputs, **kwargs):
r"""
Returns:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.XxxConfig`) 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 XXX 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 ``output_hidden_states=True`` is passed or when ``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 ``output_attentions=True`` is passed or 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.
"""
outputs = self.transformer(inputs, **kwargs)
return outputs
@ -317,84 +320,89 @@ class TFXxxModel(TFXxxPreTrainedModel):
TFXxxMLMHead = tf.keras.layers.Layer
@add_start_docstrings(
"""Xxx Model with a `language modeling` head on top. """, XXX_START_DOCSTRING, XXX_INPUTS_DOCSTRING
)
class TFXxxForMaskedLM(TFXxxPreTrainedModel):
r"""
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**prediction_scores**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length, config.vocab_size)``
Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``Numpy array`` or ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``Numpy array`` or ``tf.Tensor`` (one for each layer) of shape ``(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 XxxTokenizer, TFXxxForMaskedLM
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = TFXxxForMaskedLM.from_pretrained('xxx-base-uncased')
input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :] # Batch size 1
outputs = model(input_ids)
prediction_scores = outputs[0]
"""
@add_start_docstrings("""Xxx Model with a `language modeling` head on top. """, XXX_START_DOCSTRING)
class TFXxxForMaskedLM(TFXxxPreTrainedModel, TFMaskedLanguageModelingLoss):
def __init__(self, config, *inputs, **kwargs):
super().__init__(config, *inputs, **kwargs)
self.transformer = TFXxxMainLayer(config, name="transformer")
self.mlm = TFXxxMLMHead(config, self.transformer.embeddings, name="mlm")
def call(self, inputs, **kwargs):
outputs = self.transformer(inputs, **kwargs)
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
@add_code_sample_docstrings(tokenizer_class=_TOKENIZER_FOR_DOC, checkpoint="xxx-base-cased")
def call(
self,
inputs=None,
attention_mask=None,
token_type_ids=None,
position_ids=None,
head_mask=None,
inputs_embeds=None,
output_attentions=None,
output_hidden_states=None,
labels=None,
training=False,
):
r"""
labels (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
Labels for computing the masked language modeling loss.
Indices should be in ``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring)
Tokens with indices set to ``-100`` are ignored (masked), the loss is only computed for the tokens with labels
in ``[0, ..., config.vocab_size]``
Return:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.XxxConfig`) and inputs:
prediction_scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, config.vocab_size)`):
Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``output_hidden_states=True`` is passed or when ``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 ``output_attentions=True`` is passed or 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.
"""
if isinstance(inputs, (tuple, list)):
labels = inputs[8] if len(inputs) > 8 else labels
if len(inputs) > 8:
inputs = inputs[:8]
elif isinstance(inputs, (dict, BatchEncoding)):
labels = inputs.pop("labels", labels)
outputs = self.transformer(
inputs,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
training=training,
)
sequence_output = outputs[0]
prediction_scores = self.mlm(sequence_output, training=kwargs.get("training", False))
prediction_scores = self.mlm(sequence_output, training=training)
outputs = (prediction_scores,) + outputs[2:] # Add hidden states and attention if they are here
return outputs # prediction_scores, (hidden_states), (attentions)
if labels is not None:
loss = self.compute_loss(labels, prediction_scores)
outputs = (loss,) + outputs
return outputs # (loss), prediction_scores, (hidden_states), (attentions)
@add_start_docstrings(
"""Xxx Model transformer with a sequence classification/regression head on top (a linear layer on top of
"""XXX Model transformer with a sequence classification/regression head on top (a linear layer on top of
the pooled output) e.g. for GLUE tasks. """,
XXX_START_DOCSTRING,
XXX_INPUTS_DOCSTRING,
)
class TFXxxForSequenceClassification(TFXxxPreTrainedModel):
r"""
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**logits**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, config.num_labels)``
Classification (or regression if config.num_labels==1) scores (before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``Numpy array`` or ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``Numpy array`` or ``tf.Tensor`` (one for each layer) of shape ``(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 XxxTokenizer, TFXxxForSequenceClassification
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = TFXxxForSequenceClassification.from_pretrained('xxx-base-uncased')
input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :] # Batch size 1
outputs = model(input_ids)
logits = outputs[0]
"""
class TFXxxForSequenceClassification(TFXxxPreTrainedModel, TFSequenceClassificationLoss):
def __init__(self, config, *inputs, **kwargs):
super().__init__(config, *inputs, **kwargs)
self.num_labels = config.num_labels
@ -405,51 +413,216 @@ class TFXxxForSequenceClassification(TFXxxPreTrainedModel):
config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="classifier"
)
def call(self, inputs, **kwargs):
outputs = self.transformer(inputs, **kwargs)
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING)
@add_code_sample_docstrings(tokenizer_class=_TOKENIZER_FOR_DOC, checkpoint="xxx-base-cased")
def call(
self,
inputs=None,
attention_mask=None,
token_type_ids=None,
position_ids=None,
head_mask=None,
inputs_embeds=None,
output_attentions=None,
output_hidden_states=None,
labels=None,
training=False,
):
r"""
labels (:obj:`tf.Tensor` of shape :obj:`(batch_size,)`, `optional`, defaults to :obj:`None`):
Labels for computing the sequence classification/regression loss.
Indices should be in :obj:`[0, ..., config.num_labels - 1]`.
If :obj:`config.num_labels == 1` a regression loss is computed (Mean-Square loss),
If :obj:`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
Return:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.XxxConfig`) and inputs:
logits (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, config.num_labels)`):
Classification (or regression if config.num_labels==1) scores (before SoftMax).
hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``output_hidden_states=True`` is passed or when ``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 ``output_attentions=True`` is passed or 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.
"""
if isinstance(inputs, (tuple, list)):
labels = inputs[8] if len(inputs) > 8 else labels
if len(inputs) > 8:
inputs = inputs[:8]
elif isinstance(inputs, (dict, BatchEncoding)):
labels = inputs.pop("labels", labels)
outputs = self.transformer(
inputs,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
training=training,
)
pooled_output = outputs[1]
pooled_output = self.dropout(pooled_output, training=kwargs.get("training", False))
pooled_output = self.dropout(pooled_output, training=training)
logits = self.classifier(pooled_output)
outputs = (logits,) + outputs[2:] # add hidden states and attention if they are here
return outputs # logits, (hidden_states), (attentions)
if labels is not None:
loss = self.compute_loss(labels, logits)
outputs = (loss,) + outputs
return outputs # (loss), logits, (hidden_states), (attentions)
@add_start_docstrings(
"""Xxx Model with a token classification head on top (a linear layer on top of
"""XXX Model with a multiple choice classification head on top (a linear layer on top of
the pooled output and a softmax) e.g. for RocStories/SWAG tasks. """,
XXX_START_DOCSTRING,
)
class TFXxxForMultipleChoice(TFXxxPreTrainedModel, TFMultipleChoiceLoss):
def __init__(self, config, *inputs, **kwargs):
super().__init__(config, *inputs, **kwargs)
self.transformer = TFXxxMainLayer(config, name="transformer")
self.dropout = tf.keras.layers.Dropout(config.hidden_dropout_prob)
self.classifier = tf.keras.layers.Dense(
1, kernel_initializer=get_initializer(config.initializer_range), name="classifier"
)
@property
def dummy_inputs(self):
""" Dummy inputs to build the network.
Returns:
tf.Tensor with dummy inputs
"""
return {"input_ids": tf.constant(MULTIPLE_CHOICE_DUMMY_INPUTS)}
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING.format("(batch_size, num_choices, sequence_length)"))
@add_code_sample_docstrings(tokenizer_class=_TOKENIZER_FOR_DOC, checkpoint="xxx-base-cased")
def call(
self,
inputs,
attention_mask=None,
token_type_ids=None,
position_ids=None,
head_mask=None,
inputs_embeds=None,
output_attentions=None,
output_hidden_states=None,
labels=None,
training=False,
):
r"""
labels (:obj:`tf.Tensor` of shape :obj:`(batch_size,)`, `optional`, defaults to :obj:`None`):
Labels for computing the multiple choice classification loss.
Indices should be in ``[0, ..., num_choices]`` where `num_choices` is the size of the second dimension
of the input tensors. (see `input_ids` above)
Return:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.XxxConfig`) and inputs:
classification_scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, num_choices)`:
`num_choices` is the size of the second dimension of the input tensors. (see `input_ids` above).
Classification scores (before SoftMax).
hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``output_hidden_states=True`` is passed or when ``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 ``output_attentions=True`` is passed or 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.
"""
if isinstance(inputs, (tuple, list)):
input_ids = inputs[0]
attention_mask = inputs[1] if len(inputs) > 1 else attention_mask
token_type_ids = inputs[2] if len(inputs) > 2 else token_type_ids
position_ids = inputs[3] if len(inputs) > 3 else position_ids
head_mask = inputs[4] if len(inputs) > 4 else head_mask
inputs_embeds = inputs[5] if len(inputs) > 5 else inputs_embeds
output_attentions = inputs[6] if len(inputs) > 6 else output_attentions
output_hidden_states = inputs[7] if len(inputs) > 7 else output_hidden_states
labels = inputs[8] if len(inputs) > 8 else labels
assert len(inputs) <= 9, "Too many inputs."
elif isinstance(inputs, (dict, BatchEncoding)):
input_ids = inputs.get("input_ids")
attention_mask = inputs.get("attention_mask", attention_mask)
token_type_ids = inputs.get("token_type_ids", token_type_ids)
position_ids = inputs.get("position_ids", position_ids)
head_mask = inputs.get("head_mask", head_mask)
inputs_embeds = inputs.get("inputs_embeds", inputs_embeds)
output_attentions = inputs.get("output_attentions", output_attentions)
output_hidden_states = inputs.get("output_hidden_states", output_hidden_states)
labels = inputs.get("labels", labels)
assert len(inputs) <= 9, "Too many inputs."
else:
input_ids = inputs
if input_ids is not None:
num_choices = shape_list(input_ids)[1]
seq_length = shape_list(input_ids)[2]
else:
num_choices = shape_list(inputs_embeds)[1]
seq_length = shape_list(inputs_embeds)[2]
flat_input_ids = tf.reshape(input_ids, (-1, seq_length)) if input_ids is not None else None
flat_attention_mask = tf.reshape(attention_mask, (-1, seq_length)) if attention_mask is not None else None
flat_token_type_ids = tf.reshape(token_type_ids, (-1, seq_length)) if token_type_ids is not None else None
flat_position_ids = tf.reshape(position_ids, (-1, seq_length)) if position_ids is not None else None
flat_inputs_embeds = (
tf.reshape(inputs_embeds, (-1, seq_length, shape_list(inputs_embeds)[3]))
if inputs_embeds is not None
else None
)
flat_inputs = [
flat_input_ids,
flat_attention_mask,
flat_token_type_ids,
flat_position_ids,
head_mask,
flat_inputs_embeds,
output_attentions,
output_hidden_states,
]
outputs = self.transformer(flat_inputs, training=training)
pooled_output = outputs[1]
pooled_output = self.dropout(pooled_output, training=training)
logits = self.classifier(pooled_output)
reshaped_logits = tf.reshape(logits, (-1, num_choices))
outputs = (reshaped_logits,) + outputs[2:] # add hidden states and attention if they are here
if labels is not None:
loss = self.compute_loss(labels, reshaped_logits)
outputs = (loss,) + outputs
return outputs # (loss), reshaped_logits, (hidden_states), (attentions)
@add_start_docstrings(
"""XXX Model with a token classification head on top (a linear layer on top of
the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. """,
XXX_START_DOCSTRING,
XXX_INPUTS_DOCSTRING,
)
class TFXxxForTokenClassification(TFXxxPreTrainedModel):
r"""
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**scores**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length, config.num_labels)``
Classification scores (before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``Numpy array`` or ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``Numpy array`` or ``tf.Tensor`` (one for each layer) of shape ``(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 XxxTokenizer, TFXxxForTokenClassification
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = TFXxxForTokenClassification.from_pretrained('xxx-base-uncased')
input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :] # Batch size 1
outputs = model(input_ids)
scores = outputs[0]
"""
class TFXxxForTokenClassification(TFXxxPreTrainedModel, TFTokenClassificationLoss):
def __init__(self, config, *inputs, **kwargs):
super().__init__(config, *inputs, **kwargs)
self.num_labels = config.num_labels
@ -460,53 +633,81 @@ class TFXxxForTokenClassification(TFXxxPreTrainedModel):
config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="classifier"
)
def call(self, inputs, **kwargs):
outputs = self.transformer(inputs, **kwargs)
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING)
@add_code_sample_docstrings(tokenizer_class=_TOKENIZER_FOR_DOC, checkpoint="xxx-base-cased")
def call(
self,
inputs=None,
attention_mask=None,
token_type_ids=None,
position_ids=None,
head_mask=None,
inputs_embeds=None,
output_attentions=None,
output_hidden_states=None,
labels=None,
training=False,
):
r"""
labels (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
Labels for computing the token classification loss.
Indices should be in ``[0, ..., config.num_labels - 1]``.
Return:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.XxxConfig`) and inputs:
scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, config.num_labels)`):
Classification scores (before SoftMax).
hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``output_hidden_states=True`` is passed or when ``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 ``output_attentions=True`` is passed or 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.
"""
if isinstance(inputs, (tuple, list)):
labels = inputs[8] if len(inputs) > 8 else labels
if len(inputs) > 8:
inputs = inputs[:8]
elif isinstance(inputs, (dict, BatchEncoding)):
labels = inputs.pop("labels", labels)
outputs = self.transformer(
inputs,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
training=training,
)
sequence_output = outputs[0]
sequence_output = self.dropout(sequence_output, training=kwargs.get("training", False))
sequence_output = self.dropout(sequence_output, training=training)
logits = self.classifier(sequence_output)
outputs = (logits,) + outputs[2:] # add hidden states and attention if they are here
return outputs # scores, (hidden_states), (attentions)
if labels is not None:
loss = self.compute_loss(labels, logits)
outputs = (loss,) + outputs
return outputs # (loss), logits, (hidden_states), (attentions)
@add_start_docstrings(
"""Xxx Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of
"""XXX Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of
the hidden-states output to compute `span start logits` and `span end logits`). """,
XXX_START_DOCSTRING,
XXX_INPUTS_DOCSTRING,
)
class TFXxxForQuestionAnswering(TFXxxPreTrainedModel):
r"""
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**start_scores**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length,)``
Span-start scores (before SoftMax).
**end_scores**: ``Numpy array`` or ``tf.Tensor`` of shape ``(batch_size, sequence_length,)``
Span-end scores (before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``Numpy array`` or ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``Numpy array`` or ``tf.Tensor`` (one for each layer) of shape ``(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 XxxTokenizer, TFXxxForQuestionAnswering
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = TFXxxForQuestionAnswering.from_pretrained('xxx-base-uncased')
input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :] # Batch size 1
outputs = model(input_ids)
start_scores, end_scores = outputs[:2]
"""
class TFXxxForQuestionAnswering(TFXxxPreTrainedModel, TFQuestionAnsweringLoss):
def __init__(self, config, *inputs, **kwargs):
super().__init__(config, *inputs, **kwargs)
self.num_labels = config.num_labels
@ -516,8 +717,70 @@ class TFXxxForQuestionAnswering(TFXxxPreTrainedModel):
config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="qa_outputs"
)
def call(self, inputs, **kwargs):
outputs = self.transformer(inputs, **kwargs)
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING)
@add_code_sample_docstrings(tokenizer_class=_TOKENIZER_FOR_DOC, checkpoint="xxx-base-cased")
def call(
self,
inputs=None,
attention_mask=None,
token_type_ids=None,
position_ids=None,
head_mask=None,
inputs_embeds=None,
output_attentions=None,
output_hidden_states=None,
start_positions=None,
end_positions=None,
training=False,
):
r"""
start_positions (:obj:`tf.Tensor` of shape :obj:`(batch_size,)`, `optional`, defaults to :obj:`None`):
Labels for position (index) of the start of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (`sequence_length`).
Position outside of the sequence are not taken into account for computing the loss.
end_positions (:obj:`tf.Tensor` of shape :obj:`(batch_size,)`, `optional`, defaults to :obj:`None`):
Labels for position (index) of the end of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (`sequence_length`).
Position outside of the sequence are not taken into account for computing the loss.
Return:
:obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.XxxConfig`) and inputs:
start_scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length,)`):
Span-start scores (before SoftMax).
end_scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length,)`):
Span-end scores (before SoftMax).
hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``output_hidden_states=True`` is passed or when ``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 ``output_attentions=True`` is passed or 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.
"""
if isinstance(inputs, (tuple, list)):
start_positions = inputs[8] if len(inputs) > 8 else start_positions
end_positions = inputs[9] if len(inputs) > 9 else end_positions
if len(inputs) > 8:
inputs = inputs[:8]
elif isinstance(inputs, (dict, BatchEncoding)):
start_positions = inputs.pop("start_positions", start_positions)
end_positions = inputs.pop("end_positions", start_positions)
outputs = self.transformer(
inputs,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
training=training,
)
sequence_output = outputs[0]
@ -528,4 +791,10 @@ class TFXxxForQuestionAnswering(TFXxxPreTrainedModel):
outputs = (start_logits, end_logits,) + outputs[2:]
return outputs # start_logits, end_logits, (hidden_states), (attentions)
if start_positions is not None and end_positions is not None:
labels = {"start_position": start_positions}
labels["end_position"] = end_positions
loss = self.compute_loss(labels, outputs[:2])
outputs = (loss,) + outputs
return outputs # (loss), start_logits, end_logits, (hidden_states), (attentions)

522
templates/adding_a_new_model/modeling_xxx.py
Просмотреть файл

@ -27,12 +27,23 @@ from torch import nn
from torch.nn import CrossEntropyLoss, MSELoss
from .configuration_xxx import XxxConfig
from .file_utils import add_start_docstrings
from .file_utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_callable
from .modeling_outputs import (
BaseModelOutputWithPooling,
MaskedLMOutput,
MultipleChoiceModelOutput,
QuestionAnsweringModelOutput,
SequenceClassifierOutput,
TokenClassifierOutput,
)
from .modeling_utils import PreTrainedModel
logger = logging.getLogger(__name__)
_CONFIG_FOR_DOC = "XXXConfig"
_TOKENIZER_FOR_DOC = "XXXTokenizer"
####################################################
# This list contrains shortcut names for some of
# the pretrained weights provided with the models
@ -197,19 +208,12 @@ class XxxPreTrainedModel(PreTrainedModel):
XXX_START_DOCSTRING = r""" The XXX model was proposed in
`XXX: Pre-training of Deep Bidirectional Transformers for Language Understanding`_
by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova. It's a bidirectional transformer
pre-trained using a combination of masked language modeling objective and next sentence prediction
on a large corpus comprising the Toronto Book Corpus and Wikipedia.
`XXX: Pre-training of Deep Bidirectional Transformers for Language Understanding
<https://arxiv.org/abs/1810.04805>`__ by....
This model is a PyTorch `torch.nn.Module`_ sub-class. Use it as a regular PyTorch Module and
refer to the PyTorch documentation for all matter related to general usage and behavior.
.. _`XXX: Pre-training of Deep Bidirectional Transformers for Language Understanding`:
https://arxiv.org/abs/1810.04805
.. _`torch.nn.Module`:
https://pytorch.org/docs/stable/nn.html#module
This model is a PyTorch `torch.nn.Module <https://pytorch.org/docs/stable/nn.html#torch.nn.Module>`_ sub-class.
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general
usage and behavior.
Parameters:
config (:class:`~transformers.XxxConfig`): Model configuration class with all the parameters of the model.
@ -219,86 +223,53 @@ XXX_START_DOCSTRING = r""" The XXX model was proposed in
XXX_INPUTS_DOCSTRING = r"""
Inputs:
**input_ids**: ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
Indices of input sequence tokens in the vocabulary.
To match pre-training, XXX input sequence should be formatted with [CLS] and [SEP] tokens as follows:
(a) For sequence pairs:
``tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]``
``token_type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1``
(b) For single sequences:
``tokens: [CLS] the dog is hairy . [SEP]``
``token_type_ids: 0 0 0 0 0 0 0``
Xxx is a model with absolute position embeddings so it's usually advised to pad the inputs on
the right rather than the left.
Indices can be obtained using :class:`transformers.XxxTokenizer`.
See :func:`transformers.PreTrainedTokenizer.encode` and
:func:`transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.
**attention_mask**: (`optional`) ``torch.FloatTensor`` of shape ``(batch_size, sequence_length)``:
:func:`transformers.PreTrainedTokenizer.__call__` for details.
`What are input IDs? <../glossary.html#input-ids>`__
attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `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.
**token_type_ids**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
`What are attention masks? <../glossary.html#attention-mask>`__
token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `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
(see `XXX: Pre-training of Deep Bidirectional Transformers for Language Understanding`_ for more details).
**position_ids**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
`What are token type IDs? <../glossary.html#token-type-ids>`_
position_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `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]``.
**head_mask**: (`optional`) ``torch.FloatTensor`` of shape ``(num_heads,)`` or ``(num_layers, num_heads)``:
`What are position IDs? <../glossary.html#position-ids>`_
head_mask (:obj:`torch.FloatTensor` 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**.
**inputs_embeds**: (`optional`) ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, embedding_dim)``:
Optionally, instead of passing ``input_ids`` you can choose to directly pass an embedded representation.
:obj:`1` indicates the head is **not masked**, :obj:`0` indicates the head is **masked**.
inputs_embeds (:obj:`torch.FloatTensor` 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.
output_attentions (:obj:`bool`, `optional`, defaults to :obj:`None`):
If set to ``True``, the attentions tensors of all attention layers are returned. See ``attentions`` under returned tensors for more detail.
output_hidden_states (:obj:`bool`, `optional`, defaults to :obj:`None`):
If set to ``True``, the hidden states of all layers are returned. See ``hidden_states`` under returned tensors for more detail.
return_tuple (:obj:`bool`, `optional`, defaults to :obj:`None`):
If set to ``True``, the output of the model will be a plain tuple instead of a ``dataclass``.
"""
@add_start_docstrings(
"The bare Xxx Model transformer outputting raw hidden-states without any specific head on top.",
"The bare XXX Model transformer outputting raw hidden-states without any specific head on top.",
XXX_START_DOCSTRING,
XXX_INPUTS_DOCSTRING,
)
class XxxModel(XxxPreTrainedModel):
r"""
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**last_hidden_state**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, hidden_size)``
Sequence of hidden-states at the output of the last layer of the model.
**pooler_output**: ``torch.FloatTensor`` of shape ``(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 Xxx 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**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``torch.FloatTensor`` (one for each layer) of shape ``(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::
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = XxxModel.from_pretrained('xxx-base-uncased')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
outputs = model(input_ids)
last_hidden_states = outputs[0] # The last hidden-state is the first element of the output tuple
"""
def __init__(self, config):
super().__init__(config)
@ -322,6 +293,13 @@ class XxxModel(XxxPreTrainedModel):
for layer, heads in heads_to_prune.items():
self.encoder.layer[layer].attention.prune_heads(heads)
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
@add_code_sample_docstrings(
tokenizer_class=_TOKENIZER_FOR_DOC,
checkpoint="xxx-base-uncased",
output_type=BaseModelOutputWithPooling,
config_class=_CONFIG_FOR_DOC,
)
def forward(
self,
input_ids=None,
@ -330,7 +308,16 @@ class XxxModel(XxxPreTrainedModel):
position_ids=None,
head_mask=None,
inputs_embeds=None,
output_attentions=None,
output_hidden_states=None,
return_tuple=None,
):
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
)
return_tuple = return_tuple if return_tuple is not None else self.config.use_return_tuple
if input_ids is not None and inputs_embeds is not None:
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
elif input_ids is not None:
@ -362,45 +349,21 @@ class XxxModel(XxxPreTrainedModel):
)
encoder_outputs = self.encoder(embedding_output, extended_attention_mask, head_mask=head_mask)
sequence_output = encoder_outputs[0]
outputs = (sequence_output,) + encoder_outputs[1:] # add hidden_states and attentions if they are here
pooled_output = self.pooler(sequence_output)
return outputs # sequence_output, (hidden_states), (attentions)
if return_tuple:
return (sequence_output, pooled_output) + encoder_outputs[1:]
return BaseModelOutputWithPooling(
last_hidden_state=sequence_output,
pooler_output=pooled_output,
hidden_states=encoder_outputs.hidden_states,
attentions=encoder_outputs.attentions,
)
@add_start_docstrings(
"""Xxx Model with a `language modeling` head on top. """, XXX_START_DOCSTRING, XXX_INPUTS_DOCSTRING
)
@add_start_docstrings("""XXX Model with a `language modeling` head on top. """, XXX_START_DOCSTRING)
class XxxForMaskedLM(XxxPreTrainedModel):
r"""
**masked_lm_labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
Labels for computing the masked language modeling loss.
Indices should be in ``[-1, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring)
Tokens with indices set to ``-100`` are ignored (masked), the loss is only computed for the tokens with labels
in ``[0, ..., config.vocab_size]``
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**loss**: (`optional`, returned when ``masked_lm_labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
Masked language modeling loss.
**prediction_scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, config.vocab_size)``
Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``torch.FloatTensor`` (one for each layer) of shape ``(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::
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = XxxForMaskedLM.from_pretrained('xxx-base-uncased')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
outputs = model(input_ids, masked_lm_labels=input_ids)
loss, prediction_scores = outputs[:2]
"""
def __init__(self, config):
super().__init__(config)
@ -412,6 +375,13 @@ class XxxForMaskedLM(XxxPreTrainedModel):
def get_output_embeddings(self):
return self.lm_head
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
@add_code_sample_docstrings(
tokenizer_class=_TOKENIZER_FOR_DOC,
checkpoint="xxx-base-uncased",
output_type=MaskedLMOutput,
config_class=_CONFIG_FOR_DOC,
)
def forward(
self,
input_ids=None,
@ -420,8 +390,19 @@ class XxxForMaskedLM(XxxPreTrainedModel):
position_ids=None,
head_mask=None,
inputs_embeds=None,
masked_lm_labels=None,
labels=None,
output_attentions=None,
output_hidden_states=None,
return_tuple=None,
):
r"""
labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
Labels for computing the masked language modeling loss.
Indices should be in ``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring)
Tokens with indices set to ``-100`` are ignored (masked), the loss is only computed for the tokens with labels
in ``[0, ..., config.vocab_size]``
"""
return_tuple = return_tuple if return_tuple is not None else self.config.use_return_tuple
outputs = self.transformer(
input_ids,
@ -430,58 +411,37 @@ class XxxForMaskedLM(XxxPreTrainedModel):
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_tuple=return_tuple,
)
sequence_output = outputs[0]
prediction_scores = self.cls(sequence_output)
outputs = (prediction_scores,) + outputs[2:] # Add hidden states and attention if they are here
if masked_lm_labels is not None:
loss_fct = CrossEntropyLoss()
masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), masked_lm_labels.view(-1))
outputs = (masked_lm_loss,) + outputs
masked_lm_loss = None
if labels is not None:
loss_fct = CrossEntropyLoss() # -100 index = padding token
masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
return outputs # (masked_lm_loss), prediction_scores, (hidden_states), (attentions)
if return_tuple:
output = (prediction_scores,) + outputs[2:]
return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output
return MaskedLMOutput(
loss=masked_lm_loss,
logits=prediction_scores,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
@add_start_docstrings(
"""Xxx Model transformer with a sequence classification/regression head on top (a linear layer on top of
"""XXX Model transformer with a sequence classification/regression head on top (a linear layer on top of
the pooled output) e.g. for GLUE tasks. """,
XXX_START_DOCSTRING,
XXX_INPUTS_DOCSTRING,
)
class XxxForSequenceClassification(XxxPreTrainedModel):
r"""
**labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size,)``:
Labels for computing the sequence classification/regression loss.
Indices should be in ``[0, ..., config.num_labels - 1]``.
If ``config.num_labels == 1`` a regression loss is computed (Mean-Square loss),
If ``config.num_labels > 1`` a classification loss is computed (Cross-Entropy).
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**loss**: (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
Classification (or regression if config.num_labels==1) loss.
**logits**: ``torch.FloatTensor`` of shape ``(batch_size, config.num_labels)``
Classification (or regression if config.num_labels==1) scores (before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``torch.FloatTensor`` (one for each layer) of shape ``(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::
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = XxxForSequenceClassification.from_pretrained('xxx-base-uncased')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
labels = torch.tensor([1]).unsqueeze(0) # Batch size 1
outputs = model(input_ids, labels=labels)
loss, logits = outputs[:2]
"""
def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
@ -492,6 +452,13 @@ class XxxForSequenceClassification(XxxPreTrainedModel):
self.init_weights()
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
@add_code_sample_docstrings(
tokenizer_class=_TOKENIZER_FOR_DOC,
checkpoint="xxx-base-uncased",
output_type=SequenceClassifierOutput,
config_class=_CONFIG_FOR_DOC,
)
def forward(
self,
input_ids=None,
@ -501,7 +468,18 @@ class XxxForSequenceClassification(XxxPreTrainedModel):
head_mask=None,
inputs_embeds=None,
labels=None,
output_attentions=None,
output_hidden_states=None,
return_tuple=None,
):
r"""
labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`, defaults to :obj:`None`):
Labels for computing the sequence classification/regression loss.
Indices should be in :obj:`[0, ..., config.num_labels - 1]`.
If :obj:`config.num_labels == 1` a regression loss is computed (Mean-Square loss),
If :obj:`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
"""
return_tuple = return_tuple if return_tuple is not None else self.config.use_return_tuple
outputs = self.transformer(
input_ids,
@ -510,6 +488,9 @@ class XxxForSequenceClassification(XxxPreTrainedModel):
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_tuple=return_tuple,
)
pooled_output = outputs[1]
@ -517,8 +498,7 @@ class XxxForSequenceClassification(XxxPreTrainedModel):
pooled_output = self.dropout(pooled_output)
logits = self.classifier(pooled_output)
outputs = (logits,) + outputs[2:] # add hidden states and attention if they are here
loss = None
if labels is not None:
if self.num_labels == 1:
# We are doing regression
@ -527,47 +507,108 @@ class XxxForSequenceClassification(XxxPreTrainedModel):
else:
loss_fct = CrossEntropyLoss()
loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
outputs = (loss,) + outputs
return outputs # (loss), logits, (hidden_states), (attentions)
if return_tuple:
output = (logits,) + outputs[2:]
return ((loss,) + output) if loss is not None else output
return SequenceClassifierOutput(
loss=loss, logits=logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions,
)
@add_start_docstrings(
"""Xxx Model with a token classification head on top (a linear layer on top of
"""XXX Model with a multiple choice classification head on top (a linear layer on top of
the pooled output and a softmax) e.g. for RocStories/SWAG tasks. """,
XXX_START_DOCSTRING,
)
class XxxForMultipleChoice(XxxPreTrainedModel):
def __init__(self, config):
super().__init__(config)
self.transformer = XxxModel(config)
self.dropout = nn.Dropout(config.hidden_dropout_prob)
self.classifier = nn.Linear(config.hidden_size, 1)
self.init_weights()
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING.format("(batch_size, num_choices, sequence_length)"))
@add_code_sample_docstrings(
tokenizer_class=_TOKENIZER_FOR_DOC,
checkpoint="xxx-base-uncased",
output_type=MultipleChoiceModelOutput,
config_class=_CONFIG_FOR_DOC,
)
def forward(
self,
input_ids=None,
attention_mask=None,
token_type_ids=None,
position_ids=None,
head_mask=None,
inputs_embeds=None,
labels=None,
output_attentions=None,
output_hidden_states=None,
return_tuple=None,
):
r"""
labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`, defaults to :obj:`None`):
Labels for computing the multiple choice classification loss.
Indices should be in ``[0, ..., num_choices-1]`` where `num_choices` is the size of the second dimension
of the input tensors. (see `input_ids` above)
"""
return_tuple = return_tuple if return_tuple is not None else self.config.use_return_tuple
num_choices = input_ids.shape[1] if input_ids is not None else inputs_embeds.shape[1]
input_ids = input_ids.view(-1, input_ids.size(-1)) if input_ids is not None else None
attention_mask = attention_mask.view(-1, attention_mask.size(-1)) if attention_mask is not None else None
token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1)) if token_type_ids is not None else None
position_ids = position_ids.view(-1, position_ids.size(-1)) if position_ids is not None else None
inputs_embeds = (
inputs_embeds.view(-1, inputs_embeds.size(-2), inputs_embeds.size(-1))
if inputs_embeds is not None
else None
)
outputs = self.transformer(
input_ids,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_tuple=return_tuple,
)
pooled_output = outputs[1]
pooled_output = self.dropout(pooled_output)
logits = self.classifier(pooled_output)
reshaped_logits = logits.view(-1, num_choices)
loss = None
if labels is not None:
loss_fct = CrossEntropyLoss()
loss = loss_fct(reshaped_logits, labels)
if return_tuple:
output = (reshaped_logits,) + outputs[2:]
return ((loss,) + output) if loss is not None else output
return MultipleChoiceModelOutput(
loss=loss, logits=reshaped_logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions,
)
@add_start_docstrings(
"""XXX Model with a token classification head on top (a linear layer on top of
the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. """,
XXX_START_DOCSTRING,
XXX_INPUTS_DOCSTRING,
)
class XxxForTokenClassification(XxxPreTrainedModel):
r"""
**labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
Labels for computing the token classification loss.
Indices should be in ``[0, ..., config.num_labels - 1]``.
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**loss**: (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
Classification loss.
**scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, config.num_labels)``
Classification scores (before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``torch.FloatTensor`` (one for each layer) of shape ``(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::
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = XxxForTokenClassification.from_pretrained('xxx-base-uncased')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
labels = torch.tensor([1] * input_ids.size(1)).unsqueeze(0) # Batch size 1
outputs = model(input_ids, labels=labels)
loss, scores = outputs[:2]
"""
def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
@ -578,6 +619,13 @@ class XxxForTokenClassification(XxxPreTrainedModel):
self.init_weights()
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
@add_code_sample_docstrings(
tokenizer_class=_TOKENIZER_FOR_DOC,
checkpoint="xxx-base-uncased",
output_type=TokenClassifierOutput,
config_class=_CONFIG_FOR_DOC,
)
def forward(
self,
input_ids=None,
@ -587,7 +635,16 @@ class XxxForTokenClassification(XxxPreTrainedModel):
head_mask=None,
inputs_embeds=None,
labels=None,
output_attentions=None,
output_hidden_states=None,
return_tuple=None,
):
r"""
labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
Labels for computing the token classification loss.
Indices should be in ``[0, ..., config.num_labels - 1]``.
"""
return_tuple = return_tuple if return_tuple is not None else self.config.use_return_tuple
outputs = self.transformer(
input_ids,
@ -596,6 +653,9 @@ class XxxForTokenClassification(XxxPreTrainedModel):
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_tuple=return_tuple,
)
sequence_output = outputs[0]
@ -603,70 +663,35 @@ class XxxForTokenClassification(XxxPreTrainedModel):
sequence_output = self.dropout(sequence_output)
logits = self.classifier(sequence_output)
outputs = (logits,) + outputs[2:] # add hidden states and attention if they are here
loss = None
if labels is not None:
loss_fct = CrossEntropyLoss()
# Only keep active parts of the loss
if attention_mask is not None:
active_loss = attention_mask.view(-1) == 1
active_logits = logits.view(-1, self.num_labels)[active_loss]
active_labels = labels.view(-1)[active_loss]
active_logits = logits.view(-1, self.num_labels)
active_labels = torch.where(
active_loss, labels.view(-1), torch.tensor(loss_fct.ignore_index).type_as(labels)
)
loss = loss_fct(active_logits, active_labels)
else:
loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
outputs = (loss,) + outputs
return outputs # (loss), scores, (hidden_states), (attentions)
if return_tuple:
output = (logits,) + outputs[2:]
return ((loss,) + output) if loss is not None else output
return TokenClassifierOutput(
loss=loss, logits=logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions,
)
@add_start_docstrings(
"""Xxx Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of
the hidden-states output to compute `span start logits` and `span end logits`). """,
"""XXX Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear
layers on top of the hidden-states output to compute `span start logits` and `span end logits`). """,
XXX_START_DOCSTRING,
XXX_INPUTS_DOCSTRING,
)
class XxxForQuestionAnswering(XxxPreTrainedModel):
r"""
**start_positions**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size,)``:
Labels for position (index) of the start of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (`sequence_length`).
Position outside of the sequence are not taken into account for computing the loss.
**end_positions**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size,)``:
Labels for position (index) of the end of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (`sequence_length`).
Position outside of the sequence are not taken into account for computing the loss.
Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
**loss**: (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.
**start_scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length,)``
Span-start scores (before SoftMax).
**end_scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length,)``
Span-end scores (before SoftMax).
**hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
of shape ``(batch_size, sequence_length, hidden_size)``:
Hidden-states of the model at the output of each layer plus the initial embedding outputs.
**attentions**: (`optional`, returned when ``output_attentions=True``)
list of ``torch.FloatTensor`` (one for each layer) of shape ``(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::
tokenizer = XxxTokenizer.from_pretrained('xxx-base-uncased')
model = XxxForQuestionAnswering.from_pretrained('xxx-large-uncased-whole-word-masking-finetuned-squad')
question, text = "Who was Jim Henson?", "Jim Henson was a nice puppet"
input_text = "[CLS] " + question + " [SEP] " + text + " [SEP]"
input_ids = tokenizer.encode(input_text)
token_type_ids = [0 if i <= input_ids.index(102) else 1 for i in range(len(input_ids))]
start_scores, end_scores = model(torch.tensor([input_ids]), token_type_ids=torch.tensor([token_type_ids]))
all_tokens = tokenizer.convert_ids_to_tokens(input_ids)
print(' '.join(all_tokens[torch.argmax(start_scores) : torch.argmax(end_scores)+1]))
# a nice puppet
"""
def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
@ -676,6 +701,13 @@ class XxxForQuestionAnswering(XxxPreTrainedModel):
self.init_weights()
@add_start_docstrings_to_callable(XXX_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
@add_code_sample_docstrings(
tokenizer_class=_TOKENIZER_FOR_DOC,
checkpoint="xxx-base-uncased",
output_type=QuestionAnsweringModelOutput,
config_class=_CONFIG_FOR_DOC,
)
def forward(
self,
input_ids=None,
@ -686,7 +718,21 @@ class XxxForQuestionAnswering(XxxPreTrainedModel):
inputs_embeds=None,
start_positions=None,
end_positions=None,
output_attentions=None,
output_hidden_states=None,
return_tuple=None,
):
r"""
start_positions (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`, defaults to :obj:`None`):
Labels for position (index) of the start of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (`sequence_length`).
Position outside of the sequence are not taken into account for computing the loss.
end_positions (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`, defaults to :obj:`None`):
Labels for position (index) of the end of the labelled span for computing the token classification loss.
Positions are clamped to the length of the sequence (`sequence_length`).
Position outside of the sequence are not taken into account for computing the loss.
"""
return_tuple = return_tuple if return_tuple is not None else self.config.use_return_tuple
outputs = self.transformer(
input_ids,
@ -695,6 +741,9 @@ class XxxForQuestionAnswering(XxxPreTrainedModel):
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_tuple=return_tuple,
)
sequence_output = outputs[0]
@ -704,7 +753,7 @@ class XxxForQuestionAnswering(XxxPreTrainedModel):
start_logits = start_logits.squeeze(-1)
end_logits = end_logits.squeeze(-1)
outputs = (start_logits, end_logits,) + outputs[2:]
total_loss = None
if start_positions is not None and end_positions is not None:
# If we are on multi-GPU, split add a dimension
if len(start_positions.size()) > 1:
@ -720,6 +769,15 @@ class XxxForQuestionAnswering(XxxPreTrainedModel):
start_loss = loss_fct(start_logits, start_positions)
end_loss = loss_fct(end_logits, end_positions)
total_loss = (start_loss + end_loss) / 2
outputs = (total_loss,) + outputs
return outputs # (loss), start_logits, end_logits, (hidden_states), (attentions)
if return_tuple:
output = (start_logits, end_logits) + outputs[2:]
return ((total_loss,) + output) if total_loss is not None else output
return QuestionAnsweringModelOutput(
loss=total_loss,
start_logits=start_logits,
end_logits=end_logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)

137
templates/adding_a_new_model/tokenization_xxx.py
Просмотреть файл

@ -18,6 +18,7 @@
import collections
import logging
import os
from typing import List, Optional
from .tokenization_utils import PreTrainedTokenizer
@ -77,12 +78,37 @@ def load_vocab(vocab_file):
class XxxTokenizer(PreTrainedTokenizer):
r"""
Constructs a XxxTokenizer.
:class:`~transformers.XxxTokenizer` runs end-to-end tokenization: punctuation splitting + wordpiece
Constructs a XXX tokenizer. Based on XXX.
This tokenizer inherits from :class:`~transformers.PreTrainedTokenizer` which contains most of the methods. Users
should refer to the superclass for more information regarding methods.
Args:
vocab_file: Path to a one-wordpiece-per-line vocabulary file
do_lower_case: Whether to lower case the input. Only has an effect when do_basic_tokenize=True
vocab_file (:obj:`str`):
File containing the vocabulary.
do_lower_case (:obj:`bool`, `optional`, defaults to :obj:`True`):
Whether to lowercase the input when tokenizing.
do_basic_tokenize (:obj:`bool`, `optional`, defaults to :obj:`True`):
Whether to do basic tokenization before WordPiece.
never_split (:obj:`Iterable`, `optional`, defaults to :obj:`None`):
Collection of tokens which will never be split during tokenization. Only has an effect when
:obj:`do_basic_tokenize=True`
unk_token (:obj:`str`, `optional`, defaults to :obj:`"[UNK]"`):
The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
token instead.
sep_token (:obj:`str`, `optional`, defaults to :obj:`"[SEP]"`):
The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences
for sequence classification or for a text and a question for question answering.
It is also used as the last token of a sequence built with special tokens.
pad_token (:obj:`str`, `optional`, defaults to :obj:`"[PAD]"`):
The token used for padding, for example when batching sequences of different lengths.
cls_token (:obj:`str`, `optional`, defaults to :obj:`"[CLS]"`):
The classifier token which is used when doing sequence classification (classification of the whole
sequence instead of per-token classification). It is the first token of the sequence when built with
special tokens.
mask_token (:obj:`str`, `optional`, defaults to :obj:`"[MASK]"`):
The token used for masking values. This is the token used when training this model with masked language
modeling. This is the token which the model will try to predict.
"""
vocab_files_names = VOCAB_FILES_NAMES
@ -94,21 +120,16 @@ class XxxTokenizer(PreTrainedTokenizer):
self,
vocab_file,
do_lower_case=True,
do_basic_tokenize=True,
never_split=None,
unk_token="[UNK]",
sep_token="[SEP]",
pad_token="[PAD]",
cls_token="[CLS]",
mask_token="[MASK]",
tokenize_chinese_chars=True,
**kwargs
):
"""Constructs a XxxTokenizer.
Args:
**vocab_file**: Path to a one-wordpiece-per-line vocabulary file
**do_lower_case**: (`optional`) boolean (default True)
Whether to lower case the input
Only has an effect when do_basic_tokenize=True
"""
super().__init__(
unk_token=unk_token,
sep_token=sep_token,
@ -121,22 +142,35 @@ class XxxTokenizer(PreTrainedTokenizer):
if not os.path.isfile(vocab_file):
raise ValueError(
"Can't find a vocabulary file at path '{}'. To load the vocabulary from a Google pretrained "
"model use `tokenizer = XxxTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`".format(vocab_file)
"model use `tokenizer = BertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`".format(vocab_file)
)
self.vocab = load_vocab(vocab_file)
self.ids_to_tokens = collections.OrderedDict([(ids, tok) for tok, ids in self.vocab.items()])
self.do_basic_tokenize = do_basic_tokenize
# Replace and adapt
# if do_basic_tokenize:
# self.basic_tokenizer = BasicTokenizer(
# do_lower_case=do_lower_case, never_split=never_split, tokenize_chinese_chars=tokenize_chinese_chars
# )
# self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab, unk_token=self.unk_token)
@property
def vocab_size(self):
return len(self.vocab)
def get_vocab(self):
return dict(self.vocab, **self.added_tokens_encoder)
def _tokenize(self, text):
""" Take as input a string and return a list of strings (tokens) for words/sub-words
"""
split_tokens = []
if self.do_basic_tokenize:
for token in self.basic_tokenizer.tokenize(text, never_split=self.all_special_tokens):
for sub_token in self.wordpiece_tokenizer.tokenize(token):
split_tokens.append(sub_token)
# If the token is part of the never_split set
if token in self.basic_tokenizer.never_split:
split_tokens.append(token)
else:
split_tokens += self.wordpiece_tokenizer.tokenize(token)
else:
split_tokens = self.wordpiece_tokenizer.tokenize(text)
return split_tokens
@ -154,13 +188,25 @@ class XxxTokenizer(PreTrainedTokenizer):
out_string = " ".join(tokens).replace(" ##", "").strip()
return out_string
def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
def build_inputs_with_special_tokens(
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
) -> List[int]:
"""
Build model inputs from a sequence or a pair of sequence for sequence classification tasks
by concatenating and adding special tokens.
A BERT sequence has the following format:
single sequence: [CLS] X [SEP]
pair of sequences: [CLS] A [SEP] B [SEP]
- single sequence: ``[CLS] X [SEP]``
- pair of sequences: ``[CLS] A [SEP] B [SEP]``
Args:
token_ids_0 (:obj:`List[int]`):
List of IDs to which the special tokens will be added
token_ids_1 (:obj:`List[int]`, `optional`, defaults to :obj:`None`):
Optional second list of IDs for sequence pairs.
Returns:
:obj:`List[int]`: list of `input IDs <../glossary.html#input-ids>`__ with the appropriate special tokens.
"""
if token_ids_1 is None:
return [self.cls_token_id] + token_ids_0 + [self.sep_token_id]
@ -168,20 +214,23 @@ class XxxTokenizer(PreTrainedTokenizer):
sep = [self.sep_token_id]
return cls + token_ids_0 + sep + token_ids_1 + sep
def get_special_tokens_mask(self, token_ids_0, token_ids_1=None, already_has_special_tokens=False):
def get_special_tokens_mask(
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
) -> List[int]:
"""
Retrieves sequence ids from a token list that has no special tokens added. This method is called when adding
special tokens using the tokenizer ``prepare_for_model`` methods.
special tokens using the tokenizer ``prepare_for_model`` method.
Args:
token_ids_0: list of ids (must not contain special tokens)
token_ids_1: Optional list of ids (must not contain special tokens), necessary when fetching sequence ids
for sequence pairs
already_has_special_tokens: (default False) Set to True if the token list is already formated with
special tokens for the model
token_ids_0 (:obj:`List[int]`):
List of ids.
token_ids_1 (:obj:`List[int]`, `optional`, defaults to :obj:`None`):
Optional second list of IDs for sequence pairs.
already_has_special_tokens (:obj:`bool`, `optional`, defaults to :obj:`False`):
Set to True if the token list is already formatted with special tokens for the model
Returns:
A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
:obj:`List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
"""
if already_has_special_tokens:
@ -196,14 +245,29 @@ class XxxTokenizer(PreTrainedTokenizer):
return [1] + ([0] * len(token_ids_0)) + [1] + ([0] * len(token_ids_1)) + [1]
return [1] + ([0] * len(token_ids_0)) + [1]
def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
def create_token_type_ids_from_sequences(
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
) -> List[int]:
"""
Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
A BERT sequence pair mask has the following format:
0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1
| first sequence | second sequence
::
0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
| first sequence | second sequence |
if token_ids_1 is None, only returns the first portion of the mask (0's).
Args:
token_ids_0 (:obj:`List[int]`):
List of ids.
token_ids_1 (:obj:`List[int]`, `optional`, defaults to :obj:`None`):
Optional second list of IDs for sequence pairs.
Returns:
:obj:`List[int]`: List of `token type IDs <../glossary.html#token-type-ids>`_ according to the given
sequence(s).
"""
sep = [self.sep_token_id]
cls = [self.cls_token_id]
@ -212,7 +276,16 @@ class XxxTokenizer(PreTrainedTokenizer):
return len(cls + token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1]
def save_vocabulary(self, vocab_path):
"""Save the tokenizer vocabulary to a directory or file."""
"""
Save the sentencepiece vocabulary (copy original file) and special tokens file to a directory.
Args:
vocab_path (:obj:`str`):
The directory in which to save the vocabulary.
Returns:
:obj:`Tuple(str)`: Paths to the files saved.
"""
index = 0
if os.path.isdir(vocab_path):
vocab_file = os.path.join(vocab_path, VOCAB_FILES_NAMES["vocab_file"])