We provide a base class, ``PretrainedConfig``, which can load a pretrained instance either from a local file or directory or from a pretrained model configuration provided by the library (downloaded from HuggingFace AWS S3 repository).
The base class ``PretrainedConfig`` implements the common methods for loading/saving a configuration either from a local file or directory, or from a pretrained model configuration provided by the library (downloaded from HuggingFace's AWS S3 repository).
The base class ``PreTrainedModel`` implements the common methods for loading/saving a model either from a local file or directory, or from a pretrained model configuration provided by the library (downloaded from HuggingFace's AWS S3 repository).
``PreTrainedModel`` also implements a few methods which are common among all the models to:
- resize the input token embeddings when new tokens are added to the vocabulary
The base class ``PreTrainedTokenizer`` implements the common methods for loading/saving a tokenizer either from a local file or directory, or from a pretrained tokenizer provided by the library (downloaded from HuggingFace's AWS S3 repository).
``PreTrainedTokenizer`` is the main entry point into tokenizers as it also implements the main methods for using all the tokenizers:
- tokenizing, converting tokens to ids and back and encoding/decoding,
- adding new tokens to the vocabulary in a way that is independant of the underlying structure (BPE, SentencePiece...),
- managing special tokens (adding them, assigning them to roles, making sure they are not split during tokenization)
- API of the configuration classes for BERT, GPT, GPT-2 and Transformer-XL
TODO Lysandre filled: Removed Models/Tokenizers/Optimizers as no single link can be made.
Configurations
^^^^^^^^^^^^^^
Models (BERT, GPT, GPT-2 and Transformer-XL) are defined and build from configuration classes which contains the
parameters of the models (number of layers, dimensionalities...) and a few utilities to read and write from JSON
configuration files. The respective configuration classes are:
* ``BertConfig`` for ``BertModel`` and BERT classes instances.
* ``OpenAIGPTConfig`` for ``OpenAIGPTModel`` and OpenAI GPT classes instances.
* ``GPT2Config`` for ``GPT2Model`` and OpenAI GPT-2 classes instances.
* ``TransfoXLConfig`` for ``TransfoXLModel`` and Transformer-XL classes instances.
These configuration classes contains a few utilities to load and save configurations:
* ``from_dict(cls, json_object)``\ : A class method to construct a configuration from a Python dictionary of parameters. Returns an instance of the configuration class.
* ``from_json_file(cls, json_file)``\ : A class method to construct a configuration from a json file of parameters. Returns an instance of the configuration class.
* ``to_dict()``\ : Serializes an instance to a Python dictionary. Returns a dictionary.
* ``to_json_string()``\ : Serializes an instance to a JSON string. Returns a string.
* ``to_json_file(json_file_path)``\ : Save an instance to a json file.
Loading Google AI or OpenAI pre-trained weights or PyTorch dump
To load one of Google AI's,OpenAI's pre-trained models or a PyTorch saved model (an instance of ``BertForPreTraining`` saved with ``torch.save()``\ ), the PyTorch model classes and the tokenizer can be instantiated using the ``from_pretrained()`` method:
.. code-block:: python
model = BERT_CLASS.from_pretrained(PRE_TRAINED_MODEL_NAME_OR_PATH, cache_dir=None, from_tf=False, state_dict=None, *input, **kwargs)
where
* ``BERT_CLASS`` is either a tokenizer to load the vocabulary (\ ``BertTokenizer`` or ``OpenAIGPTTokenizer`` classes) or one of the eight BERT or three OpenAI GPT PyTorch model classes (to load the pre-trained weights): ``BertModel``\ , ``BertForMaskedLM``\ , ``BertForNextSentencePrediction``\ , ``BertForPreTraining``\ , ``BertForSequenceClassification``\ , ``BertForTokenClassification``\ , ``BertForMultipleChoice``\ , ``BertForQuestionAnswering``\ , ``OpenAIGPTModel``\ , ``OpenAIGPTLMHeadModel`` or ``OpenAIGPTDoubleHeadsModel``\ , and
*
``PRE_TRAINED_MODEL_NAME_OR_PATH`` is either:
*
the shortcut name of a Google AI'sorOpenAI's pre-trained model selected in the list:
* ``bert-base-chinese``: Chinese Simplified and Traditional, 12-layer, 768-hidden, 12-heads, 110M parameters
* ``bert-base-german-cased``: Trained on German data only, 12-layer, 768-hidden, 12-heads, 110M parameters `Performance Evaluation <https://deepset.ai/german-bert>`__
* ``bert-large-uncased-whole-word-masking``: 24-layer, 1024-hidden, 16-heads, 340M parameters - Trained with Whole Word Masking (mask all of the the tokens corresponding to a word at once)
* ``bert-large-cased-whole-word-masking``: 24-layer, 1024-hidden, 16-heads, 340M parameters - Trained with Whole Word Masking (mask all of the the tokens corresponding to a word at once)
* ``bert-large-uncased-whole-word-masking-finetuned-squad``: The ``bert-large-uncased-whole-word-masking`` model finetuned on SQuAD (using the ``run_bert_squad.py`` examples). Results: *exact_match: 86.91579943235573, f1: 93.1532499015869*
* ``transfo-xl-wt103``: Transformer-XL English model trained on wikitext-103, 18-layer, 1024-hidden, 16-heads, 257M parameters
*
a path or url to a pretrained model archive containing:
* ``bert_config.json`` or ``openai_gpt_config.json`` a configuration file for the model, and
* ``pytorch_model.bin`` a PyTorch dump of a pre-trained instance of ``BertForPreTraining``\ , ``OpenAIGPTModel``\ , ``TransfoXLModel``\ , ``GPT2LMHeadModel`` (saved with the usual ``torch.save()``\ )
If ``PRE_TRAINED_MODEL_NAME_OR_PATH`` is a shortcut name, the pre-trained weights will be downloaded from AWS S3 (see the links `here <https://github.com/huggingface/pytorch-transformers/blob/master/pytorch_transformers/modeling_bert.py>`__\ ) and stored in a cache folder to avoid future download (the cache folder can be found at ``~/.pytorch_pretrained_bert/``\ ).
*
``cache_dir`` can be an optional path to a specific directory to download and cache the pre-trained model weights. This option is useful in particular when you are using distributed training: to avoid concurrent access to the same weights you can set for example ``cache_dir='./pretrained_model_{}'.format(args.local_rank)`` (see the section on distributed training for more information).
* ``from_tf``\ : should we load the weights from a locally saved TensorFlow checkpoint
* ``state_dict``\ : an optional state dictionary (collections.OrderedDict object) to use instead of Google pre-trained models
* ``*inputs``\ , `**kwargs`: additional input for the specific Bert class (ex: num_labels for BertForSequenceClassification)
``Uncased`` means that the text has been lowercased before WordPiece tokenization, e.g., ``John Smith`` becomes ``john smith``. The Uncased model also strips out any accent markers. ``Cased`` means that the true case and accent markers are preserved. Typically, the Uncased model is better unless you know that case information is important for your task (e.g., Named Entity Recognition or Part-of-Speech tagging). For information about the Multilingual and Chinese model, see the `Multilingual README <https://github.com/google-research/bert/blob/master/multilingual.md>`__ or the original TensorFlow repository.
When using an ``uncased model``\ , make sure to pass ``--do_lower_case`` to the example training scripts (or pass ``do_lower_case=True`` to FullTokenizer if you'reusingyourownscriptandloadingthetokenizeryour-self.).
Usually,ifyoudon't set any specific environment variable, ``pytorch_pretrained_bert`` cache will be at ``~/.cache/torch/pytorch_pretrained_bert/``.
You can alsways safely delete ``pytorch_pretrained_bert`` cache but the pretrained model weights and vocabulary files wil have to be re-downloaded from our S3.
Serialization best-practices
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This section explain how you can save and re-load a fine-tuned model (BERT, GPT, GPT-2 and Transformer-XL).
There are three types of files you need to save to be able to reload a fine-tuned model:
* the model it-self which should be saved following PyTorch serialization `best practices <https://pytorch.org/docs/stable/notes/serialization.html#best-practices>`__\ ,
* the configuration file of the model which is saved as a JSON file, and
* the vocabulary (and the merges for the BPE-based models GPT and GPT-2).
The *default filenames* of these files are as follow:
* the model weights file: ``pytorch_model.bin``\ ,
* the configuration file: ``config.json``\ ,
* the vocabulary file: ``vocab.txt`` for BERT and Transformer-XL, ``vocab.json`` for GPT/GPT-2 (BPE vocabulary),
* for GPT/GPT-2 (BPE vocabulary) the additional merges file: ``merges.txt``.
**If you save a model using these *default filenames*\ , you can then re-load the model and tokenizer using the ``from_pretrained()`` method.**
Here is the recommended way of saving the model, configuration and vocabulary to an ``output_dir`` directory and reloading the model and tokenizer afterwards:
.. code-block:: python
from pytorch_pretrained_bert import WEIGHTS_NAME, CONFIG_NAME
output_dir = "./models/"
# Step 1: Save a model, configuration and vocabulary that you have fine-tuned
# If we have a distributed model, save only the encapsulated model
# (it was wrapped in PyTorch DistributedDataParallel or DataParallel)
model_to_save = model.module if hasattr(model, 'module') else model
# If we save using the predefined names, we can load using `from_pretrained`
@@ -8,13 +8,13 @@ The library was designed with two strong goals in mind:
...
@@ -8,13 +8,13 @@ The library was designed with two strong goals in mind:
- be as easy and fast to use as possible:
- be as easy and fast to use as possible:
- we strongly limited the number of abstractions to learn, in fact there are almost no abstractions, just three standard classes for each model: configuration, models and tokenizer,
- we strongly limited the number of user-facing abstractions to learn, in fact there are almost no abstractions, just three standard classes required to use each model: configuration, models and tokenizer,
-each pretrained model configuration, weights and vocabulary can be downloaded, cached and loaded in the related class in a simple way by using a common `from_pretrained()` instantiation method.
-all of these classes can be initialized in a simple and unified way from pretrained instances by using a common `from_pretrained()` instantiation method which will take care of downloading (if needed), caching and loading the related class from a pretrained instance supplied in the library or your own saved instance.
- this library is NOT a modular toolbox of building blocks for neural nets, to extend/build-upon the library, just use your regular Python/PyTorch modules and inherit from the base classes of the library to reuse functionalities like model loading/saving.
-as a consequence, this library is NOT a modular toolbox of building blocks for neural nets. If you want to extend/build-upon the library, just use regular Python/PyTorch modules and inherit from the base classes of the library to reuse functionalities like model loading/saving.
- provide state-of-the-art models with performances as close as possible to the original models:
- provide state-of-the-art models with performances as close as possible to the original models:
- we provide at least one example for each model which reproduces a result provided by the official authors of said model,
- we provide at least one example for each architecture which reproduces a result provided by the official authors of said architecture,
- the code is usually as close to the original code base as possible which means some PyTorch code may be not as *pytorchic* as it could be as a result of being converted TensorFlow code.
- the code is usually as close to the original code base as possible which means some PyTorch code may be not as *pytorchic* as it could be as a result of being converted TensorFlow code.
A few other goals:
A few other goals:
...
@@ -34,15 +34,18 @@ A few other goals:
...
@@ -34,15 +34,18 @@ A few other goals:
The library is build around three type of classes for each models:
The library is build around three type of classes for each models:
-**model classes** which are PyTorch models (`torch.nn.Modules`) of the 6 models architectures currently provided in the library, e.g. `BertModel`
-**model classes** which are PyTorch models (`torch.nn.Modules`) of the 6 models architectures currently provided in the library, e.g. `BertModel`
-**configuration classes** which store all the parameters required to build a model, e.g. `BertConfig`
-**configuration classes** which store all the parameters required to build a model, e.g. `BertConfig`. You don't always need to instantiate these your-self, in particular if you are using a pretrained model without any modification, creating the model will automatically take care of instantiating the configuration (which is part of the model)
-**tokenizer classes** which store the vocabulary for each model and provide methods for encoding strings in list of token embeddings indices to be fed to a model, e.g. `BertTokenizer`
-**tokenizer classes** which store the vocabulary for each model and provide methods for encoding/decoding strings in list of token embeddings indices to be fed to a model, e.g. `BertTokenizer`
All these classes can be instantiated from pretrained instances and saved locally using two methods:
All these classes can be instantiated from pretrained instances and saved locally using two methods:
-`from_pretrained()` let you instantiate a model/configuration/tokenizer from a pretrained version either provided by the library itself (currently 27 models are provided as listed [here](https://huggingface.co/pytorch-transformers/pretrained_models.html)) or stored locally (or on a server) by the user,
-`from_pretrained()` let you instantiate a model/configuration/tokenizer from a pretrained version either provided by the library itself (currently 27 models are provided as listed [here](https://huggingface.co/pytorch-transformers/pretrained_models.html)) or stored locally (or on a server) by the user,
-`save_pretrained()` let you save a model/configuration/tokenizer locally so that it can be reloaded using `from_pretrained()`.
-`save_pretrained()` let you save a model/configuration/tokenizer locally so that it can be reloaded using `from_pretrained()`.
Let's go through a few simple quick-start examples to see how we can instantiate and use these classes.
We'll finish this quickstart tour by going through a few simple quick-start examples to see how we can instantiate and use these classes. The rest of the documentation is organized in two parts:
- the **MAIN CLASSES** section details the common functionalities/method/attributes of the three main type of classes (configuration, model, tokenizer) plus some optimization related classes provided as utilities for training,
- the **PACKAGE REFERENCE** section details all the variants of each class for each model architectures and in particular the input/output that you should expect when calling each of them.
Handle a few common attributes and methods for loading/downloading/saving configurations.
Handles a few parameters common to all models' configurations as well as methods for loading/downloading/saving configurations.
Class attributes (overridden by derived classes):
- ``pretrained_config_archive_map``: a python ``dict`` of with `short-cut-names` (string) as keys and `url` (string) of associated pretrained model configurations as values.
Parameters:
``finetuning_task``: string, default `None`. Name of the task used to fine-tune the model. This can be used when converting from an original (TensorFlow or PyTorch) checkpoint.
``num_labels``: integer, default `2`. Number of classes to use when the model is a classification model (sequences/tokens)
``output_attentions``: boolean, default `False`. Should the model returns attentions weights.
``output_hidden_states``: string, default `False`. Should the model returns all hidden-states.
``torchscript``: string, default `False`. Is the model used with Torchscript.
"""
"""
pretrained_config_archive_map={}
pretrained_config_archive_map={}
def__init__(self,**kwargs):
def__init__(self,**kwargs):
r""" The initialization of :class:`~pytorch_transformers.PretrainedConfig` extracts
a few configuration attributes from `**kwargs` which are common to all models:
- `finetuning_task`: string, default `None`. Name of the task used to fine-tune the model (used to convert from original checkpoint)
- `num_labels`: integer, default `2`. Number of classes to use when the model is a classification model (sequences/tokens)
- `output_attentions`: boolean, default `False`. Should the model returns attentions weights.
- `output_hidden_states`: string, default `False`. Should the model returns all hidden-states.
- `torchscript`: string, default `False`. Is the model used with Torchscript.
r""" Instantiate a PretrainedConfig from a pre-trained model configuration.
r""" Instantiate a :class:`~pytorch_transformers.PretrainedConfig` (or a derived class) from a pre-trained model configuration.
Parameters:
Parameters:
**pretrained_model_name_or_path**: either:
pretrained_model_name_or_path: either:
- a string with the `shortcut name` of a pre-trained model configuration to load from cache or download, e.g.: ``bert-base-uncased``.
- a string with the `shortcut name` of a pre-trained model configuration to load from cache or download, e.g.: ``bert-base-uncased``.
- a path to a `directory` containing a configuration file saved using the `save_pretrained(save_directory)` method, e.g.: ``./my_model_directory/``.
- a path to a `directory` containing a configuration file saved using the :func:`~pytorch_transformers.PretrainedConfig.save_pretrained` method, e.g.: ``./my_model_directory/``.
- a path or url to a saved configuration JSON `file`, e.g.: ``./my_model_directory/configuration.json``.
- a path or url to a saved configuration JSON `file`, e.g.: ``./my_model_directory/configuration.json``.
**cache_dir**: (`optional`) string:
cache_dir: (`optional`) string:
Path to a directory in which a downloaded pre-trained model
Path to a directory in which a downloaded pre-trained model
configuration should be cached if the standard cache should not be used.
configuration should be cached if the standard cache should not be used.
**return_unused_kwargs**: (`optional`) bool:
kwargs: (`optional`) dict: key/value pairs with which to update the configuration object after loading.
- If False, then this function returns just the final configuration object.
- The values in kwargs of any keys which are configuration attributes will be used to override the loaded values.
- If True, then this functions returns a tuple `(config, unused_kwargs)` where `unused_kwargs` is a dictionary consisting of the key/value pairs whose keys are not configuration attributes: ie the part of kwargs which has not been used to update `config` and is otherwise ignored.
- Behavior concerning key/value pairs whose keys are *not* configuration attributes is controlled by the `return_unused_kwargs` keyword parameter.
**kwargs**: (`optional`) dict:
return_unused_kwargs: (`optional`) bool:
Dictionary of key/value pairs with which to update the configuration object after loading.
- The values in kwargs of any keys which are configuration attributes will be used
- If False, then this function returns just the final configuration object.
to override the loaded values.
- If True, then this functions returns a tuple `(config, unused_kwargs)` where `unused_kwargs` is a dictionary consisting of the key/value pairs whose keys are not configuration attributes: ie the part of kwargs which has not been used to update `config` and is otherwise ignored.
- Behavior concerning key/value pairs whose keys are *not* configuration attributes is controlled
by the `return_unused_kwargs` keyword parameter.
Examples::
Examples::
...
@@ -215,14 +214,26 @@ class PretrainedConfig(object):
...
@@ -215,14 +214,26 @@ class PretrainedConfig(object):
classPreTrainedModel(nn.Module):
classPreTrainedModel(nn.Module):
""" Base class for all models. Handle loading/storing model config and
r""" Base class for all models.
a simple interface for dowloading and loading pretrained models.
:class:`~pytorch_transformers.PreTrainedModel` takes care of storing the configuration of the models and handles methods for loading/downloading/saving models
as well as a few methods commons to all models to (i) resize the input embeddings and (ii) prune heads in the self-attention heads.
Class attributes (overridden by derived classes):
- ``config_class``: a class derived from :class:`~pytorch_transformers.PretrainedConfig` to use as configuration class for this model architecture.
- ``pretrained_model_archive_map``: a python ``dict`` of with `short-cut-names` (string) as keys and `url` (string) of associated pretrained weights as values.
- ``load_tf_weights``: a python ``method`` for loading a TensorFlow checkpoint in a PyTorch model, taking as arguments:
- ``model``: an instance of the relevant subclass of :class:`~pytorch_transformers.PreTrainedModel`,
- ``config``: an instance of the relevant subclass of :class:`~pytorch_transformers.PretrainedConfig`,
- ``path``: a path (string) to the TensorFlow checkpoint.
- ``base_model_prefix``: a string indicating the attribute associated to the base model in derived classes of the same architecture adding modules on top of the base model.
"""
"""
config_class=PretrainedConfig
config_class=None
pretrained_model_archive_map={}
pretrained_model_archive_map={}
load_tf_weights=lambdamodel,config,path:None
load_tf_weights=lambdamodel,config,path:None
base_model_prefix=""
base_model_prefix=""
input_embeddings=None
def__init__(self,config,*inputs,**kwargs):
def__init__(self,config,*inputs,**kwargs):
super(PreTrainedModel,self).__init__()
super(PreTrainedModel,self).__init__()
...
@@ -282,15 +293,14 @@ class PreTrainedModel(nn.Module):
...
@@ -282,15 +293,14 @@ class PreTrainedModel(nn.Module):
""" Resize input token embeddings matrix of the model if new_num_tokens != config.vocab_size.
""" Resize input token embeddings matrix of the model if new_num_tokens != config.vocab_size.
Take care of tying weights embeddings afterwards if the model class has a `tie_weights()` method.
Take care of tying weights embeddings afterwards if the model class has a `tie_weights()` method.
Args:
Arguments:
new_num_tokens: (`optional`) int
New number of tokens in the embedding matrix.
new_num_tokens: (`optional`) int:
Increasing the size will add newly initialized vectors at the end
New number of tokens in the embedding matrix. Increasing the size will add newly initialized vectors at the end. Reducing the size will remove vectors from the end.
Reducing the size will remove vectors from the end
If not provided or None: does nothing and just returns a pointer to the input tokens ``torch.nn.Embeddings`` Module of the model.
If not provided or None: does nothing and just returns a pointer to the input tokens Embedding Module of the model.
Return: ``torch.nn.Embeddings``
Return: ``torch.nn.Embeddings``
Pointer to the input tokens Embedding Module of the model
Pointer to the input tokens Embeddings Module of the model
"""
"""
base_model=getattr(self,self.base_model_prefix,self)# get the base model if needed
base_model=getattr(self,self.base_model_prefix,self)# get the base model if needed
@@ -309,15 +319,17 @@ class PreTrainedModel(nn.Module):
...
@@ -309,15 +319,17 @@ class PreTrainedModel(nn.Module):
defprune_heads(self,heads_to_prune):
defprune_heads(self,heads_to_prune):
""" Prunes heads of the base model.
""" Prunes heads of the base model.
Args:
heads_to_prune: dict of {layer_num (int): list of heads to prune in this layer (list of int)}
Arguments:
heads_to_prune: dict with keys being selected layer indices (`int`) and associated values being the list of heads to prune in said layer (list of `int`).
"""
"""
base_model=getattr(self,self.base_model_prefix,self)# get the base model if needed
base_model=getattr(self,self.base_model_prefix,self)# get the base model if needed
base_model._prune_heads(heads_to_prune)
base_model._prune_heads(heads_to_prune)
defsave_pretrained(self,save_directory):
defsave_pretrained(self,save_directory):
""" Save a model with its configuration file to a directory, so that it
""" Save a model and its configuration file to a directory, so that it
can be re-loaded using the `from_pretrained(save_directory)` class method.
can be re-loaded using the `:func:`~pytorch_transformers.PreTrainedModel.from_pretrained`` class method.
"""
"""
assertos.path.isdir(save_directory),"Saving path should be a directory where the model and configuration can be saved"
assertos.path.isdir(save_directory),"Saving path should be a directory where the model and configuration can be saved"
...
@@ -336,50 +348,45 @@ class PreTrainedModel(nn.Module):
...
@@ -336,50 +348,45 @@ class PreTrainedModel(nn.Module):
r"""Instantiate a pretrained pytorch model from a pre-trained model configuration.
r"""Instantiate a pretrained pytorch model from a pre-trained model configuration.
The model is set in evaluation mode by default using `model.eval()` (Dropout modules are deactivated)
The model is set in evaluation mode by default using ``model.eval()`` (Dropout modules are deactivated)
To train the model, you should first set it back in training mode with `model.train()`
To train the model, you should first set it back in training mode with ``model.train()``
Params:
Parameters:
**pretrained_model_name_or_path**: either:
pretrained_model_name_or_path: either:
- a string with the `shortcut name` of a pre-trained model to load from cache
or download and cache if not already stored in cache (e.g. 'bert-base-uncased').
- a string with the `shortcut name` of a pre-trained model to load from cache or download, e.g.: ``bert-base-uncased``.
- a path to a `directory` containing a configuration file saved
- a path to a `directory` containing model weights saved using :func:`~pytorch_transformers.PreTrainedModel.save_pretrained`, e.g.: ``./my_model_directory/``.
using the `save_pretrained(save_directory)` method.
- a path or url to a `tensorflow index checkpoint file` (e.g. `./tf_model/model.ckpt.index`). In this case, ``from_tf`` should be set to True and a configuration object should be provided as ``config`` argument. This loading path is slower than converting the TensorFlow checkpoint in a PyTorch model using the provided conversion scripts and loading the PyTorch model afterwards.
- a path or url to a tensorflow index checkpoint `file` (e.g. `./tf_model/model.ckpt.index`).
In this case, ``from_tf`` should be set to True and a configuration object should be
model_args: (`optional`) Sequence of positional arguments:
provided as `config` argument. This loading option is slower than converting the TensorFlow
All remaning positional arguments will be passed to the underlying model's ``__init__`` method
checkpoint in a PyTorch model using the provided conversion scripts and loading
the PyTorch model afterwards.
config: (`optional`) instance of a class derived from :class:`~pytorch_transformers.PretrainedConfig`:
**model_args**: (`optional`) Sequence:
Configuration for the model to use instead of an automatically loaded configuation. Configuration can be automatically loaded when:
All remaning positional arguments will be passed to the underlying model's __init__ function
**config**: an optional configuration for the model to use instead of an automatically loaded configuation.
- the model is a model provided by the library (loaded with the ``shortcut-name`` string of a pretrained model), or
Configuration can be automatically loaded when:
- the model was saved using :func:`~pytorch_transformers.PreTrainedModel.save_pretrained` and is reloaded by suppling the save directory.
- the model is a model provided by the library (loaded with a `shortcut name` of a pre-trained model), or
- the model is loaded by suppling a local directory as ``pretrained_model_name_or_path`` and a configuration JSON file named `config.json` is found in the directory.
- the model was saved using the `save_pretrained(save_directory)` (loaded by suppling the save directory).
**state_dict**: an optional state dictionnary for the model to use instead of a state dictionary loaded
state_dict: (`optional`) dict:
from saved weights file.
an optional state dictionnary for the model to use instead of a state dictionary loaded from saved weights file.
This option can be used if you want to create a model from a pretrained configuration but load your own weights.
This option can be used if you want to create a model from a pretrained configuration but load your own weights.
In this case though, you should check if using `save_pretrained(dir)` and `from_pretrained(save_directory)` is not
In this case though, you should check if using :func:`~pytorch_transformers.PreTrainedModel.save_pretrained` and :func:`~pytorch_transformers.PreTrainedModel.from_pretrained` is not a simpler option.
a simpler option.
**cache_dir**: (`optional`) string:
cache_dir: (`optional`) string:
Path to a directory in which a downloaded pre-trained model
Path to a directory in which a downloaded pre-trained model
configuration should be cached if the standard cache should not be used.
configuration should be cached if the standard cache should not be used.
**output_loading_info**: (`optional`) boolean:
output_loading_info: (`optional`) boolean:
Set to ``True`` to also return a dictionnary containing missing keys, unexpected keys and error messages.
Set to ``True`` to also return a dictionnary containing missing keys, unexpected keys and error messages.
**kwargs**: (`optional`) dict:
Dictionary of key, values to update the configuration object after loading.
kwargs: (`optional`) Remaining dictionary of keyword arguments:
Can be used to override selected configuration parameters. E.g. ``output_attention=True``.
Can be used to update the configuration object (after it being loaded) and initiate the model. (e.g. ``output_attention=True``). Behave differently depending on whether a `config` is provided or automatically loaded:
- If a configuration is providedictionaryfig`, **kwargs will be directly passed
- If a configuration is provided with ``config``, ``**kwargs`` will be directly passed to the underlying model's ``__init__`` method (we assume all relevant updates to the configuration have already been done)
to the underlying model's __init__ method.
- If a configuration is not provided, ``kwargs`` will be first passed to the configuration class initialization function (:func:`~pytorch_transformers.PretrainedConfig.from_pretrained`). Each key of ``kwargs`` that corresponds to a configuration attribute will be used to override said attribute with the supplied ``kwargs`` value. Remaining keys that do not correspond to any configuration attribute will be passed to the underlying model's ``__init__`` function.
- If a configuration is not provided, **kwargs will be first passed to the pretrained
model configuration class loading function (`PretrainedConfig.from_pretrained`).
Examples::
Each key of **kwargs that corresponds to a configuration attribute
will be used to override said attribute with the supplied **kwargs value.
Remaining keys that do not correspond to any configuration attribute will
be passed to the underlying model's __init__ function.
Examples::dictionary
model = BertModel.from_pretrained('bert-base-uncased') # Download model and configuration from S3 and cache.
model = BertModel.from_pretrained('bert-base-uncased') # Download model and configuration from S3 and cache.
model = BertModel.from_pretrained('./test/saved_model/') # E.g. model was saved using `save_pretrained('./test/saved_model/')`
model = BertModel.from_pretrained('./test/saved_model/') # E.g. model was saved using `save_pretrained('./test/saved_model/')`
""" An abstract class to handle dowloading and loading pretrained tokenizers and adding tokens to the vocabulary.
""" Base class for all tokenizers.
Handle all the shared methods for tokenization and special tokens as well as methods dowloading/caching/loading pretrained tokenizers as well as adding tokens to the vocabulary.
Derived class can set up a few special tokens to be used in common scripts and internals:
This class also contain the added tokens in a unified way on top of all tokenizers so we don't have to handle the specific vocabulary augmentation methods of the various underlying dictionary structures (BPE, sentencepiece...).
We defined an added_tokens_encoder to add new tokens to the vocabulary without having to handle the
Class attributes (overridden by derived classes):
specific vocabulary augmentation methods of the various underlying dictionary structures (BPE, sentencepiece...).
- ``vocab_files_names``: a python ``dict`` with, as keys, the ``__init__`` keyword name of each vocabulary file required by the model, and as associated values, the filename for saving the associated file (string).
- ``pretrained_vocab_files_map``: a python ``dict of dict`` the high-level keys being the ``__init__`` keyword name of each vocabulary file required by the model, the low-level being the `short-cut-names` (string) of the pretrained models with, as associated values, the `url` (string) to the associated pretrained vocabulary file.
- ``max_model_input_sizes``: a python ``dict`` with, as keys, the `short-cut-names` (string) of the pretrained models, and as associated values, the maximum length of the sequence inputs of this model, or None if the model has no maximum input size.
Parameters:
- ``bos_token``: (`Optional`) string: a beginning of sentence token. Will be associated to ``self.bos_token``
- ``eos_token``: (`Optional`) string: an end of sentence token. Will be associated to ``self.eos_token``
- ``unk_token``: (`Optional`) string: an unknown token. Will be associated to ``self.unk_token``
- ``sep_token``: (`Optional`) string: a separation token (e.g. to separate context and query in an input sequence). Will be associated to ``self.sep_token``
- ``pad_token``: (`Optional`) string: a padding token. Will be associated to ``self.pad_token``
- ``cls_token``: (`Optional`) string: a classification token (e.g. to extract a summary of an input sequence leveraging self-attention along the full depth of the model). Will be associated to ``self.cls_token``
- ``mask_token``: (`Optional`) string: a masking token (e.g. when training a model with masked-language modeling). Will be associated to ``self.mask_token``
- ``additional_special_tokens``: (`Optional`) list: a list of additional special tokens. Adding all special tokens here ensure they won't be split by the tokenization process. Will be associated to ``self.additional_special_tokens``
"""
"""
vocab_files_names={}
vocab_files_names={}
pretrained_vocab_files_map={}
pretrained_vocab_files_map={}
...
@@ -49,82 +69,98 @@ class PreTrainedTokenizer(object):
...
@@ -49,82 +69,98 @@ class PreTrainedTokenizer(object):
@property
@property
defbos_token(self):
defbos_token(self):
""" Beginning of sentence token (string). Log an error if used while not having been set. """
ifself._bos_tokenisNone:
ifself._bos_tokenisNone:
logger.error("Using bos_token, but it is not set yet.")
logger.error("Using bos_token, but it is not set yet.")
returnself._bos_token
returnself._bos_token
@property
@property
defeos_token(self):
defeos_token(self):
""" End of sentence token (string). Log an error if used while not having been set. """
ifself._eos_tokenisNone:
ifself._eos_tokenisNone:
logger.error("Using eos_token, but it is not set yet.")
logger.error("Using eos_token, but it is not set yet.")
returnself._eos_token
returnself._eos_token
@property
@property
defunk_token(self):
defunk_token(self):
""" Unknown token (string). Log an error if used while not having been set. """
ifself._unk_tokenisNone:
ifself._unk_tokenisNone:
logger.error("Using unk_token, but it is not set yet.")
logger.error("Using unk_token, but it is not set yet.")
returnself._unk_token
returnself._unk_token
@property
@property
defsep_token(self):
defsep_token(self):
""" Separation token (string). E.g. separate context and query in an input sequence. Log an error if used while not having been set. """
ifself._sep_tokenisNone:
ifself._sep_tokenisNone:
logger.error("Using sep_token, but it is not set yet.")
logger.error("Using sep_token, but it is not set yet.")
returnself._sep_token
returnself._sep_token
@property
@property
defpad_token(self):
defpad_token(self):
""" Padding token (string). Log an error if used while not having been set. """
ifself._pad_tokenisNone:
ifself._pad_tokenisNone:
logger.error("Using pad_token, but it is not set yet.")
logger.error("Using pad_token, but it is not set yet.")
returnself._pad_token
returnself._pad_token
@property
@property
defcls_token(self):
defcls_token(self):
""" Classification token (string). E.g. to extract a summary of an input sequence leveraging self-attention along the full depth of the model. Log an error if used while not having been set. """
ifself._cls_tokenisNone:
ifself._cls_tokenisNone:
logger.error("Using cls_token, but it is not set yet.")
logger.error("Using cls_token, but it is not set yet.")
returnself._cls_token
returnself._cls_token
@property
@property
defmask_token(self):
defmask_token(self):
""" Mask token (string). E.g. when training a model with masked-language modeling. Log an error if used while not having been set. """
ifself._mask_tokenisNone:
ifself._mask_tokenisNone:
logger.error("Using mask_token, but it is not set yet.")
logger.error("Using mask_token, but it is not set yet.")
returnself._mask_token
returnself._mask_token
@property
@property
defadditional_special_tokens(self):
defadditional_special_tokens(self):
""" All the additional special tokens you may want to use (list of strings). Log an error if used while not having been set. """
ifself._additional_special_tokensisNone:
ifself._additional_special_tokensisNone:
logger.error("Using additional_special_tokens, but it is not set yet.")
logger.error("Using additional_special_tokens, but it is not set yet.")
returnself._additional_special_tokens
returnself._additional_special_tokens
@bos_token.setter
@bos_token.setter
defbos_token(self,value):
defbos_token(self,value):
self.add_tokens([value])
self._bos_token=value
self._bos_token=value
@eos_token.setter
@eos_token.setter
defeos_token(self,value):
defeos_token(self,value):
self.add_tokens([value])
self._eos_token=value
self._eos_token=value
@unk_token.setter
@unk_token.setter
defunk_token(self,value):
defunk_token(self,value):
self.add_tokens([value])
self._unk_token=value
self._unk_token=value
@sep_token.setter
@sep_token.setter
defsep_token(self,value):
defsep_token(self,value):
self.add_tokens([value])
self._sep_token=value
self._sep_token=value
@pad_token.setter
@pad_token.setter
defpad_token(self,value):
defpad_token(self,value):
self.add_tokens([value])
self._pad_token=value
self._pad_token=value
@cls_token.setter
@cls_token.setter
defcls_token(self,value):
defcls_token(self,value):
self.add_tokens([value])
self._cls_token=value
self._cls_token=value
@mask_token.setter
@mask_token.setter
defmask_token(self,value):
defmask_token(self,value):
self.add_tokens([value])
self._mask_token=value
self._mask_token=value
@additional_special_tokens.setter
@additional_special_tokens.setter
defadditional_special_tokens(self,value):
defadditional_special_tokens(self,value):
self.add_tokens(value)
self._additional_special_tokens=value
self._additional_special_tokens=value
def__init__(self,max_len=None,**kwargs):
def__init__(self,max_len=None,**kwargs):
...
@@ -148,15 +184,47 @@ class PreTrainedTokenizer(object):
...
@@ -148,15 +184,47 @@ class PreTrainedTokenizer(object):
@classmethod
@classmethod
deffrom_pretrained(cls,*inputs,**kwargs):
deffrom_pretrained(cls,*inputs,**kwargs):
r""" Instantiate a :class:`~pytorch_transformers.PreTrainedTokenizer` (or a derived class) from a predefined tokenizer.
Parameters:
pretrained_model_name_or_path: either:
- a string with the `shortcut name` of a predefined tokenizer to load from cache or download, e.g.: ``bert-base-uncased``.
- a path to a `directory` containing vocabulary files required by the tokenizer, for instance saved using the :func:`~pytorch_transformers.PreTrainedTokenizer.save_pretrained` method, e.g.: ``./my_model_directory/``.
- (not applicable to all derived classes) a path or url to a single saved vocabulary file if and only if the tokenizer only requires a single vocabulary file (e.g. Bert, XLNet), e.g.: ``./my_model_directory/vocab.txt``.
cache_dir: (`optional`) string:
Path to a directory in which a downloaded predefined tokenizer vocabulary files should be cached if the standard cache should not be used.
inputs: (`optional`) positional arguments: will be passed to the Tokenizer ``__init__`` method.
kwargs: (`optional`) keyword arguments: will be passed to the Tokenizer ``__init__`` method. Can be used to set special tokens like ``bos_token``, ``eos_token``, ``unk_token``, ``sep_token``, ``pad_token``, ``cls_token``, ``mask_token``, ``additional_special_tokens``. See parameters in the doc string of :class:`~pytorch_transformers.PreTrainedTokenizer` for details.
Examples::
# We can't instantiate directly the base class `PreTrainedTokenizer` so let's show our examples on a derived class: BertTokenizer
Instantiate a PreTrainedTokenizer from pre-trained vocabulary files.
Download and cache the vocabulary files if needed.
"""
cache_dir=kwargs.pop('cache_dir',None)
cache_dir=kwargs.pop('cache_dir',None)
s3_models=list(cls.max_model_input_sizes.keys())
s3_models=list(cls.max_model_input_sizes.keys())
...
@@ -253,8 +321,9 @@ class PreTrainedTokenizer(object):
...
@@ -253,8 +321,9 @@ class PreTrainedTokenizer(object):
defsave_pretrained(self,save_directory):
defsave_pretrained(self,save_directory):
""" Save the tokenizer vocabulary files (with added tokens) and the
""" Save the tokenizer vocabulary files (with added tokens) and the
special-tokens-to-class-attributes-mapping to a directory, so that it
special-tokens-to-class-attributes-mapping to a directory.
can be re-loaded using the `from_pretrained(save_directory)` class method.
This method make sure the full tokenizer can then be re-loaded using the :func:`~pytorch_transformers.PreTrainedTokenizer.from_pretrained` class method.
"""
"""
ifnotos.path.isdir(save_directory):
ifnotos.path.isdir(save_directory):
logger.error("Saving directory ({}) should be a directory".format(save_directory))
logger.error("Saving directory ({}) should be a directory".format(save_directory))
...
@@ -279,37 +348,50 @@ class PreTrainedTokenizer(object):
...
@@ -279,37 +348,50 @@ class PreTrainedTokenizer(object):
defsave_vocabulary(self,save_directory):
defsave_vocabulary(self,save_directory):
""" Save the tokenizer vocabulary to a directory. This method doesn't save added tokens
""" Save the tokenizer vocabulary to a directory. This method does *NOT* save added tokens
and special token mappings.
and special token mappings.
Please use `save_pretrained()` to save the full Tokenizer state so that it can be
Please use :func:`~pytorch_transformers.PreTrainedTokenizer.save_pretrained` `()` to save the full Tokenizer state if you want to reload it using the :func:`~pytorch_transformers.PreTrainedTokenizer.from_pretrained` class method.
reloaded using the `from_pretrained(save_directory)` class method.
"""
"""
raiseNotImplementedError
raiseNotImplementedError
defvocab_size(self):
defvocab_size(self):
""" Size of the base vocabulary (without the added tokens) """
raiseNotImplementedError
raiseNotImplementedError
def__len__(self):
def__len__(self):
""" Size of the full vocabulary with the added tokens """
""" Add a list of new tokens to the tokenizer class. If the new tokens are not in the
""" Add a list of new tokens to the tokenizer class. If the new tokens are not in the
vocabulary, they are added to the added_tokens_encoder with indices starting from
vocabulary, they are added to it with indices starting from length of the current vocabulary.
the last index of the current vocabulary.
Parameters:
new_tokens: list of string. Each string is a token to add. Tokens are only added if they are not already in the vocabulary (tested by checking if the tokenizer assign the index of the ``unk_token`` to them).
Returns:
Returns:
Number of tokens added to the vocabulary which can be used to correspondingly
Number of tokens added to the vocabulary.
increase the size of the associated model embedding matrices.
Examples::
# Let's see how to increase the vocabulary of Bert model and tokenizer
model.resize_token_embeddings(len(tokenizer)) # Notice: resize_token_embeddings expect to receive the full size of the new vocabulary, i.e. the length of the tokenizer.
@@ -325,23 +407,23 @@ class PreTrainedTokenizer(object):
...
@@ -325,23 +407,23 @@ class PreTrainedTokenizer(object):
defadd_special_tokens(self,special_tokens_dict):
defadd_special_tokens(self,special_tokens_dict):
""" Add a dictionary of special tokens (eos, pad, cls...) to the encoder and link them
""" Add a dictionary of special tokens (eos, pad, cls...) to the encoder and link them
to class attributes. If the special tokens are not in the vocabulary, they are added
to class attributes. If special tokens are NOT in the vocabulary, they are added
to it and indexed starting from the last index of the current vocabulary.
to it (indexed starting from the last index of the current vocabulary).
Parameters:
special_tokens_dict: dict of string. Keys should be in the list of predefined special attributes: [``bos_token``, ``eos_token``, ``unk_token``, ``sep_token``, ``pad_token``, ``cls_token``, ``mask_token``, ``additional_special_tokens``].
Tokens are only added if they are not already in the vocabulary (tested by checking if the tokenizer assign the index of the ``unk_token`` to them).
Returns:
Number of tokens added to the vocabulary which can be used to correspondingly
increase the size of the associated model embedding matrices.
