SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Merge branch 'master' into RoBERTa

Browse Source
This commit is contained in:
LysandreJik
2019-08-08 09:42:05 -04:00
parent 39d72bcc7b 7729ef7381
commit d2cc6b101e
62 changed files with 2082 additions and 1387 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
pytorch_transformers/__init__.py
View File

@@ -1,4 +1,5 @@
__version__ = "1.0.0"
from .tokenization_auto import AutoTokenizer
from .tokenization_bert import BertTokenizer, BasicTokenizer, WordpieceTokenizer
from .tokenization_openai import OpenAIGPTTokenizer
from .tokenization_transfo_xl import (TransfoXLTokenizer, TransfoXLCorpus)
@@ -8,6 +9,10 @@ from .tokenization_xlm import XLMTokenizer
from .tokenization_roberta import RobertaTokenizer
from .tokenization_utils import (PreTrainedTokenizer, clean_up_tokenization)
from .tokenization_utils import (PreTrainedTokenizer)
from .modeling_auto import (AutoConfig, AutoModel)
from .modeling_bert import (BertConfig, BertPreTrainedModel, BertModel, BertForPreTraining,
BertForMaskedLM, BertForNextSentencePrediction,
BertForSequenceClassification, BertForMultipleChoice,
@@ -42,4 +47,4 @@ from .modeling_utils import (WEIGHTS_NAME, CONFIG_NAME, TF_WEIGHTS_NAME,
from .optimization import (AdamW, ConstantLRSchedule, WarmupConstantSchedule, WarmupCosineSchedule,
WarmupCosineWithHardRestartsSchedule, WarmupLinearSchedule)
from .file_utils import (PYTORCH_PRETRAINED_BERT_CACHE, cached_path)
from .file_utils import (PYTORCH_TRANSFORMERS_CACHE, PYTORCH_PRETRAINED_BERT_CACHE, cached_path)

2
pytorch_transformers/convert_gpt2_checkpoint_to_pytorch.py
View File

@@ -58,7 +58,7 @@ if __name__ == "__main__":
default = None,
type = str,
required = True,
help = "Path the TensorFlow checkpoint path.")
help = "Path to the TensorFlow checkpoint path.")
parser.add_argument("--pytorch_dump_folder_path",
default = None,
type = str,

2
pytorch_transformers/convert_openai_checkpoint_to_pytorch.py
View File

@@ -58,7 +58,7 @@ if __name__ == "__main__":
default = None,
type = str,
required = True,
help = "Path the TensorFlow checkpoint path.")
help = "Path to the TensorFlow checkpoint path.")
parser.add_argument("--pytorch_dump_folder_path",
default = None,
type = str,

40
pytorch_transformers/convert_pytorch_checkpoint_to_tf.py
View File

@@ -20,7 +20,7 @@ import argparse
import torch
import numpy as np
import tensorflow as tf
from pytorch_pretrained_bert.modeling import BertModel
from pytorch_transformers.modeling import BertModel
def convert_pytorch_checkpoint_to_tf(model:BertModel, ckpt_dir:str, model_name:str):
@@ -41,7 +41,7 @@ def convert_pytorch_checkpoint_to_tf(model:BertModel, ckpt_dir:str, model_name:s
N BertForQuestionAnswering
"""
tensors_to_transopse = (
tensors_to_transpose = (
"dense.weight",
"attention.self.query",
"attention.self.key",
@@ -62,34 +62,34 @@ def convert_pytorch_checkpoint_to_tf(model:BertModel, ckpt_dir:str, model_name:s
if not os.path.isdir(ckpt_dir):
os.makedirs(ckpt_dir)
session = tf.Session()
state_dict = model.state_dict()
tf_vars = []
def to_tf_var_name(name:str):
for patt, repl in iter(var_map):
name = name.replace(patt, repl)
return 'bert/{}'.format(name)
def assign_tf_var(tensor:np.ndarray, name:str):
tmp_var = tf.Variable(initial_value=tensor)
tf_var = tf.get_variable(dtype=tmp_var.dtype, shape=tmp_var.shape, name=name)
op = tf.assign(ref=tf_var, value=tmp_var)
session.run(tf.variables_initializer([tmp_var, tf_var]))
session.run(fetches=[op, tf_var])
def create_tf_var(tensor:np.ndarray, name:str, session:tf.Session):
tf_dtype = tf.dtypes.as_dtype(tensor.dtype)
tf_var = tf.get_variable(dtype=tf_dtype, shape=tensor.shape, name=name, initializer=tf.zeros_initializer())
session.run(tf.variables_initializer([tf_var]))
session.run(tf_var)
return tf_var
for var_name in state_dict:
tf_name = to_tf_var_name(var_name)
torch_tensor = state_dict[var_name].numpy()
if any([x in var_name for x in tensors_to_transopse]):
torch_tensor = torch_tensor.T
tf_tensor = assign_tf_var(tensor=torch_tensor, name=tf_name)
tf_vars.append(tf_tensor)
print("{0}{1}initialized".format(tf_name, " " * (60 - len(tf_name))))
tf.reset_default_graph()
with tf.Session() as session:
for var_name in state_dict:
tf_name = to_tf_var_name(var_name)
torch_tensor = state_dict[var_name].numpy()
if any([x in var_name for x in tensors_to_transpose]):
torch_tensor = torch_tensor.T
tf_var = create_tf_var(tensor=torch_tensor, name=tf_name, session=session)
tf.keras.backend.set_value(tf_var, torch_tensor)
tf_weight = session.run(tf_var)
print("Successfully created {}: {}".format(tf_name, np.allclose(tf_weight, torch_tensor)))
saver = tf.train.Saver(tf_vars)
saver.save(session, os.path.join(ckpt_dir, model_name.replace("-", "_") + ".ckpt"))
saver = tf.train.Saver(tf.trainable_variables())
saver.save(session, os.path.join(ckpt_dir, model_name.replace("-", "_") + ".ckpt"))
def main(raw_args=None):

2
pytorch_transformers/convert_tf_checkpoint_to_pytorch.py
View File

@@ -47,7 +47,7 @@ if __name__ == "__main__":
default = None,
type = str,
required = True,
help = "Path the TensorFlow checkpoint path.")
help = "Path to the TensorFlow checkpoint path.")
parser.add_argument("--bert_config_file",
default = None,
type = str,

9
pytorch_transformers/convert_transfo_xl_checkpoint_to_pytorch.py
View File

@@ -24,11 +24,10 @@ from io import open
import torch
import pytorch_transformers.tokenization_transfo_xl as data_utils
from pytorch_transformers.modeling_transfo_xl import (CONFIG_NAME,
WEIGHTS_NAME,
TransfoXLConfig,
TransfoXLLMHeadModel,
load_tf_weights_in_transfo_xl)
from pytorch_transformers import CONFIG_NAME, WEIGHTS_NAME
from pytorch_transformers.modeling_transfo_xl import (TransfoXLConfig, TransfoXLLMHeadModel,
load_tf_weights_in_transfo_xl)
from pytorch_transformers.tokenization_transfo_xl import (CORPUS_NAME, VOCAB_FILES_NAMES)
if sys.version_info[0] == 2:

2
pytorch_transformers/convert_xlnet_checkpoint_to_pytorch.py
View File

@@ -79,7 +79,7 @@ if __name__ == "__main__":
default = None,
type = str,
required = True,
help = "Path the TensorFlow checkpoint path.")
help = "Path to the TensorFlow checkpoint path.")
parser.add_argument("--xlnet_config_file",
default = None,
type = str,

15
pytorch_transformers/file_utils.py
View File

@@ -38,10 +38,13 @@ except ImportError:
try:
from pathlib import Path
PYTORCH_PRETRAINED_BERT_CACHE = Path(
os.getenv('PYTORCH_PRETRAINED_BERT_CACHE', default_cache_path))
os.getenv('PYTORCH_TRANSFORMERS_CACHE', os.getenv('PYTORCH_PRETRAINED_BERT_CACHE', default_cache_path)))
except (AttributeError, ImportError):
PYTORCH_PRETRAINED_BERT_CACHE = os.getenv('PYTORCH_PRETRAINED_BERT_CACHE',
default_cache_path)
PYTORCH_PRETRAINED_BERT_CACHE = os.getenv('PYTORCH_TRANSFORMERS_CACHE',
os.getenv('PYTORCH_PRETRAINED_BERT_CACHE',
default_cache_path))
PYTORCH_TRANSFORMERS_CACHE = PYTORCH_PRETRAINED_BERT_CACHE # Kept for backward compatibility
logger = logging.getLogger(__name__) # pylint: disable=invalid-name
@@ -70,7 +73,7 @@ def filename_to_url(filename, cache_dir=None):
Raise ``EnvironmentError`` if `filename` or its stored metadata do not exist.
"""
if cache_dir is None:
cache_dir = PYTORCH_PRETRAINED_BERT_CACHE
cache_dir = PYTORCH_TRANSFORMERS_CACHE
if sys.version_info[0] == 3 and isinstance(cache_dir, Path):
cache_dir = str(cache_dir)
@@ -98,7 +101,7 @@ def cached_path(url_or_filename, cache_dir=None):
make sure the file exists and then return the path.
"""
if cache_dir is None:
cache_dir = PYTORCH_PRETRAINED_BERT_CACHE
cache_dir = PYTORCH_TRANSFORMERS_CACHE
if sys.version_info[0] == 3 and isinstance(url_or_filename, Path):
url_or_filename = str(url_or_filename)
if sys.version_info[0] == 3 and isinstance(cache_dir, Path):
@@ -187,7 +190,7 @@ def get_from_cache(url, cache_dir=None):
If it's not there, download it. Then return the path to the cached file.
"""
if cache_dir is None:
cache_dir = PYTORCH_PRETRAINED_BERT_CACHE
cache_dir = PYTORCH_TRANSFORMERS_CACHE
if sys.version_info[0] == 3 and isinstance(cache_dir, Path):
cache_dir = str(cache_dir)
if sys.version_info[0] == 2 and not isinstance(cache_dir, str):

236
pytorch_transformers/modeling_auto.py Normal file
View File

@@ -0,0 +1,236 @@
# coding=utf-8
# Copyright 2018 The HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Auto Model class. """
from __future__ import absolute_import, division, print_function, unicode_literals
import logging
import torch
import torch.nn as nn
from torch.nn import CrossEntropyLoss, MSELoss
from torch.nn.parameter import Parameter
from .modeling_bert import BertConfig, BertModel
from .modeling_openai import OpenAIGPTConfig, OpenAIGPTModel
from .modeling_gpt2 import GPT2Config, GPT2Model
from .modeling_transfo_xl import TransfoXLConfig, TransfoXLModel
from .modeling_xlnet import XLNetConfig, XLNetModel
from .modeling_xlm import XLMConfig, XLMModel
from .modeling_utils import PreTrainedModel, SequenceSummary
logger = logging.getLogger(__name__)
class AutoConfig(object):
r""":class:`~pytorch_transformers.AutoConfig` is a generic configuration class
that will be instantiated as one of the configuration classes of the library
when created with the `AutoConfig.from_pretrained(pretrained_model_name_or_path)`
class method.
The `from_pretrained()` method take care of returning the correct model class instance
using pattern matching on the `pretrained_model_name_or_path` string.
The base model class to instantiate is selected as the first pattern matching
in the `pretrained_model_name_or_path` string (in the following order):
- contains `bert`: BertConfig (Bert model)
- contains `openai-gpt`: OpenAIGPTConfig (OpenAI GPT model)
- contains `gpt2`: GPT2Config (OpenAI GPT-2 model)
- contains `transfo-xl`: TransfoXLConfig (Transformer-XL model)
- contains `xlnet`: XLNetConfig (XLNet model)
- contains `xlm`: XLMConfig (XLM model)
This class cannot be instantiated using `__init__()` (throw an error).
"""
def __init__(self):
raise EnvironmentError("AutoConfig is designed to be instantiated "
"using the `AutoConfig.from_pretrained(pretrained_model_name_or_path)` method.")
@classmethod
def from_pretrained(cls, pretrained_model_name_or_path, **kwargs):
r""" Instantiate a one of the configuration classes of the library
from a pre-trained model configuration.
The configuration class to instantiate is selected as the first pattern matching
in the `pretrained_model_name_or_path` string (in the following order):
- contains `bert`: BertConfig (Bert model)
- contains `openai-gpt`: OpenAIGPTConfig (OpenAI GPT model)
- contains `gpt2`: GPT2Config (OpenAI GPT-2 model)
- contains `transfo-xl`: TransfoXLConfig (Transformer-XL model)
- contains `xlnet`: XLNetConfig (XLNet model)
- contains `xlm`: XLMConfig (XLM model)
Params:
**pretrained_model_name_or_path**: either:
- a string with the `shortcut name` of a pre-trained model configuration to load from cache
or download and cache if not already stored in cache (e.g. 'bert-base-uncased').
- a path to a `directory` containing a configuration file saved
using the `save_pretrained(save_directory)` method.
- a path or url to a saved configuration `file`.
**cache_dir**: (`optional`) string:
Path to a directory in which a downloaded pre-trained model
configuration should be cached if the standard cache should not be used.
**return_unused_kwargs**: (`optional`) bool:
- If False, then this function returns just the final configuration object.
- 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.
**kwargs**: (`optional`) dict:
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
to override the loaded values.
- Behavior concerning key/value pairs whose keys are *not* configuration attributes is controlled
by the `return_unused_kwargs` keyword parameter.
Examples::
config = AutoConfig.from_pretrained('bert-base-uncased') # Download configuration from S3 and cache.
config = AutoConfig.from_pretrained('./test/bert_saved_model/') # E.g. config (or model) was saved using `save_pretrained('./test/saved_model/')`
config = AutoConfig.from_pretrained('./test/bert_saved_model/my_configuration.json')
config = AutoConfig.from_pretrained('bert-base-uncased', output_attention=True, foo=False)
assert config.output_attention == True
config, unused_kwargs = AutoConfig.from_pretrained('bert-base-uncased', output_attention=True,
foo=False, return_unused_kwargs=True)
assert config.output_attention == True
assert unused_kwargs == {'foo': False}
"""
if 'bert' in pretrained_model_name_or_path:
return BertConfig.from_pretrained(pretrained_model_name_or_path, **kwargs)
elif 'openai-gpt' in pretrained_model_name_or_path:
return OpenAIGPTConfig.from_pretrained(pretrained_model_name_or_path, **kwargs)
elif 'gpt2' in pretrained_model_name_or_path:
return GPT2Config.from_pretrained(pretrained_model_name_or_path, **kwargs)
elif 'transfo-xl' in pretrained_model_name_or_path:
return TransfoXLConfig.from_pretrained(pretrained_model_name_or_path, **kwargs)
elif 'xlnet' in pretrained_model_name_or_path:
return XLNetConfig.from_pretrained(pretrained_model_name_or_path, **kwargs)
elif 'xlm' in pretrained_model_name_or_path:
return XLMConfig.from_pretrained(pretrained_model_name_or_path, **kwargs)
raise ValueError("Unrecognized model identifier in {}. Should contains one of "
"'bert', 'openai-gpt', 'gpt2', 'transfo-xl', 'xlnet', "
"'xlm'".format(pretrained_model_name_or_path))
class AutoModel(object):
r"""
:class:`~pytorch_transformers.AutoModel` is a generic model class
that will be instantiated as one of the base model classes of the library
when created with the `AutoModel.from_pretrained(pretrained_model_name_or_path)`
class method.
The `from_pretrained()` method take care of returning the correct model class instance
using pattern matching on the `pretrained_model_name_or_path` string.
The base model class to instantiate is selected as the first pattern matching
in the `pretrained_model_name_or_path` string (in the following order):
- contains `bert`: BertConfig (Bert model)
- contains `openai-gpt`: OpenAIGPTConfig (OpenAI GPT model)
- contains `gpt2`: GPT2Config (OpenAI GPT-2 model)
- contains `transfo-xl`: TransfoXLConfig (Transformer-XL model)
- contains `xlnet`: XLNetConfig (XLNet model)
- contains `xlm`: XLMConfig (XLM model)
This class cannot be instantiated using `__init__()` (throw an error).
"""
def __init__(self):
raise EnvironmentError("AutoModel is designed to be instantiated "
"using the `AutoModel.from_pretrained(pretrained_model_name_or_path)` method.")
@classmethod
def from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs):
r""" Instantiate a one of the base model classes of the library
from a pre-trained model configuration.
The base model class to instantiate is selected as the first pattern matching
in the `pretrained_model_name_or_path` string (in the following order):
- contains `bert`: BertConfig (Bert model)
- contains `openai-gpt`: OpenAIGPTConfig (OpenAI GPT model)
- contains `gpt2`: GPT2Config (OpenAI GPT-2 model)
- contains `transfo-xl`: TransfoXLConfig (Transformer-XL model)
- contains `xlnet`: XLNetConfig (XLNet model)
- contains `xlm`: XLMConfig (XLM model)
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()`
Params:
**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 path to a `directory` containing a configuration file saved
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 option is slower than converting the TensorFlow
checkpoint in a PyTorch model using the provided conversion scripts and loading
the PyTorch model afterwards.
**model_args**: (`optional`) Sequence:
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.
Configuration can be automatically loaded when:
- the model is a model provided by the library (loaded with a `shortcut name` of a pre-trained model), or
- 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
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.
In this case though, you should check if using `save_pretrained(dir)` and `from_pretrained(save_directory)` is not
a simpler option.
**cache_dir**: (`optional`) string:
Path to a directory in which a downloaded pre-trained model
configuration should be cached if the standard cache should not be used.
**output_loading_info**: (`optional`) boolean:
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.
Can be used to override selected configuration parameters. E.g. ``output_attention=True``.
- If a configuration is provided with `config`, **kwargs will be directly passed
to the underlying model's __init__ method.
- If a configuration is not provided, **kwargs will be first passed to the pretrained
model configuration class loading function (`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.
Examples::
model = AutoModel.from_pretrained('bert-base-uncased') # Download model and configuration from S3 and cache.
model = AutoModel.from_pretrained('./test/bert_model/') # E.g. model was saved using `save_pretrained('./test/saved_model/')`
model = AutoModel.from_pretrained('bert-base-uncased', output_attention=True) # Update configuration during loading
assert model.config.output_attention == True
# Loading from a TF checkpoint file instead of a PyTorch model (slower)
config = AutoConfig.from_json_file('./tf_model/bert_tf_model_config.json')
model = AutoModel.from_pretrained('./tf_model/bert_tf_checkpoint.ckpt.index', from_tf=True, config=config)
"""
if 'bert' in pretrained_model_name_or_path:
return BertModel.from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)
elif 'openai-gpt' in pretrained_model_name_or_path:
return OpenAIGPTModel.from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)
elif 'gpt2' in pretrained_model_name_or_path:
return GPT2Model.from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)
elif 'transfo-xl' in pretrained_model_name_or_path:
return TransfoXLModel.from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)
elif 'xlnet' in pretrained_model_name_or_path:
return XLNetModel.from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)
elif 'xlm' in pretrained_model_name_or_path:
return XLMModel.from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)
raise ValueError("Unrecognized model identifier in {}. Should contains one of "
"'bert', 'openai-gpt', 'gpt2', 'transfo-xl', 'xlnet', "
"'xlm'".format(pretrained_model_name_or_path))

111
pytorch_transformers/modeling_bert.py
View File

@@ -222,7 +222,7 @@ class BertConfig(PretrainedConfig):
try:
from apex.normalization.fused_layer_norm import FusedLayerNorm as BertLayerNorm
except ImportError:
except (ImportError, AttributeError) as e:
logger.info("Better speed can be achieved with apex installed from https://www.github.com/nvidia/apex .")
class BertLayerNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-12):
@@ -643,12 +643,11 @@ class BertModel(BertPreTrainedModel):
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>> model = BertModel(config)
>>> 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
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertModel.from_pretrained('bert-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):
@@ -754,13 +753,11 @@ class BertForPreTraining(BertPreTrainedModel):
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForPreTraining(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids)
>>> prediction_scores, seq_relationship_scores = outputs[:2]
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertForPreTraining.from_pretrained('bert-base-uncased')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
outputs = model(input_ids)
prediction_scores, seq_relationship_scores = outputs[:2]
"""
def __init__(self, config):
@@ -824,13 +821,11 @@ class BertForMaskedLM(BertPreTrainedModel):
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForMaskedLM(config)
>>> 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]
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertForMaskedLM.from_pretrained('bert-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):
@@ -857,7 +852,7 @@ class BertForMaskedLM(BertPreTrainedModel):
sequence_output = outputs[0]
prediction_scores = self.cls(sequence_output)
outputs = (prediction_scores,) + outputs[2:] # Add hidden states and attention is they are here
outputs = (prediction_scores,) + outputs[2:] # Add hidden states and attention if they are here
if masked_lm_labels is not None:
loss_fct = CrossEntropyLoss(ignore_index=-1)
masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), masked_lm_labels.view(-1))
@@ -891,13 +886,11 @@ class BertForNextSentencePrediction(BertPreTrainedModel):
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForNextSentencePrediction(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids)
>>> seq_relationship_scores = outputs[0]
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertForNextSentencePrediction.from_pretrained('bert-base-uncased')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
outputs = model(input_ids)
seq_relationship_scores = outputs[0]
"""
def __init__(self, config):
@@ -951,14 +944,12 @@ class BertForSequenceClassification(BertPreTrainedModel):
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForSequenceClassification(config)
>>> 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]
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertForSequenceClassification.from_pretrained('bert-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):
@@ -1057,15 +1048,13 @@ class BertForMultipleChoice(BertPreTrainedModel):
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForMultipleChoice(config)
>>> choices = ["Hello, my dog is cute", "Hello, my cat is amazing"]
>>> input_ids = torch.tensor([tokenizer.encode(s) for s in choices]).unsqueeze(0) # Batch size 1, 2 choices
>>> labels = torch.tensor(1).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids, labels=labels)
>>> loss, classification_scores = outputs[:2]
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertForMultipleChoice.from_pretrained('bert-base-uncased')
choices = ["Hello, my dog is cute", "Hello, my cat is amazing"]
input_ids = torch.tensor([tokenizer.encode(s) for s in choices]).unsqueeze(0) # Batch size 1, 2 choices
labels = torch.tensor(1).unsqueeze(0) # Batch size 1
outputs = model(input_ids, labels=labels)
loss, classification_scores = outputs[:2]
"""
def __init__(self, config):
@@ -1127,14 +1116,12 @@ class BertForTokenClassification(BertPreTrainedModel):
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForTokenClassification(config)
>>> 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]
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertForTokenClassification.from_pretrained('bert-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):
@@ -1203,15 +1190,13 @@ class BertForQuestionAnswering(BertPreTrainedModel):
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased')
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
>>>
>>> model = BertForQuestionAnswering(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> start_positions = torch.tensor([1])
>>> end_positions = torch.tensor([3])
>>> outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)
>>> loss, start_scores, end_scores = outputs[:2]
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertForQuestionAnswering.from_pretrained('bert-base-uncased')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
start_positions = torch.tensor([1])
end_positions = torch.tensor([3])
outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)
loss, start_scores, end_scores = outputs[:2]
"""
def __init__(self, config):

39
pytorch_transformers/modeling_gpt2.py
View File

@@ -137,7 +137,7 @@ class GPT2Config(PretrainedConfig):
initializer_range=0.02,
num_labels=1,
summary_type='token_ids',
summary_type='cls_index',
summary_use_proj=True,
summary_activation=None,
summary_proj_to_labels=True,
@@ -433,12 +433,11 @@ class GPT2Model(GPT2PreTrainedModel):
Examples::
>>> config = GPT2Config.from_pretrained('gpt2')
>>> tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
>>> model = GPT2Model(config)
>>> 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
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
model = GPT2Model.from_pretrained('gpt2')
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):
@@ -567,12 +566,11 @@ class GPT2LMHeadModel(GPT2PreTrainedModel):
Examples::
>>> config = GPT2Config.from_pretrained('gpt2')
>>> tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
>>> model = GPT2LMHeadModel(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids, labels=input_ids)
>>> loss, logits = outputs[:2]
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
model = GPT2LMHeadModel.from_pretrained('gpt2')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
outputs = model(input_ids, labels=input_ids)
loss, logits = outputs[:2]
"""
def __init__(self, config):
@@ -683,14 +681,13 @@ class GPT2DoubleHeadsModel(GPT2PreTrainedModel):
Examples::
>>> config = GPT2Config.from_pretrained('gpt2')
>>> tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
>>> model = GPT2DoubleHeadsModel(config)
>>> choices = ["Hello, my dog is cute [CLS]", "Hello, my cat is cute [CLS]"] # Assume you've added [CLS] to the vocabulary
>>> input_ids = torch.tensor([tokenizer.encode(s) for s in choices]).unsqueeze(0) # Batch size 1, 2 choices
>>> mc_token_ids = torch.tensor([-1, -1]).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids, mc_token_ids)
>>> lm_prediction_scores, mc_prediction_scores = outputs[:2]
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
model = GPT2DoubleHeadsModel.from_pretrained('gpt2')
choices = ["Hello, my dog is cute [CLS]", "Hello, my cat is cute [CLS]"] # Assume you've added [CLS] to the vocabulary
input_ids = torch.tensor([tokenizer.encode(s) for s in choices]).unsqueeze(0) # Batch size 1, 2 choices
mc_token_ids = torch.tensor([-1, -1]).unsqueeze(0) # Batch size 1
outputs = model(input_ids, mc_token_ids)
lm_prediction_scores, mc_prediction_scores = outputs[:2]
"""
def __init__(self, config):

39
pytorch_transformers/modeling_openai.py
View File

@@ -171,7 +171,7 @@ class OpenAIGPTConfig(PretrainedConfig):
predict_special_tokens=True,
num_labels=1,
summary_type='token_ids',
summary_type='cls_index',
summary_use_proj=True,
summary_activation=None,
summary_proj_to_labels=True,
@@ -439,12 +439,11 @@ class OpenAIGPTModel(OpenAIGPTPreTrainedModel):
Examples::
>>> config = OpenAIGPTConfig.from_pretrained('openai-gpt')
>>> tokenizer = OpenAIGPTTokenizer.from_pretrained('openai-gpt')
>>> model = OpenAIGPTModel(config)
>>> 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
tokenizer = OpenAIGPTTokenizer.from_pretrained('openai-gpt')
model = OpenAIGPTModel.from_pretrained('openai-gpt')
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):
@@ -558,12 +557,11 @@ class OpenAIGPTLMHeadModel(OpenAIGPTPreTrainedModel):
Examples::
>>> config = OpenAIGPTConfig.from_pretrained('openai-gpt')
>>> tokenizer = OpenAIGPTTokenizer.from_pretrained('openai-gpt')
>>> model = OpenAIGPTLMHeadModel(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids, labels=input_ids)
>>> loss, logits = outputs[:2]
tokenizer = OpenAIGPTTokenizer.from_pretrained('openai-gpt')
model = OpenAIGPTLMHeadModel.from_pretrained('openai-gpt')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
outputs = model(input_ids, labels=input_ids)
loss, logits = outputs[:2]
"""
def __init__(self, config):
@@ -665,14 +663,13 @@ class OpenAIGPTDoubleHeadsModel(OpenAIGPTPreTrainedModel):
Examples::
>>> config = OpenAIGPTConfig.from_pretrained('openai-gpt')
>>> tokenizer = OpenAIGPTTokenizer.from_pretrained('openai-gpt')
>>> model = OpenAIGPTDoubleHeadsModel(config)
>>> choices = ["Hello, my dog is cute [CLS]", "Hello, my cat is cute [CLS]"] # Assume you've added [CLS] to the vocabulary
>>> input_ids = torch.tensor([tokenizer.encode(s) for s in choices]).unsqueeze(0) # Batch size 1, 2 choices
>>> mc_token_ids = torch.tensor([-1, -1]).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids, mc_token_ids)
>>> lm_prediction_scores, mc_prediction_scores = outputs[:2]
tokenizer = OpenAIGPTTokenizer.from_pretrained('openai-gpt')
model = OpenAIGPTDoubleHeadsModel.from_pretrained('openai-gpt')
choices = ["Hello, my dog is cute [CLS]", "Hello, my cat is cute [CLS]"] # Assume you've added [CLS] to the vocabulary
input_ids = torch.tensor([tokenizer.encode(s) for s in choices]).unsqueeze(0) # Batch size 1, 2 choices
mc_token_ids = torch.tensor([-1, -1]).unsqueeze(0) # Batch size 1
outputs = model(input_ids, mc_token_ids)
lm_prediction_scores, mc_prediction_scores = outputs[:2]
"""
def __init__(self, config):

22
pytorch_transformers/modeling_transfo_xl.py
View File

@@ -968,12 +968,11 @@ class TransfoXLModel(TransfoXLPreTrainedModel):
Examples::
>>> config = TransfoXLConfig.from_pretrained('transfo-xl-wt103')
>>> tokenizer = TransfoXLTokenizer.from_pretrained('transfo-xl-wt103')
>>> model = TransfoXLModel(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids)
>>> last_hidden_states, mems = outputs[:2]
tokenizer = TransfoXLTokenizer.from_pretrained('transfo-xl-wt103')
model = TransfoXLModel.from_pretrained('transfo-xl-wt103')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
outputs = model(input_ids)
last_hidden_states, mems = outputs[:2]
"""
def __init__(self, config):
@@ -1284,12 +1283,11 @@ class TransfoXLLMHeadModel(TransfoXLPreTrainedModel):
Examples::
>>> config = TransfoXLConfig.from_pretrained('transfo-xl-wt103')
>>> tokenizer = TransfoXLTokenizer.from_pretrained('transfo-xl-wt103')
>>> model = TransfoXLLMHeadModel(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> outputs = model(input_ids)
>>> prediction_scores, mems = outputs[:2]
tokenizer = TransfoXLTokenizer.from_pretrained('transfo-xl-wt103')
model = TransfoXLLMHeadModel.from_pretrained('transfo-xl-wt103')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
outputs = model(input_ids)
prediction_scores, mems = outputs[:2]
"""
def __init__(self, config):

229
pytorch_transformers/modeling_utils.py
View File

@@ -68,8 +68,18 @@ else:
class PretrainedConfig(object):
""" Base class for all configuration classes.
Handle a few common parameters and methods for loading/downloading/saving configurations.
r""" Base class for all configuration classes.
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 = {}
@@ -81,8 +91,8 @@ class PretrainedConfig(object):
self.torchscript = kwargs.pop('torchscript', False)
def save_pretrained(self, save_directory):
""" Save a configuration object to a directory, so that it
can be re-loaded using the `from_pretrained(save_directory)` class method.
""" Save a configuration object to the directory `save_directory`, so that it
can be re-loaded using the :func:`~pytorch_transformers.PretrainedConfig.from_pretrained` class method.
"""
assert os.path.isdir(save_directory), "Saving path should be a directory where the model and configuration can be saved"
@@ -93,41 +103,42 @@ class PretrainedConfig(object):
@classmethod
def from_pretrained(cls, pretrained_model_name_or_path, **kwargs):
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.
Params:
**pretrained_model_name_or_path**: either:
- a string with the `shortcut name` of a pre-trained model configuration to load from cache
or download and cache if not already stored in cache (e.g. 'bert-base-uncased').
- a path to a `directory` containing a configuration file saved
using the `save_pretrained(save_directory)` method.
- a path or url to a saved configuration `file`.
**cache_dir**: (`optional`) string:
Parameters:
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 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``.
cache_dir: (`optional`) string:
Path to a directory in which a downloaded pre-trained model
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.
- The values in kwargs of any keys which are configuration attributes will be used to override the loaded values.
- Behavior concerning key/value pairs whose keys are *not* configuration attributes is controlled by the `return_unused_kwargs` keyword parameter.
return_unused_kwargs: (`optional`) bool:
- If False, then this function returns just the final configuration object.
- 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.
**kwargs**: (`optional`) dict:
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
to override the loaded values.
- Behavior concerning key/value pairs whose keys are *not* configuration attributes is controlled
by the `return_unused_kwargs` keyword parameter.
- 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.
Examples::
>>> config = BertConfig.from_pretrained('bert-base-uncased') # Download configuration from S3 and cache.
>>> config = BertConfig.from_pretrained('./test/saved_model/') # E.g. config (or model) was saved using `save_pretrained('./test/saved_model/')`
>>> config = BertConfig.from_pretrained('./test/saved_model/my_configuration.json')
>>> config = BertConfig.from_pretrained('bert-base-uncased', output_attention=True, foo=False)
>>> assert config.output_attention == True
>>> config, unused_kwargs = BertConfig.from_pretrained('bert-base-uncased', output_attention=True,
>>> foo=False, return_unused_kwargs=True)
>>> assert config.output_attention == True
>>> assert unused_kwargs == {'foo': False}
# We can't instantiate directly the base class `PretrainedConfig` so let's show the examples on a
# derived class: BertConfig
config = BertConfig.from_pretrained('bert-base-uncased') # Download configuration from S3 and cache.
config = BertConfig.from_pretrained('./test/saved_model/') # E.g. config (or model) was saved using `save_pretrained('./test/saved_model/')`
config = BertConfig.from_pretrained('./test/saved_model/my_configuration.json')
config = BertConfig.from_pretrained('bert-base-uncased', output_attention=True, foo=False)
assert config.output_attention == True
config, unused_kwargs = BertConfig.from_pretrained('bert-base-uncased', output_attention=True,
foo=False, return_unused_kwargs=True)
assert config.output_attention == True
assert unused_kwargs == {'foo': False}
"""
cache_dir = kwargs.pop('cache_dir', None)
@@ -217,14 +228,26 @@ class PretrainedConfig(object):
class PreTrainedModel(nn.Module):
""" Base class for all models. Handle loading/storing model config and
a simple interface for dowloading and loading pretrained models.
r""" Base class for all 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 = {}
load_tf_weights = lambda model, config, path: None
base_model_prefix = ""
input_embeddings = None
def __init__(self, config, *inputs, **kwargs):
super(PreTrainedModel, self).__init__()
@@ -282,17 +305,16 @@ class PreTrainedModel(nn.Module):
def resize_token_embeddings(self, new_num_tokens=None):
""" 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:
new_num_tokens: (`optional`) int
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
If not provided or None: does nothing and just returns a pointer to the input tokens Embedding Module of the model.
Arguments:
new_num_tokens: (`optional`) int:
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.
If not provided or None: does nothing and just returns a pointer to the input tokens ``torch.nn.Embeddings`` Module of the model.
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
model_embeds = base_model._resize_token_embeddings(new_num_tokens)
@@ -311,15 +333,17 @@ class PreTrainedModel(nn.Module):
def prune_heads(self, heads_to_prune):
""" 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._prune_heads(heads_to_prune)
def save_pretrained(self, save_directory):
""" Save a model with its configuration file to a directory, so that it
can be re-loaded using the `from_pretrained(save_directory)` class method.
""" Save a model and its configuration file to a directory, so that it
can be re-loaded using the `:func:`~pytorch_transformers.PreTrainedModel.from_pretrained`` class method.
"""
assert os.path.isdir(save_directory), "Saving path should be a directory where the model and configuration can be saved"
@@ -338,58 +362,53 @@ class PreTrainedModel(nn.Module):
def from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs):
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 desactivated)
To train the model, you should first set it back in training mode with `model.train()`
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()``
Params:
**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 path to a `directory` containing a configuration file saved
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 option is slower than converting the TensorFlow
checkpoint in a PyTorch model using the provided conversion scripts and loading
the PyTorch model afterwards.
**model_args**: (`optional`) Sequence:
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.
Configuration can be automatically loaded when:
- the model is a model provided by the library (loaded with a `shortcut name` of a pre-trained model), or
- 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
from saved weights file.
This option can be used if you want to create a model from a pretrained configuraton but load your own weights.
In this case though, you should check if using `save_pretrained(dir)` and `from_pretrained(save_directory)` is not
a simpler option.
**cache_dir**: (`optional`) string:
Parameters:
pretrained_model_name_or_path: either:
- 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 model weights saved using :func:`~pytorch_transformers.PreTrainedModel.save_pretrained`, e.g.: ``./my_model_directory/``.
- 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.
model_args: (`optional`) Sequence of positional arguments:
All remaning positional arguments will be passed to the underlying model's ``__init__`` method
config: (`optional`) instance of a class derived from :class:`~pytorch_transformers.PretrainedConfig`:
Configuration for the model to use instead of an automatically loaded configuation. Configuration can be automatically loaded when:
- the model is a model provided by the library (loaded with the ``shortcut-name`` string of a pretrained model), or
- the model was saved using :func:`~pytorch_transformers.PreTrainedModel.save_pretrained` and is reloaded by suppling the save directory.
- 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.
state_dict: (`optional`) dict:
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.
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.
cache_dir: (`optional`) string:
Path to a directory in which a downloaded pre-trained model
configuration should be cached if the standard cache should not be used.
**output_loading_info**: (`optional`) boolean:
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.
Can be used to override selected configuration parameters. E.g. ``output_attention=True``.
- If a configuration is provided with `config`, **kwargs will be directly passed
to the underlying model's __init__ method.
- If a configuration is not provided, **kwargs will be first passed to the pretrained
model configuration class loading function (`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.
output_loading_info: (`optional`) boolean:
Set to ``True`` to also return a dictionnary containing missing keys, unexpected keys and error messages.
kwargs: (`optional`) Remaining dictionary of keyword arguments:
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 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)
- 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.
Examples::
>>> 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('bert-base-uncased', output_attention=True) # Update configuration during loading
>>> assert model.config.output_attention == True
>>> # Loading from a TF checkpoint file instead of a PyTorch model (slower)
>>> config = BertConfig.from_json_file('./tf_model/my_tf_model_config.json')
>>> model = BertModel.from_pretrained('./tf_model/my_tf_checkpoint.ckpt.index', from_tf=True, config=config)
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('bert-base-uncased', output_attention=True) # Update configuration during loading
assert model.config.output_attention == True
# Loading from a TF checkpoint file instead of a PyTorch model (slower)
config = BertConfig.from_json_file('./tf_model/my_tf_model_config.json')
model = BertModel.from_pretrained('./tf_model/my_tf_checkpoint.ckpt.index', from_tf=True, config=config)
"""
config = kwargs.pop('config', None)
@@ -760,7 +779,7 @@ class SequenceSummary(nn.Module):
- 'last' => [default] take the last token hidden state (like XLNet)
- 'first' => take the first token hidden state (like Bert)
- 'mean' => take the mean of all tokens hidden states
- 'token_ids' => supply a Tensor of classification token indices (GPT/GPT-2)
- 'cls_index' => supply a Tensor of classification token position (GPT/GPT-2)
- 'attn' => Not implemented now, use multi-head attention
summary_use_proj: Add a projection after the vector extraction
summary_proj_to_labels: If True, the projection outputs to config.num_labels classes (otherwise to hidden_size). Default: False.
@@ -772,7 +791,7 @@ class SequenceSummary(nn.Module):
super(SequenceSummary, self).__init__()
self.summary_type = config.summary_type if hasattr(config, 'summary_use_proj') else 'last'
if config.summary_type == 'attn':
if self.summary_type == 'attn':
# We should use a standard multi-head attention module with absolute positional embedding for that.
# Cf. https://github.com/zihangdai/xlnet/blob/master/modeling.py#L253-L276
# We can probably just use the multi-head attention module of PyTorch >=1.1.0
@@ -798,11 +817,11 @@ class SequenceSummary(nn.Module):
if hasattr(config, 'summary_last_dropout') and config.summary_last_dropout > 0:
self.last_dropout = nn.Dropout(config.summary_last_dropout)
def forward(self, hidden_states, token_ids=None):
def forward(self, hidden_states, cls_index=None):
""" hidden_states: float Tensor in shape [bsz, seq_len, hidden_size], the hidden-states of the last layer.
token_ids: [optional] index of the classification token if summary_type == 'token_ids',
cls_index: [optional] position of the classification token if summary_type == 'cls_index',
shape (bsz,) or more generally (bsz, ...) where ... are optional leading dimensions of hidden_states.
if summary_type == 'token_ids' and token_ids is None:
if summary_type == 'cls_index' and cls_index is None:
we take the last token of the sequence as classification token
"""
if self.summary_type == 'last':
@@ -811,14 +830,14 @@ class SequenceSummary(nn.Module):
output = hidden_states[:, 0]
elif self.summary_type == 'mean':
output = hidden_states.mean(dim=1)
elif self.summary_type == 'token_ids':
if token_ids is None:
token_ids = torch.full_like(hidden_states[..., :1, :], hidden_states.shape[-2]-1, dtype=torch.long)
elif self.summary_type == 'cls_index':
if cls_index is None:
cls_index = torch.full_like(hidden_states[..., :1, :], hidden_states.shape[-2]-1, dtype=torch.long)
else:
token_ids = token_ids.unsqueeze(-1).unsqueeze(-1)
token_ids = token_ids.expand((-1,) * (token_ids.dim()-1) + (hidden_states.size(-1),))
# shape of token_ids: (bsz, XX, 1, hidden_size) where XX are optional leading dim of hidden_states
output = hidden_states.gather(-2, token_ids).squeeze(-2) # shape (bsz, XX, hidden_size)
cls_index = cls_index.unsqueeze(-1).unsqueeze(-1)
cls_index = cls_index.expand((-1,) * (cls_index.dim()-1) + (hidden_states.size(-1),))
# shape of cls_index: (bsz, XX, 1, hidden_size) where XX are optional leading dim of hidden_states
output = hidden_states.gather(-2, cls_index).squeeze(-2) # shape (bsz, XX, hidden_size)
elif self.summary_type == 'attn':
raise NotImplementedError

52
pytorch_transformers/modeling_xlm.py
View File

@@ -472,12 +472,11 @@ class XLMModel(XLMPreTrainedModel):
Examples::
>>> config = XLMConfig.from_pretrained('xlm-mlm-en-2048')
>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = XLMModel(config)
>>> 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
tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
model = XLMModel.from_pretrained('xlm-mlm-en-2048')
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
"""
ATTRIBUTES = ['encoder', 'eos_index', 'pad_index', # 'with_output',
@@ -745,12 +744,11 @@ class XLMWithLMHeadModel(XLMPreTrainedModel):
Examples::
>>> config = XLMConfig.from_pretrained('xlm-mlm-en-2048')
>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>> model = XLMWithLMHeadModel(config)
>>> 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
tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
model = XLMWithLMHeadModel.from_pretrained('xlm-mlm-en-2048')
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):
@@ -805,14 +803,12 @@ class XLMForSequenceClassification(XLMPreTrainedModel):
Examples::
>>> config = XLMConfig.from_pretrained('xlm-mlm-en-2048')
>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>>
>>> model = XLMForSequenceClassification(config)
>>> 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]
tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
model = XLMForSequenceClassification.from_pretrained('xlm-mlm-en-2048')
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):
@@ -885,15 +881,13 @@ class XLMForQuestionAnswering(XLMPreTrainedModel):
Examples::
>>> config = XLMConfig.from_pretrained('xlm-mlm-en-2048')
>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>>
>>> model = XLMForQuestionAnswering(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> start_positions = torch.tensor([1])
>>> end_positions = torch.tensor([3])
>>> outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)
>>> loss, start_scores, end_scores = outputs[:2]
tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
model = XLMForQuestionAnswering.from_pretrained('xlm-mlm-en-2048')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
start_positions = torch.tensor([1])
end_positions = torch.tensor([3])
outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)
loss, start_scores, end_scores = outputs[:2]
"""
def __init__(self, config):

64
pytorch_transformers/modeling_xlnet.py
View File

@@ -335,7 +335,7 @@ class XLNetConfig(PretrainedConfig):
try:
from apex.normalization.fused_layer_norm import FusedLayerNorm as XLNetLayerNorm
except ImportError:
except (ImportError, AttributeError) as e:
logger.info("Better speed can be achieved with apex installed from https://www.github.com/nvidia/apex .")
class XLNetLayerNorm(nn.Module):
def __init__(self, d_model, eps=1e-12):
@@ -712,12 +712,11 @@ class XLNetModel(XLNetPreTrainedModel):
Examples::
>>> config = XLNetConfig.from_pretrained('xlnet-large-cased')
>>> tokenizer = XLNetTokenizer.from_pretrained('xlnet-large-cased')
>>> model = XLNetModel(config)
>>> 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
tokenizer = XLNetTokenizer.from_pretrained('xlnet-large-cased')
model = XLNetModel.from_pretrained('xlnet-large-cased')
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):
@@ -1019,17 +1018,16 @@ class XLNetLMHeadModel(XLNetPreTrainedModel):
Examples::
>>> config = XLNetConfig.from_pretrained('xlnet-large-cased')
>>> tokenizer = XLNetTokenizer.from_pretrained('xlnet-large-cased')
>>> model = XLNetLMHeadModel(config)
>>> # We show how to setup inputs to predict a next token using a bi-directional context.
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is very <mask>")).unsqueeze(0) # We will predict the masked token
>>> perm_mask = torch.zeros((1, input_ids.shape[1], input_ids.shape[1]), dtype=torch.float)
>>> perm_mask[:, :, -1] = 1.0 # Previous tokens don't see last token
>>> target_mapping = torch.zeros((1, 1, input_ids.shape[1]), dtype=torch.float) # Shape [1, 1, seq_length] => let's predict one token
>>> target_mapping[0, 0, -1] = 1.0 # Our first (and only) prediction will be the last token of the sequence (the masked token)
>>> outputs = model(input_ids, perm_mask=perm_mask, target_mapping=target_mapping)
>>> next_token_logits = outputs[0] # Output has shape [target_mapping.size(0), target_mapping.size(1), config.vocab_size]
tokenizer = XLNetTokenizer.from_pretrained('xlnet-large-cased')
model = XLNetLMHeadModel.from_pretrained('xlnet-large-cased')
# We show how to setup inputs to predict a next token using a bi-directional context.
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is very <mask>")).unsqueeze(0) # We will predict the masked token
perm_mask = torch.zeros((1, input_ids.shape[1], input_ids.shape[1]), dtype=torch.float)
perm_mask[:, :, -1] = 1.0 # Previous tokens don't see last token
target_mapping = torch.zeros((1, 1, input_ids.shape[1]), dtype=torch.float) # Shape [1, 1, seq_length] => let's predict one token
target_mapping[0, 0, -1] = 1.0 # Our first (and only) prediction will be the last token of the sequence (the masked token)
outputs = model(input_ids, perm_mask=perm_mask, target_mapping=target_mapping)
next_token_logits = outputs[0] # Output has shape [target_mapping.size(0), target_mapping.size(1), config.vocab_size]
"""
def __init__(self, config):
@@ -1100,14 +1098,12 @@ class XLNetForSequenceClassification(XLNetPreTrainedModel):
Examples::
>>> config = XLNetConfig.from_pretrained('xlnet-large-cased')
>>> tokenizer = XLNetTokenizer.from_pretrained('xlnet-large-cased')
>>>
>>> model = XLNetForSequenceClassification(config)
>>> 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]
tokenizer = XLNetTokenizer.from_pretrained('xlnet-large-cased')
model = XLNetForSequenceClassification.from_pretrained('xlnet-large-cased')
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):
@@ -1200,15 +1196,13 @@ class XLNetForQuestionAnswering(XLNetPreTrainedModel):
Examples::
>>> config = XLMConfig.from_pretrained('xlm-mlm-en-2048')
>>> tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
>>>
>>> model = XLMForQuestionAnswering(config)
>>> input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
>>> start_positions = torch.tensor([1])
>>> end_positions = torch.tensor([3])
>>> outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)
>>> loss, start_scores, end_scores = outputs[:2]
tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
model = XLMForQuestionAnswering.from_pretrained('xlnet-large-cased')
input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0) # Batch size 1
start_positions = torch.tensor([1])
end_positions = torch.tensor([3])
outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)
loss, start_scores, end_scores = outputs[:2]
"""
def __init__(self, config):

47
pytorch_transformers/tests/modeling_auto_test.py Normal file
View File

@@ -0,0 +1,47 @@
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import unittest
import shutil
import pytest
import logging
from pytorch_transformers import AutoConfig, BertConfig, AutoModel, BertModel
from pytorch_transformers.modeling_bert import BERT_PRETRAINED_MODEL_ARCHIVE_MAP
from .modeling_common_test import (CommonTestCases, ConfigTester, ids_tensor)
class AutoModelTest(unittest.TestCase):
def test_model_from_pretrained(self):
logging.basicConfig(level=logging.INFO)
for model_name in list(BERT_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model = AutoModel.from_pretrained(model_name)
model, loading_info = AutoModel.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertModel)
for value in loading_info.values():
self.assertEqual(len(value), 0)
if __name__ == "__main__":
unittest.main()

46
pytorch_transformers/tests/tokenization_auto_test.py Normal file
View File

@@ -0,0 +1,46 @@
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import unittest
import shutil
import pytest
import logging
from pytorch_transformers import AutoTokenizer, BertTokenizer, AutoTokenizer, GPT2Tokenizer
from pytorch_transformers.modeling_bert import BERT_PRETRAINED_MODEL_ARCHIVE_MAP
from pytorch_transformers.modeling_gpt2 import GPT2_PRETRAINED_MODEL_ARCHIVE_MAP
class AutoTokenizerTest(unittest.TestCase):
def test_tokenizer_from_pretrained(self):
logging.basicConfig(level=logging.INFO)
for model_name in list(BERT_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
tokenizer = AutoTokenizer.from_pretrained(model_name)
self.assertIsNotNone(tokenizer)
self.assertIsInstance(tokenizer, BertTokenizer)
self.assertGreater(len(tokenizer), 0)
for model_name in list(GPT2_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
tokenizer = AutoTokenizer.from_pretrained(model_name)
self.assertIsNotNone(tokenizer)
self.assertIsInstance(tokenizer, GPT2Tokenizer)
self.assertGreater(len(tokenizer), 0)
if __name__ == "__main__":
unittest.main()

35
pytorch_transformers/tests/tokenization_bert_test.py
View File

@@ -24,30 +24,37 @@ from pytorch_transformers.tokenization_bert import (BasicTokenizer,
_is_control, _is_punctuation,
_is_whitespace, VOCAB_FILES_NAMES)
from .tokenization_tests_commons import create_and_check_tokenizer_commons, TemporaryDirectory
from .tokenization_tests_commons import CommonTestCases
class TokenizationTest(unittest.TestCase):
class BertTokenizationTest(CommonTestCases.CommonTokenizerTester):
tokenizer_class = BertTokenizer
def setUp(self):
super(BertTokenizationTest, self).setUp()
def test_full_tokenizer(self):
vocab_tokens = [
"[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn",
"##ing", ",", "low", "lowest",
]
with TemporaryDirectory() as tmpdirname:
vocab_file = os.path.join(tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
with open(vocab_file, "w", encoding='utf-8') as vocab_writer:
vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
with open(self.vocab_file, "w", encoding='utf-8') as vocab_writer:
vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
input_text = u"UNwant\u00E9d,running"
output_text = u"unwanted, running"
def get_tokenizer(self):
return BertTokenizer.from_pretrained(self.tmpdirname)
create_and_check_tokenizer_commons(self, input_text, output_text, BertTokenizer, tmpdirname)
def get_input_output_texts(self):
input_text = u"UNwant\u00E9d,running"
output_text = u"unwanted, running"
return input_text, output_text
tokenizer = BertTokenizer(vocab_file)
def test_full_tokenizer(self):
tokenizer = BertTokenizer(self.vocab_file)
tokens = tokenizer.tokenize(u"UNwant\u00E9d,running")
self.assertListEqual(tokens, ["un", "##want", "##ed", ",", "runn", "##ing"])
self.assertListEqual(tokenizer.convert_tokens_to_ids(tokens), [7, 4, 5, 10, 8, 9])
tokens = tokenizer.tokenize(u"UNwant\u00E9d,running")
self.assertListEqual(tokens, ["un", "##want", "##ed", ",", "runn", "##ing"])
self.assertListEqual(tokenizer.convert_tokens_to_ids(tokens), [7, 4, 5, 10, 8, 9])
def test_chinese(self):
tokenizer = BasicTokenizer()

55
pytorch_transformers/tests/tokenization_gpt2_test.py
View File

@@ -20,42 +20,49 @@ import json
from pytorch_transformers.tokenization_gpt2 import GPT2Tokenizer, VOCAB_FILES_NAMES
from .tokenization_tests_commons import create_and_check_tokenizer_commons, TemporaryDirectory
from .tokenization_tests_commons import CommonTestCases
class GPT2TokenizationTest(unittest.TestCase):
class GPT2TokenizationTest(CommonTestCases.CommonTokenizerTester):
def test_full_tokenizer(self):
""" Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt """
tokenizer_class = GPT2Tokenizer
def setUp(self):
super(GPT2TokenizationTest, self).setUp()
# Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt
vocab = ["l", "o", "w", "e", "r", "s", "t", "i", "d", "n",
"lo", "low", "er",
"low", "lowest", "newer", "wider", "<unk>"]
vocab_tokens = dict(zip(vocab, range(len(vocab))))
merges = ["#version: 0.2", "l o", "lo w", "e r", ""]
special_tokens_map = {"unk_token": "<unk>"}
self.special_tokens_map = {"unk_token": "<unk>"}
with TemporaryDirectory() as tmpdirname:
vocab_file = os.path.join(tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
merges_file = os.path.join(tmpdirname, VOCAB_FILES_NAMES['merges_file'])
with open(vocab_file, "w") as fp:
fp.write(json.dumps(vocab_tokens))
with open(merges_file, "w") as fp:
fp.write("\n".join(merges))
self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
self.merges_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['merges_file'])
with open(self.vocab_file, "w") as fp:
fp.write(json.dumps(vocab_tokens))
with open(self.merges_file, "w") as fp:
fp.write("\n".join(merges))
input_text = u"lower newer"
output_text = u"lower<unk>newer"
def get_tokenizer(self):
return GPT2Tokenizer.from_pretrained(self.tmpdirname, **self.special_tokens_map)
create_and_check_tokenizer_commons(self, input_text, output_text, GPT2Tokenizer, tmpdirname, **special_tokens_map)
def get_input_output_texts(self):
input_text = u"lower newer"
output_text = u"lower<unk>newer"
return input_text, output_text
tokenizer = GPT2Tokenizer(vocab_file, merges_file, **special_tokens_map)
text = "lower"
bpe_tokens = ["low", "er"]
tokens = tokenizer.tokenize(text)
self.assertListEqual(tokens, bpe_tokens)
def test_full_tokenizer(self):
tokenizer = GPT2Tokenizer(self.vocab_file, self.merges_file, **self.special_tokens_map)
text = "lower"
bpe_tokens = ["low", "er"]
tokens = tokenizer.tokenize(text)
self.assertListEqual(tokens, bpe_tokens)
input_tokens = tokens + [tokenizer.unk_token]
input_bpe_tokens = [13, 12, 17]
self.assertListEqual(
tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
input_tokens = tokens + [tokenizer.unk_token]
input_bpe_tokens = [13, 12, 17]
self.assertListEqual(
tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
if __name__ == '__main__':

54
pytorch_transformers/tests/tokenization_openai_test.py
View File

@@ -20,13 +20,17 @@ import json
from pytorch_transformers.tokenization_openai import OpenAIGPTTokenizer, VOCAB_FILES_NAMES
from .tokenization_tests_commons import create_and_check_tokenizer_commons, TemporaryDirectory
from .tokenization_tests_commons import CommonTestCases
class OpenAIGPTTokenizationTest(unittest.TestCase):
class OpenAIGPTTokenizationTest(CommonTestCases.CommonTokenizerTester):
def test_full_tokenizer(self):
""" Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt """
tokenizer_class = OpenAIGPTTokenizer
def setUp(self):
super(OpenAIGPTTokenizationTest, self).setUp()
# Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt
vocab = ["l", "o", "w", "e", "r", "s", "t", "i", "d", "n",
"w</w>", "r</w>", "t</w>",
"lo", "low", "er</w>",
@@ -34,30 +38,34 @@ class OpenAIGPTTokenizationTest(unittest.TestCase):
vocab_tokens = dict(zip(vocab, range(len(vocab))))
merges = ["#version: 0.2", "l o", "lo w", "e r</w>", ""]
with TemporaryDirectory() as tmpdirname:
vocab_file = os.path.join(tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
merges_file = os.path.join(tmpdirname, VOCAB_FILES_NAMES['merges_file'])
with open(vocab_file, "w") as fp:
fp.write(json.dumps(vocab_tokens))
with open(merges_file, "w") as fp:
fp.write("\n".join(merges))
self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
self.merges_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['merges_file'])
with open(self.vocab_file, "w") as fp:
fp.write(json.dumps(vocab_tokens))
with open(self.merges_file, "w") as fp:
fp.write("\n".join(merges))
input_text = u"lower newer"
output_text = u"lower newer"
def get_tokenizer(self):
return OpenAIGPTTokenizer.from_pretrained(self.tmpdirname)
create_and_check_tokenizer_commons(self, input_text, output_text, OpenAIGPTTokenizer, tmpdirname)
def get_input_output_texts(self):
input_text = u"lower newer"
output_text = u"lower newer"
return input_text, output_text
tokenizer = OpenAIGPTTokenizer(vocab_file, merges_file)
text = "lower"
bpe_tokens = ["low", "er</w>"]
tokens = tokenizer.tokenize(text)
self.assertListEqual(tokens, bpe_tokens)
def test_full_tokenizer(self):
tokenizer = OpenAIGPTTokenizer(self.vocab_file, self.merges_file)
input_tokens = tokens + ["<unk>"]
input_bpe_tokens = [14, 15, 20]
self.assertListEqual(
tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
text = "lower"
bpe_tokens = ["low", "er</w>"]
tokens = tokenizer.tokenize(text)
self.assertListEqual(tokens, bpe_tokens)
input_tokens = tokens + ["<unk>"]
input_bpe_tokens = [14, 15, 20]
self.assertListEqual(
tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
if __name__ == '__main__':

172
pytorch_transformers/tests/tokenization_tests_commons.py
View File

@@ -19,6 +19,7 @@ import sys
from io import open
import tempfile
import shutil
import unittest
if sys.version_info[0] == 2:
import cPickle as pickle
@@ -36,113 +37,124 @@ else:
unicode = str
def create_and_check_save_and_load_tokenizer(tester, tokenizer_class, *inputs, **kwargs):
tokenizer = tokenizer_class.from_pretrained(*inputs, **kwargs)
class CommonTestCases:
before_tokens = tokenizer.encode(u"He is very happy, UNwant\u00E9d,running")
class CommonTokenizerTester(unittest.TestCase):
with TemporaryDirectory() as tmpdirname:
tokenizer.save_pretrained(tmpdirname)
tokenizer = tokenizer.from_pretrained(tmpdirname)
tokenizer_class = None
after_tokens = tokenizer.encode(u"He is very happy, UNwant\u00E9d,running")
tester.assertListEqual(before_tokens, after_tokens)
def setUp(self):
self.tmpdirname = tempfile.mkdtemp()
def create_and_check_pickle_tokenizer(tester, tokenizer_class, *inputs, **kwargs):
tokenizer = tokenizer_class.from_pretrained(*inputs, **kwargs)
tester.assertIsNotNone(tokenizer)
def tearDown(self):
shutil.rmtree(self.tmpdirname)
text = u"Munich and Berlin are nice cities"
subwords = tokenizer.tokenize(text)
def get_tokenizer(self):
raise NotImplementedError
with TemporaryDirectory() as tmpdirname:
def get_input_output_texts(self):
raise NotImplementedError
filename = os.path.join(tmpdirname, u"tokenizer.bin")
pickle.dump(tokenizer, open(filename, "wb"))
def test_save_and_load_tokenizer(self):
tokenizer = self.get_tokenizer()
tokenizer_new = pickle.load(open(filename, "rb"))
before_tokens = tokenizer.encode(u"He is very happy, UNwant\u00E9d,running")
subwords_loaded = tokenizer_new.tokenize(text)
with TemporaryDirectory() as tmpdirname:
tokenizer.save_pretrained(tmpdirname)
tokenizer = tokenizer.from_pretrained(tmpdirname)
tester.assertListEqual(subwords, subwords_loaded)
after_tokens = tokenizer.encode(u"He is very happy, UNwant\u00E9d,running")
self.assertListEqual(before_tokens, after_tokens)
def test_pickle_tokenizer(self):
tokenizer = self.get_tokenizer()
self.assertIsNotNone(tokenizer)
text = u"Munich and Berlin are nice cities"
subwords = tokenizer.tokenize(text)
with TemporaryDirectory() as tmpdirname:
filename = os.path.join(tmpdirname, u"tokenizer.bin")
pickle.dump(tokenizer, open(filename, "wb"))
tokenizer_new = pickle.load(open(filename, "rb"))
subwords_loaded = tokenizer_new.tokenize(text)
self.assertListEqual(subwords, subwords_loaded)
def create_and_check_add_tokens_tokenizer(tester, tokenizer_class, *inputs, **kwargs):
tokenizer = tokenizer_class.from_pretrained(*inputs, **kwargs)
def test_add_tokens_tokenizer(self):
tokenizer = self.get_tokenizer()
vocab_size = tokenizer.vocab_size
all_size = len(tokenizer)
vocab_size = tokenizer.vocab_size
all_size = len(tokenizer)
tester.assertNotEqual(vocab_size, 0)
tester.assertEqual(vocab_size, all_size)
self.assertNotEqual(vocab_size, 0)
self.assertEqual(vocab_size, all_size)
new_toks = ["aaaaabbbbbb", "cccccccccdddddddd"]
added_toks = tokenizer.add_tokens(new_toks)
vocab_size_2 = tokenizer.vocab_size
all_size_2 = len(tokenizer)
new_toks = ["aaaaabbbbbb", "cccccccccdddddddd"]
added_toks = tokenizer.add_tokens(new_toks)
vocab_size_2 = tokenizer.vocab_size
all_size_2 = len(tokenizer)
tester.assertNotEqual(vocab_size_2, 0)
tester.assertEqual(vocab_size, vocab_size_2)
tester.assertEqual(added_toks, len(new_toks))
tester.assertEqual(all_size_2, all_size + len(new_toks))
self.assertNotEqual(vocab_size_2, 0)
self.assertEqual(vocab_size, vocab_size_2)
self.assertEqual(added_toks, len(new_toks))
self.assertEqual(all_size_2, all_size + len(new_toks))
tokens = tokenizer.encode("aaaaabbbbbb low cccccccccdddddddd l")
tester.assertGreaterEqual(len(tokens), 4)
tester.assertGreater(tokens[0], tokenizer.vocab_size - 1)
tester.assertGreater(tokens[-2], tokenizer.vocab_size - 1)
tokens = tokenizer.encode("aaaaabbbbbb low cccccccccdddddddd l")
self.assertGreaterEqual(len(tokens), 4)
self.assertGreater(tokens[0], tokenizer.vocab_size - 1)
self.assertGreater(tokens[-2], tokenizer.vocab_size - 1)
new_toks_2 = {'eos_token': ">>>>|||<||<<|<<",
'pad_token': "<<<<<|||>|>>>>|>"}
added_toks_2 = tokenizer.add_special_tokens(new_toks_2)
vocab_size_3 = tokenizer.vocab_size
all_size_3 = len(tokenizer)
new_toks_2 = {'eos_token': ">>>>|||<||<<|<<",
'pad_token': "<<<<<|||>|>>>>|>"}
added_toks_2 = tokenizer.add_special_tokens(new_toks_2)
vocab_size_3 = tokenizer.vocab_size
all_size_3 = len(tokenizer)
tester.assertNotEqual(vocab_size_3, 0)
tester.assertEqual(vocab_size, vocab_size_3)
tester.assertEqual(added_toks_2, len(new_toks_2))
tester.assertEqual(all_size_3, all_size_2 + len(new_toks_2))
self.assertNotEqual(vocab_size_3, 0)
self.assertEqual(vocab_size, vocab_size_3)
self.assertEqual(added_toks_2, len(new_toks_2))
self.assertEqual(all_size_3, all_size_2 + len(new_toks_2))
tokens = tokenizer.encode(">>>>|||<||<<|<< aaaaabbbbbb low cccccccccdddddddd <<<<<|||>|>>>>|> l")
tokens = tokenizer.encode(">>>>|||<||<<|<< aaaaabbbbbb low cccccccccdddddddd <<<<<|||>|>>>>|> l")
tester.assertGreaterEqual(len(tokens), 6)
tester.assertGreater(tokens[0], tokenizer.vocab_size - 1)
tester.assertGreater(tokens[0], tokens[1])
tester.assertGreater(tokens[-2], tokenizer.vocab_size - 1)
tester.assertGreater(tokens[-2], tokens[-3])
tester.assertEqual(tokens[0], tokenizer.convert_tokens_to_ids(tokenizer.eos_token))
tester.assertEqual(tokens[-2], tokenizer.convert_tokens_to_ids(tokenizer.pad_token))
self.assertGreaterEqual(len(tokens), 6)
self.assertGreater(tokens[0], tokenizer.vocab_size - 1)
self.assertGreater(tokens[0], tokens[1])
self.assertGreater(tokens[-2], tokenizer.vocab_size - 1)
self.assertGreater(tokens[-2], tokens[-3])
self.assertEqual(tokens[0], tokenizer.convert_tokens_to_ids(tokenizer.eos_token))
self.assertEqual(tokens[-2], tokenizer.convert_tokens_to_ids(tokenizer.pad_token))
def create_and_check_required_methods_tokenizer(tester, input_text, output_text, tokenizer_class, *inputs, **kwargs):
tokenizer = tokenizer_class.from_pretrained(*inputs, **kwargs)
def test_required_methods_tokenizer(self):
tokenizer = self.get_tokenizer()
input_text, output_text = self.get_input_output_texts()
tokens = tokenizer.tokenize(input_text)
ids = tokenizer.convert_tokens_to_ids(tokens)
ids_2 = tokenizer.encode(input_text)
tester.assertListEqual(ids, ids_2)
tokens = tokenizer.tokenize(input_text)
ids = tokenizer.convert_tokens_to_ids(tokens)
ids_2 = tokenizer.encode(input_text)
self.assertListEqual(ids, ids_2)
tokens_2 = tokenizer.convert_ids_to_tokens(ids)
text_2 = tokenizer.decode(ids)
tokens_2 = tokenizer.convert_ids_to_tokens(ids)
text_2 = tokenizer.decode(ids)
tester.assertEqual(text_2, output_text)
self.assertEqual(text_2, output_text)
tester.assertNotEqual(len(tokens_2), 0)
tester.assertIsInstance(text_2, (str, unicode))
self.assertNotEqual(len(tokens_2), 0)
self.assertIsInstance(text_2, (str, unicode))
def create_and_check_pretrained_model_lists(tester, input_text, output_text, tokenizer_class, *inputs, **kwargs):
weights_list = list(tokenizer_class.max_model_input_sizes.keys())
weights_lists_2 = []
for file_id, map_list in tokenizer_class.pretrained_vocab_files_map.items():
weights_lists_2.append(list(map_list.keys()))
def test_pretrained_model_lists(self):
weights_list = list(self.tokenizer_class.max_model_input_sizes.keys())
weights_lists_2 = []
for file_id, map_list in self.tokenizer_class.pretrained_vocab_files_map.items():
weights_lists_2.append(list(map_list.keys()))
for weights_list_2 in weights_lists_2:
tester.assertListEqual(weights_list, weights_list_2)
def create_and_check_tokenizer_commons(tester, input_text, output_text, tokenizer_class, *inputs, **kwargs):
create_and_check_pretrained_model_lists(tester, input_text, output_text, tokenizer_class, *inputs, **kwargs)
create_and_check_required_methods_tokenizer(tester, input_text, output_text, tokenizer_class, *inputs, **kwargs)
create_and_check_add_tokens_tokenizer(tester, tokenizer_class, *inputs, **kwargs)
create_and_check_save_and_load_tokenizer(tester, tokenizer_class, *inputs, **kwargs)
create_and_check_pickle_tokenizer(tester, tokenizer_class, *inputs, **kwargs)
for weights_list_2 in weights_lists_2:
self.assertListEqual(weights_list, weights_list_2)

37
pytorch_transformers/tests/tokenization_transfo_xl_test.py
View File

@@ -20,32 +20,39 @@ from io import open
from pytorch_transformers.tokenization_transfo_xl import TransfoXLTokenizer, VOCAB_FILES_NAMES
from.tokenization_tests_commons import create_and_check_tokenizer_commons, TemporaryDirectory
from.tokenization_tests_commons import CommonTestCases
class TransfoXLTokenizationTest(unittest.TestCase):
class TransfoXLTokenizationTest(CommonTestCases.CommonTokenizerTester):
tokenizer_class = TransfoXLTokenizer
def setUp(self):
super(TransfoXLTokenizationTest, self).setUp()
def test_full_tokenizer(self):
vocab_tokens = [
"<unk>", "[CLS]", "[SEP]", "want", "unwanted", "wa", "un",
"running", ",", "low", "l",
]
with TemporaryDirectory() as tmpdirname:
vocab_file = os.path.join(tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
with open(vocab_file, "w", encoding='utf-8') as vocab_writer:
vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
with open(self.vocab_file, "w", encoding='utf-8') as vocab_writer:
vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
input_text = u"<unk> UNwanted , running"
output_text = u"<unk> unwanted, running"
def get_tokenizer(self):
return TransfoXLTokenizer.from_pretrained(self.tmpdirname, lower_case=True)
create_and_check_tokenizer_commons(self, input_text, output_text, TransfoXLTokenizer, tmpdirname, lower_case=True)
def get_input_output_texts(self):
input_text = u"<unk> UNwanted , running"
output_text = u"<unk> unwanted, running"
return input_text, output_text
tokenizer = TransfoXLTokenizer(vocab_file=vocab_file, lower_case=True)
def test_full_tokenizer(self):
tokenizer = TransfoXLTokenizer(vocab_file=self.vocab_file, lower_case=True)
tokens = tokenizer.tokenize(u"<unk> UNwanted , running")
self.assertListEqual(tokens, ["<unk>", "unwanted", ",", "running"])
tokens = tokenizer.tokenize(u"<unk> UNwanted , running")
self.assertListEqual(tokens, ["<unk>", "unwanted", ",", "running"])
self.assertListEqual(
tokenizer.convert_tokens_to_ids(tokens), [0, 4, 8, 7])
self.assertListEqual(
tokenizer.convert_tokens_to_ids(tokens), [0, 4, 8, 7])
def test_full_tokenizer_lower(self):
tokenizer = TransfoXLTokenizer(lower_case=True)

54
pytorch_transformers/tests/tokenization_xlm_test.py
View File

@@ -20,12 +20,16 @@ import json
from pytorch_transformers.tokenization_xlm import XLMTokenizer, VOCAB_FILES_NAMES
from .tokenization_tests_commons import create_and_check_tokenizer_commons, TemporaryDirectory
from .tokenization_tests_commons import CommonTestCases
class XLMTokenizationTest(unittest.TestCase):
class XLMTokenizationTest(CommonTestCases.CommonTokenizerTester):
def test_full_tokenizer(self):
""" Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt """
tokenizer_class = XLMTokenizer
def setUp(self):
super(XLMTokenizationTest, self).setUp()
# Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt
vocab = ["l", "o", "w", "e", "r", "s", "t", "i", "d", "n",
"w</w>", "r</w>", "t</w>",
"lo", "low", "er</w>",
@@ -33,30 +37,34 @@ class XLMTokenizationTest(unittest.TestCase):
vocab_tokens = dict(zip(vocab, range(len(vocab))))
merges = ["l o 123", "lo w 1456", "e r</w> 1789", ""]
with TemporaryDirectory() as tmpdirname:
vocab_file = os.path.join(tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
merges_file = os.path.join(tmpdirname, VOCAB_FILES_NAMES['merges_file'])
with open(vocab_file, "w") as fp:
fp.write(json.dumps(vocab_tokens))
with open(merges_file, "w") as fp:
fp.write("\n".join(merges))
self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
self.merges_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['merges_file'])
with open(self.vocab_file, "w") as fp:
fp.write(json.dumps(vocab_tokens))
with open(self.merges_file, "w") as fp:
fp.write("\n".join(merges))
input_text = u"lower newer"
output_text = u"lower newer"
def get_tokenizer(self):
return XLMTokenizer.from_pretrained(self.tmpdirname)
create_and_check_tokenizer_commons(self, input_text, output_text, XLMTokenizer, tmpdirname)
def get_input_output_texts(self):
input_text = u"lower newer"
output_text = u"lower newer"
return input_text, output_text
tokenizer = XLMTokenizer(vocab_file, merges_file)
def test_full_tokenizer(self):
""" Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt """
tokenizer = XLMTokenizer(self.vocab_file, self.merges_file)
text = "lower"
bpe_tokens = ["low", "er</w>"]
tokens = tokenizer.tokenize(text)
self.assertListEqual(tokens, bpe_tokens)
text = "lower"
bpe_tokens = ["low", "er</w>"]
tokens = tokenizer.tokenize(text)
self.assertListEqual(tokens, bpe_tokens)
input_tokens = tokens + ["<unk>"]
input_bpe_tokens = [14, 15, 20]
self.assertListEqual(
tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
input_tokens = tokens + ["<unk>"]
input_bpe_tokens = [14, 15, 20]
self.assertListEqual(
tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
if __name__ == '__main__':

72
pytorch_transformers/tests/tokenization_xlnet_test.py
View File

@@ -19,48 +19,58 @@ import unittest
from pytorch_transformers.tokenization_xlnet import (XLNetTokenizer, SPIECE_UNDERLINE)
from .tokenization_tests_commons import create_and_check_tokenizer_commons, TemporaryDirectory
from .tokenization_tests_commons import CommonTestCases
SAMPLE_VOCAB = os.path.join(os.path.dirname(os.path.abspath(__file__)),
'fixtures/test_sentencepiece.model')
class XLNetTokenizationTest(unittest.TestCase):
class XLNetTokenizationTest(CommonTestCases.CommonTokenizerTester):
tokenizer_class = XLNetTokenizer
def setUp(self):
super(XLNetTokenizationTest, self).setUp()
# We have a SentencePiece fixture for testing
tokenizer = XLNetTokenizer(SAMPLE_VOCAB, keep_accents=True)
tokenizer.save_pretrained(self.tmpdirname)
def get_tokenizer(self):
return XLNetTokenizer.from_pretrained(self.tmpdirname)
def get_input_output_texts(self):
input_text = u"This is a test"
output_text = u"This is a test"
return input_text, output_text
def test_full_tokenizer(self):
tokenizer = XLNetTokenizer(SAMPLE_VOCAB, keep_accents=True)
with TemporaryDirectory() as tmpdirname:
tokenizer.save_pretrained(tmpdirname)
tokens = tokenizer.tokenize(u'This is a test')
self.assertListEqual(tokens, [u'▁This', u'▁is', u'▁a', u'▁t', u'est'])
input_text = u"This is a test"
output_text = u"This is a test"
self.assertListEqual(
tokenizer.convert_tokens_to_ids(tokens), [285, 46, 10, 170, 382])
create_and_check_tokenizer_commons(self, input_text, output_text, XLNetTokenizer, tmpdirname)
tokens = tokenizer.tokenize(u"I was born in 92000, and this is falsé.")
self.assertListEqual(tokens, [SPIECE_UNDERLINE + u'I', SPIECE_UNDERLINE + u'was', SPIECE_UNDERLINE + u'b',
u'or', u'n', SPIECE_UNDERLINE + u'in', SPIECE_UNDERLINE + u'',
u'9', u'2', u'0', u'0', u'0', u',', SPIECE_UNDERLINE + u'and', SPIECE_UNDERLINE + u'this',
SPIECE_UNDERLINE + u'is', SPIECE_UNDERLINE + u'f', u'al', u's', u'é', u'.'])
ids = tokenizer.convert_tokens_to_ids(tokens)
self.assertListEqual(
ids, [8, 21, 84, 55, 24, 19, 7, 0,
602, 347, 347, 347, 3, 12, 66,
46, 72, 80, 6, 0, 4])
tokens = tokenizer.tokenize(u'This is a test')
self.assertListEqual(tokens, [u'▁This', u'▁is', u'▁a', u'▁t', u'est'])
self.assertListEqual(
tokenizer.convert_tokens_to_ids(tokens), [285, 46, 10, 170, 382])
tokens = tokenizer.tokenize(u"I was born in 92000, and this is falsé.")
self.assertListEqual(tokens, [SPIECE_UNDERLINE + u'I', SPIECE_UNDERLINE + u'was', SPIECE_UNDERLINE + u'b',
u'or', u'n', SPIECE_UNDERLINE + u'in', SPIECE_UNDERLINE + u'',
u'9', u'2', u'0', u'0', u'0', u',', SPIECE_UNDERLINE + u'and', SPIECE_UNDERLINE + u'this',
SPIECE_UNDERLINE + u'is', SPIECE_UNDERLINE + u'f', u'al', u's', u'é', u'.'])
ids = tokenizer.convert_tokens_to_ids(tokens)
self.assertListEqual(
ids, [8, 21, 84, 55, 24, 19, 7, 0,
602, 347, 347, 347, 3, 12, 66,
46, 72, 80, 6, 0, 4])
back_tokens = tokenizer.convert_ids_to_tokens(ids)
self.assertListEqual(back_tokens, [SPIECE_UNDERLINE + u'I', SPIECE_UNDERLINE + u'was', SPIECE_UNDERLINE + u'b',
u'or', u'n', SPIECE_UNDERLINE + u'in',
SPIECE_UNDERLINE + u'', u'<unk>', u'2', u'0', u'0', u'0', u',',
SPIECE_UNDERLINE + u'and', SPIECE_UNDERLINE + u'this',
SPIECE_UNDERLINE + u'is', SPIECE_UNDERLINE + u'f', u'al', u's',
u'<unk>', u'.'])
back_tokens = tokenizer.convert_ids_to_tokens(ids)
self.assertListEqual(back_tokens, [SPIECE_UNDERLINE + u'I', SPIECE_UNDERLINE + u'was', SPIECE_UNDERLINE + u'b',
u'or', u'n', SPIECE_UNDERLINE + u'in',
SPIECE_UNDERLINE + u'', u'<unk>', u'2', u'0', u'0', u'0', u',',
SPIECE_UNDERLINE + u'and', SPIECE_UNDERLINE + u'this',
SPIECE_UNDERLINE + u'is', SPIECE_UNDERLINE + u'f', u'al', u's',
u'<unk>', u'.'])
def test_tokenizer_lower(self):
tokenizer = XLNetTokenizer(SAMPLE_VOCAB, do_lower_case=True)

100
pytorch_transformers/tokenization_auto.py Normal file
View File

@@ -0,0 +1,100 @@
# coding=utf-8
# Copyright 2018 The HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Auto Model class. """
from __future__ import absolute_import, division, print_function, unicode_literals
import logging
from .tokenization_bert import BertTokenizer
from .tokenization_openai import OpenAIGPTTokenizer
from .tokenization_gpt2 import GPT2Tokenizer
from .tokenization_transfo_xl import TransfoXLTokenizer
from .tokenization_xlnet import XLNetTokenizer
from .tokenization_xlm import XLMTokenizer
logger = logging.getLogger(__name__)
class AutoTokenizer(object):
r""":class:`~pytorch_transformers.AutoTokenizer` is a generic tokenizer class
that will be instantiated as one of the tokenizer classes of the library
when created with the `AutoTokenizer.from_pretrained(pretrained_model_name_or_path)`
class method.
The `from_pretrained()` method take care of returning the correct tokenizer class instance
using pattern matching on the `pretrained_model_name_or_path` string.
The tokenizer class to instantiate is selected as the first pattern matching
in the `pretrained_model_name_or_path` string (in the following order):
- contains `bert`: BertTokenizer (Bert model)
- contains `openai-gpt`: OpenAIGPTTokenizer (OpenAI GPT model)
- contains `gpt2`: GPT2Tokenizer (OpenAI GPT-2 model)
- contains `transfo-xl`: TransfoXLTokenizer (Transformer-XL model)
- contains `xlnet`: XLNetTokenizer (XLNet model)
- contains `xlm`: XLMTokenizer (XLM model)
This class cannot be instantiated using `__init__()` (throw an error).
"""
def __init__(self):
raise EnvironmentError("AutoTokenizer is designed to be instantiated "
"using the `AutoTokenizer.from_pretrained(pretrained_model_name_or_path)` method.")
@classmethod
def from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs):
r""" Instantiate a one of the tokenizer classes of the library
from a pre-trained model vocabulary.
The tokenizer class to instantiate is selected as the first pattern matching
in the `pretrained_model_name_or_path` string (in the following order):
- contains `bert`: BertTokenizer (Bert model)
- contains `openai-gpt`: OpenAIGPTTokenizer (OpenAI GPT model)
- contains `gpt2`: GPT2Tokenizer (OpenAI GPT-2 model)
- contains `transfo-xl`: TransfoXLTokenizer (Transformer-XL model)
- contains `xlnet`: XLNetTokenizer (XLNet model)
- contains `xlm`: XLMTokenizer (XLM model)
Params:
**pretrained_model_name_or_path**: either:
- a string with the `shortcut name` of a pre-trained model configuration to load from cache
or download and cache if not already stored in cache (e.g. 'bert-base-uncased').
- a path to a `directory` containing a configuration file saved
using the `save_pretrained(save_directory)` method.
- a path or url to a saved configuration `file`.
**cache_dir**: (`optional`) string:
Path to a directory in which a downloaded pre-trained model
configuration should be cached if the standard cache should not be used.
Examples::
config = AutoTokenizer.from_pretrained('bert-base-uncased') # Download vocabulary from S3 and cache.
config = AutoTokenizer.from_pretrained('./test/bert_saved_model/') # E.g. tokenizer was saved using `save_pretrained('./test/saved_model/')`
"""
if 'bert' in pretrained_model_name_or_path:
return BertTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
elif 'openai-gpt' in pretrained_model_name_or_path:
return OpenAIGPTTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
elif 'gpt2' in pretrained_model_name_or_path:
return GPT2Tokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
elif 'transfo-xl' in pretrained_model_name_or_path:
return TransfoXLTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
elif 'xlnet' in pretrained_model_name_or_path:
return XLNetTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
elif 'xlm' in pretrained_model_name_or_path:
return XLMTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
raise ValueError("Unrecognized model identifier in {}. Should contains one of "
"'bert', 'openai-gpt', 'gpt2', 'transfo-xl', 'xlnet', "
"'xlm'".format(pretrained_model_name_or_path))

8
pytorch_transformers/tokenization_bert.py
View File

@@ -22,7 +22,7 @@ import os
import unicodedata
from io import open
from .tokenization_utils import PreTrainedTokenizer, clean_up_tokenization
from .tokenization_utils import PreTrainedTokenizer
logger = logging.getLogger(__name__)
@@ -86,7 +86,7 @@ def whitespace_tokenize(text):
class BertTokenizer(PreTrainedTokenizer):
r"""
Constructs a BertTokenizer.
:class:`~pytorch_pretrained_bert.BertTokenizer` runs end-to-end tokenization: punctuation splitting + wordpiece
:class:`~pytorch_transformers.BertTokenizer` runs end-to-end tokenization: punctuation splitting + wordpiece
Args:
vocab_file: Path to a one-wordpiece-per-line vocabulary file
@@ -119,7 +119,7 @@ class BertTokenizer(PreTrainedTokenizer):
Only has an effect when do_basic_tokenize=True
**tokenize_chinese_chars**: (`optional`) boolean (default True)
Whether to tokenize Chinese characters.
This should likely be desactivated for Japanese:
This should likely be deactivated for Japanese:
see: https://github.com/huggingface/pytorch-pretrained-BERT/issues/328
"""
super(BertTokenizer, self).__init__(unk_token=unk_token, sep_token=sep_token,
@@ -214,7 +214,7 @@ class BasicTokenizer(object):
List of token not to split.
**tokenize_chinese_chars**: (`optional`) boolean (default True)
Whether to tokenize Chinese characters.
This should likely be desactivated for Japanese:
This should likely be deactivated for Japanese:
see: https://github.com/huggingface/pytorch-pretrained-BERT/issues/328
"""
if never_split is None:

10
pytorch_transformers/tokenization_gpt2.py
View File

@@ -31,7 +31,7 @@ except ImportError:
def lru_cache():
return lambda func: func
from .tokenization_utils import PreTrainedTokenizer, clean_up_tokenization
from .tokenization_utils import PreTrainedTokenizer
logger = logging.getLogger(__name__)
@@ -102,9 +102,9 @@ class GPT2Tokenizer(PreTrainedTokenizer):
pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
def __init__(self, vocab_file, merges_file, errors='replace',
def __init__(self, vocab_file, merges_file, errors='replace', unk_token="<|endoftext|>",
bos_token="<|endoftext|>", eos_token="<|endoftext|>", **kwargs):
super(GPT2Tokenizer, self).__init__(bos_token=bos_token, eos_token=eos_token, **kwargs)
super(GPT2Tokenizer, self).__init__(bos_token=bos_token, eos_token=eos_token, unk_token=unk_token, **kwargs)
self.encoder = json.load(open(vocab_file))
self.decoder = {v:k for k,v in self.encoder.items()}
@@ -177,9 +177,7 @@ class GPT2Tokenizer(PreTrainedTokenizer):
def _convert_token_to_id(self, token):
""" Converts a token (str/unicode) in an id using the vocab. """
if token in self.encoder:
return self.encoder.get(token)
return self.encoder.get(self.unk_token)
return self.encoder.get(token, self.encoder.get(self.unk_token))
def _convert_id_to_token(self, index):
"""Converts an index (integer) in a token (string/unicode) using the vocab."""

2
pytorch_transformers/tokenization_transfo_xl.py
View File

@@ -30,7 +30,7 @@ import torch
import numpy as np
from .file_utils import cached_path
from .tokenization_utils import PreTrainedTokenizer, clean_up_tokenization
from .tokenization_utils import PreTrainedTokenizer
if sys.version_info[0] == 2:
import cPickle as pickle

193
pytorch_transformers/tokenization_utils.py
View File

@@ -30,14 +30,34 @@ SPECIAL_TOKENS_MAP_FILE = 'special_tokens_map.json'
ADDED_TOKENS_FILE = 'added_tokens.json'
class PreTrainedTokenizer(object):
""" 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:
bos_token, eos_token, EOP_TOKEN, EOD_TOKEN, unk_token, sep_token, pad_token, cls_token, mask_token
additional_special_tokens = []
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
specific vocabulary augmentation methods of the various underlying dictionnary structures (BPE, sentencepiece...).
Class attributes (overridden by derived classes):
- ``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 = {}
pretrained_vocab_files_map = {}
@@ -49,48 +69,56 @@ class PreTrainedTokenizer(object):
@property
def bos_token(self):
""" Beginning of sentence token (string). Log an error if used while not having been set. """
if self._bos_token is None:
logger.error("Using bos_token, but it is not set yet.")
return self._bos_token
@property
def eos_token(self):
""" End of sentence token (string). Log an error if used while not having been set. """
if self._eos_token is None:
logger.error("Using eos_token, but it is not set yet.")
return self._eos_token
@property
def unk_token(self):
""" Unknown token (string). Log an error if used while not having been set. """
if self._unk_token is None:
logger.error("Using unk_token, but it is not set yet.")
return self._unk_token
@property
def sep_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. """
if self._sep_token is None:
logger.error("Using sep_token, but it is not set yet.")
return self._sep_token
@property
def pad_token(self):
""" Padding token (string). Log an error if used while not having been set. """
if self._pad_token is None:
logger.error("Using pad_token, but it is not set yet.")
return self._pad_token
@property
def cls_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. """
if self._cls_token is None:
logger.error("Using cls_token, but it is not set yet.")
return self._cls_token
@property
def mask_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. """
if self._mask_token is None:
logger.error("Using mask_token, but it is not set yet.")
return self._mask_token
@property
def additional_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. """
if self._additional_special_tokens is None:
logger.error("Using additional_special_tokens, but it is not set yet.")
return self._additional_special_tokens
@@ -143,20 +171,58 @@ class PreTrainedTokenizer(object):
for key, value in kwargs.items():
if key in self.SPECIAL_TOKENS_ATTRIBUTES:
if key == 'additional_special_tokens':
assert isinstance(value, (list, tuple)) and all(isinstance(t, str) or (six.PY2 and isinstance(t, unicode)) for t in value)
else:
assert isinstance(value, str) or (six.PY2 and isinstance(value, unicode))
setattr(self, key, value)
@classmethod
def from_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
# Download vocabulary from S3 and cache.
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
# If vocabulary files are in a directory (e.g. tokenizer was saved using `save_pretrained('./test/saved_model/')`)
tokenizer = BertTokenizer.from_pretrained('./test/saved_model/')
# If the tokenizer uses a single vocabulary file, you can point directly to this file
tokenizer = BertTokenizer.from_pretrained('./test/saved_model/my_vocab.txt')
# You can link tokens to special vocabulary when instantiating
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased', unk_token='<unk>')
# You should be sure '<unk>' is in the vocabulary when doing that.
# Otherwise use tokenizer.add_special_tokens({'unk_token': '<unk>'}) instead)
assert tokenizer.unk_token == '<unk>'
"""
return cls._from_pretrained(*inputs, **kwargs)
@classmethod
def _from_pretrained(cls, pretrained_model_name_or_path, cache_dir=None, *inputs, **kwargs):
"""
Instantiate a PreTrainedTokenizer from pre-trained vocabulary files.
Download and cache the vocabulary files if needed.
"""
def _from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs):
cache_dir = kwargs.pop('cache_dir', None)
s3_models = list(cls.max_model_input_sizes.keys())
vocab_files = {}
if pretrained_model_name_or_path in s3_models:
@@ -271,8 +337,9 @@ class PreTrainedTokenizer(object):
def save_pretrained(self, save_directory):
""" Save the tokenizer vocabulary files (with added tokens) and the
special-tokens-to-class-attributes-mapping to a directory, so that it
can be re-loaded using the `from_pretrained(save_directory)` class method.
special-tokens-to-class-attributes-mapping to a directory.
This method make sure the full tokenizer can then be re-loaded using the :func:`~pytorch_transformers.PreTrainedTokenizer.from_pretrained` class method.
"""
if not os.path.isdir(save_directory):
logger.error("Saving directory ({}) should be a directory".format(save_directory))
@@ -297,38 +364,52 @@ class PreTrainedTokenizer(object):
def save_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.
Please use `save_pretrained()` to save the full Tokenizer state so that it can be
reloaded using the `from_pretrained(save_directory)` class method.
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.
"""
raise NotImplementedError
def vocab_size(self):
""" Size of the base vocabulary (without the added tokens) """
raise NotImplementedError
def __len__(self):
""" Size of the full vocabulary with the added tokens """
return self.vocab_size + len(self.added_tokens_encoder)
def add_tokens(self, new_tokens):
""" 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
the last index of the current vocabulary.
vocabulary, they are added to it with indices starting from length 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:
Number of tokens added to the vocabulary which can be used to correspondingly
increase the size of the associated model embedding matrices.
Number of tokens added to the vocabulary.
Examples::
# Let's see how to increase the vocabulary of Bert model and tokenizer
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertModel.from_pretrained('bert-base-uncased')
num_added_toks = tokenizer.add_tokens(['new_tok1', 'my_new-tok2'])
print('We have added', num_added_toks, 'tokens')
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.
"""
if not new_tokens:
return 0
to_add_tokens = []
for token in new_tokens:
if self.convert_tokens_to_ids(token) == self.convert_tokens_to_ids(self.unk_token):
assert isinstance(token, str) or (six.PY2 and isinstance(token, unicode))
if token != self.unk_token and \
self.convert_tokens_to_ids(token) == self.convert_tokens_to_ids(self.unk_token):
to_add_tokens.append(token)
logger.info("Adding %s to the vocabulary", token)
@@ -341,24 +422,48 @@ class PreTrainedTokenizer(object):
def add_special_tokens(self, special_tokens_dict):
""" Add a dictionnary 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 it and indexed starting from the last index of the current vocabulary.
""" Add a dictionary of special tokens (eos, pad, cls...) to the encoder and link them
to class attributes. If special tokens are NOT in the vocabulary, they are added
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.
Number of tokens added to the vocabulary.
Examples::
# Let's see how to add a new classification token to GPT-2
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
model = GPT2Model.from_pretrained('gpt2')
special_tokens_dict = {'cls_token': '<CLS>'}
num_added_toks = tokenizer.add_special_tokens(special_tokens_dict)
print('We have added', num_added_toks, 'tokens')
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.
assert tokenizer.cls_token == '<CLS>'
"""
if not special_tokens_dict:
return 0
added_special_tokens = self.add_tokens(special_tokens_dict.values())
added_tokens = 0
for key, value in special_tokens_dict.items():
assert key in self.SPECIAL_TOKENS_ATTRIBUTES
if key == 'additional_special_tokens':
assert isinstance(value, (list, tuple)) and all(isinstance(t, str) or (six.PY2 and isinstance(t, unicode)) for t in value)
added_tokens += self.add_tokens(value)
else:
assert isinstance(value, str) or (six.PY2 and isinstance(value, unicode))
added_tokens += self.add_tokens([value])
logger.info("Assigning %s to the %s key of the tokenizer", value, key)
setattr(self, key, value)
return added_special_tokens
return added_tokens
def tokenize(self, text, **kwargs):
""" Converts a string in a sequence of tokens (string), using the tokenizer.
@@ -386,13 +491,13 @@ class PreTrainedTokenizer(object):
Split in words for word-based vocabulary or sub-words for sub-word-based
vocabularies (BPE/SentencePieces/WordPieces).
Don't take care of added tokens.
Do NOT take care of added tokens.
"""
raise NotImplementedError
def convert_tokens_to_ids(self, tokens):
""" Converts a single token or a sequence of tokens (str/unicode) in a integer id
(resp.) a sequence of ids, using the vocabulary.
""" Converts a single token, or a sequence of tokens, (str/unicode) in a single integer id
(resp. a sequence of ids), using the vocabulary.
"""
if isinstance(tokens, str) or (six.PY2 and isinstance(tokens, unicode)):
return self._convert_token_to_id_with_added_voc(tokens)
@@ -417,7 +522,8 @@ class PreTrainedTokenizer(object):
def encode(self, text):
""" Converts a string in a sequence of ids (integer), using the tokenizer and vocabulary.
same as self.convert_tokens_to_ids(self.tokenize(text)).
Same doing ``self.convert_tokens_to_ids(self.tokenize(text))``.
"""
return self.convert_tokens_to_ids(self.tokenize(text))
@@ -457,11 +563,13 @@ class PreTrainedTokenizer(object):
def decode(self, token_ids, skip_special_tokens=False, clean_up_tokenization_spaces=True):
""" Converts a sequence of ids (integer) in a string, using the tokenizer and vocabulary
with options to remove special tokens and clean up tokenization spaces.
Similar to doing ``self.convert_tokens_to_string(self.convert_ids_to_tokens(token_ids))``.
"""
filtered_tokens = self.convert_ids_to_tokens(token_ids, skip_special_tokens=skip_special_tokens)
text = self.convert_tokens_to_string(filtered_tokens)
if clean_up_tokenization_spaces:
text = clean_up_tokenization(text)
text = self.clean_up_tokenization(text)
return text
@property
@@ -497,10 +605,11 @@ class PreTrainedTokenizer(object):
all_ids = list(self.convert_tokens_to_ids(t) for t in all_toks)
return all_ids
def clean_up_tokenization(out_string):
out_string = out_string.replace(' .', '.').replace(' ?', '?').replace(' !', '!').replace(' ,', ','
).replace(" ' ", "'").replace(" n't", "n't").replace(" 'm", "'m").replace(" do not", " don't"
).replace(" 's", "'s").replace(" 've", "'ve").replace(" 're", "'re")
return out_string
@staticmethod
def clean_up_tokenization(out_string):
""" Clean up a list of simple English tokenization artifacts like spaces before punctuations and abreviated forms.
"""
out_string = out_string.replace(' .', '.').replace(' ?', '?').replace(' !', '!').replace(' ,', ','
).replace(" ' ", "'").replace(" n't", "n't").replace(" 'm", "'m").replace(" do not", " don't"
).replace(" 's", "'s").replace(" 've", "'ve").replace(" 're", "'re")
return out_string

2
pytorch_transformers/tokenization_xlnet.py
View File

@@ -23,7 +23,7 @@ from shutil import copyfile
import unicodedata
import six
from .tokenization_utils import PreTrainedTokenizer, clean_up_tokenization
from .tokenization_utils import PreTrainedTokenizer
logger = logging.getLogger(__name__)