spaCy pipelines for pre-trained BERT and other transformers

# spaCy wrapper for PyTorch Transformers

This package provides spaCy model pipelines that wrap Hugging Face's pytorch-transformers package, so you can use them in spaCy. The result is convenient access to state-of-the-art transformer architectures, such as BERT, GPT-2, XLNet, etc. For more details and background, check out our blog post.

## Features

• Use BERT, RoBERTa, XLNet and GPT-2 directly in your spaCy pipeline.
• Custom component for text classification using transformer features.
• Automatic alignment of wordpieces and outputs to linguistic tokens.
• Process multi-sentence documents with intelligent per-sentence prediction.
• Built-in hooks for context-sensitive vectors and similarity.
• Out-of-the-box serialization and model packaging.

## 🚀 Quickstart

Installing the package from pip will automatically install all dependencies, including PyTorch and spaCy. Make sure you install this package before you install the models. Also note that this package requires Python 3.6+ and the latest version of spaCy, v2.1.7 or above.

pip install spacy-pytorch-transformers


For GPU installation, find your CUDA version using nvcc --version and add the version in brackets, e.g. spacy-pytorch-transformers[cuda92] for CUDA9.2 or spacy-pytorch-transformers[cuda100] for CUDA10.0.

We've also pre-packaged some of the pretrained models as spaCy model packages. You can either use the spacy download command or download the packages from the model releases.

Package name Pretrained model Language Author Size Release
en_pytt_bertbaseuncased_lg bert-base-uncased English Google Research 406MB 📦️
de_pytt_bertbasecased_lg bert-base-german-cased German deepset 406MB 📦️
en_pytt_xlnetbasecased_lg xlnet-base-cased English CMU/Google Brain 434MB 📦️
en_pytt_robertabase_lg roberta-base English Facebook 292MB 📦️
en_pytt_distilbertbaseuncased_lg distilbert-base-uncased English Hugging Face 245MB 📦️
python -m spacy download en_pytt_bertbaseuncased_lg


Once the model is installed, you can load it in spaCy like any other model package.

import spacy

doc = nlp("Apple shares rose on the news. Apple pie is delicious.")
print(doc[0].similarity(doc[7]))
print(doc._.pytt_last_hidden_state.shape)


💡 If you're seeing an error like No module named 'spacy.lang.pytt', double-check that spacy-pytorch-transformers is installed. It needs to be available so it can register its language entry points. Also make sure that you're running spaCy v2.1.7 or higher.

## 📖 Usage

### Transfer learning

The main use case for pretrained transformer models is transfer learning. You load in a large generic model pretrained on lots of text, and start training on your smaller dataset with labels specific to your problem. This package has custom pipeline components that make this especially easy. We provide an example component for text categorization. Development of analogous components for other tasks should be quite straight-forward.

The pytt_textcat component is based on spaCy's built-in TextCategorizer and supports using the features assigned by the PyTorch-Transformers models, via the pytt_tok2vec component. This lets you use a model like BERT to predict contextual token representations, and then learn a text categorizer on top as a task-specific "head". The API is the same as any other spaCy pipeline:

TRAIN_DATA = [
("text1", {"cats": {"POSITIVE": 1.0, "NEGATIVE": 0.0}})
]

import spacy
from spacy.util import minibatch
import random
import torch

is_using_gpu = spacy.prefer_gpu()
if is_using_gpu:
torch.set_default_tensor_type("torch.cuda.FloatTensor")

print(nlp.pipe_names) # ["sentencizer", "pytt_wordpiecer", "pytt_tok2vec"]
textcat = nlp.create_pipe("pytt_textcat", config={"exclusive_classes": True})
for label in ("POSITIVE", "NEGATIVE"):

optimizer = nlp.resume_training()
for i in range(10):
random.shuffle(TRAIN_DATA)
losses = {}
for batch in minibatch(TRAIN_DATA, size=8):
texts, cats = zip(*batch)
nlp.update(texts, cats, sgd=optimizer, losses=losses)
print(i, losses)
nlp.to_disk("/bert-textcat")


For a full example, see the examples/train_textcat.py script.

### Vectors and similarity

The PyTT_TokenVectorEncoder component of the model sets custom hooks that override the default behaviour of the .vector attribute and .similarity method of the Token, Span and Doc objects. By default, these usually refer to the word vectors table at nlp.vocab.vectors. Naturally, in the transformer models we'd rather use the doc.tensor attribute, since it holds a much more informative context-sensitive representation.

apple1 = nlp("Apple shares rose on the news.")
apple2 = nlp("Apple sold fewer iPhones this quarter.")
apple3 = nlp("Apple pie is delicious.")
print(apple1[0].similarity(apple2[0]))
print(apple1[0].similarity(apple3[0]))


### Serialization

Saving and loading pretrained transformer models and packaging them as spaCy models ✨just works ✨ (at least, it should). The wrapper and components follow spaCy's API, so when you save and load the nlp object, it...

• Writes the pretrained weights to disk / bytes and loads them back in.
• Adds "lang_factory": "pytt" in the meta.json so spaCy knows how to initialize the Language class when you load the model.
• Adds this package and its version to the "requirements" in the meta.json, so when you run spacy package to create an installable Python package it's automatically added to the setup's install_requires.

For example, if you've trained your own text classifier, you can package it like this:

python -m spacy package /bert-textcat /output
cd /output/en_pytt_bertbaseuncased_lg-1.0.0
python setup.py sdist
pip install dist/en_pytt_bertbaseuncased_lg-1.0.0.tar.gz


### Extension attributes

This wrapper sets the following custom extension attributes on the Doc, Span and Token objects:

Name Type Description
._.pytt_alignment List[List[int]] Alignment between wordpieces and spaCy tokens. Contains lists of wordpiece token indices (one per spaCy token) or a list of indices (if called on a Token).
._.pytt_word_pieces List[int] The wordpiece IDs.
._.pytt_word_pieces_ List[str] The string forms of the wordpiece IDs.
._.pytt_last_hidden_state ndarray The last_hidden_state output from the PyTorch-Transformers model.
._.pytt_pooler_output List[ndarray] The pooler_output output from the PyTorch-Transformers model.
._.pytt_all_hidden_states List[ndarray] The all_hidden_states output from the PyTorch-Transformers model.
._.all_attentions List[ndarray] The all_attentions output from the PyTorch-Transformers model.
._.pytt_d_last_hidden_state ndarray The gradient of the last_hidden_state output from the PyTorch-Transformers model.
._.pytt_d_pooler_output List[ndarray] The gradient of the pooler_output output from the PyTorch-Transformers model.
._.pytt_d_all_hidden_states List[ndarray] The gradient of the all_hidden_states output from the PyTorch-Transformers model.
._.pytt_d_all_attentions List[ndarray] The gradient of the all_attentions output from the PyTorch-Transformers model.

The values can be accessed via the ._ attribute. For example:

doc = nlp("This is a text.")
print(doc._.pytt_word_pieces_)


### Setting up the pipeline

In order to run, the nlp object created using PyTT_Language requires a few components to run in order: a component that assigns sentence boundaries (e.g. spaCy's built-in Sentencizer), the PyTT_WordPiecer, which assigns the wordpiece tokens and the PyTT_TokenVectorEncoder, which assigns the token vectors. The pytt_name argument defines the name of the pretrained model to use. The from_pretrained methods load the pretrained model via pytorch-transformers.

from spacy_pytorch_transformers import PyTT_Language, PyTT_WordPiecer, PyTT_TokenVectorEncoder

name = "bert-base-uncased"
nlp = PyTT_Language(pytt_name=name, meta={"lang": "en"})
print(nlp.pipe_names)  # ['sentencizer', 'pytt_wordpiecer', 'pytt_tok2vec']


You can also use the init_model.py script in the examples.

Pytorch-Transformers models can also be loaded from a file path instead of just a name. For instance, let's say you want to use Allen AI's scibert. First, download the PyTorch model files, unpack them them, unpack the weights.tar, rename the bert_config.json to config.json and put everything into one directory. Your directory should now have a pytorch_model.bin, vocab.txt and config.json. Also make sure that your path includes the name of the model. You can then initialize the nlp object like this:

from spacy_pytorch_transformers import PyTT_Language, PyTT_WordPiecer, PyTT_TokenVectorEncoder

name = "scibert-scivocab-uncased"
path = "/path/to/scibert-scivocab-uncased"

nlp = PyTT_Language(pytt_name=name, meta={"lang": "en"})


### Tokenization alignment

Transformer models are usually trained on text preprocessed with the "wordpiece" algorithm, which limits the number of distinct token-types the model needs to consider. Wordpiece is convenient for training neural networks, but it doesn't produce segmentations that match up to any linguistic notion of a "word". Most rare words will map to multiple wordpiece tokens, and occasionally the alignment will be many-to-many. spacy-pytorch-transformers calculates this alignment, which you can access at doc._.pytt_alignment. It's a list of length equal to the number of spaCy tokens. Each value in the list is a list of consecutive integers, which are indexes into the wordpieces list.

If you can work on representations that aren't aligned to actual words, it's best to use the raw outputs of the transformer, which can be accessed at doc._.pytt_last_hidden_state. This variable gives you a tensor with one row per wordpiece token.

If you're working on token-level tasks such as part-of-speech tagging or spelling correction, you'll want to work on the token-aligned features, which are stored in the doc.tensor variable.

We've taken care to calculate the aligned doc.tensor representation as faithfully as possible, with priority given to avoid information loss. The alignment has been calculated such that doc.tensor.sum(axis=1) == doc._.pytt_last_hidden_state.sum(axis=1). To make this work, each row of the doc.tensor (which corresponds to a spaCy token) is set to a weighted sum of the rows of the last_hidden_state tensor that the token is aligned to, where the weighting is proportional to the number of other spaCy tokens aligned to that row. To include the information from the (often important --- see Clark et al., 2019) boundary tokens, we imagine that these are also "aligned" to all of the tokens in the sentence.

### Batching, padding and per-sentence processing

Transformer models have cubic runtime and memory complexity with respect to sequence length. This means that longer texts need to be divided into sentences in order to achieve reasonable efficiency.

spacy-pytorch-transformers handles this internally, and requires that sort of sentence-boundary detection component has been added to the pipeline. We recommend:

sentencizer = nlp.create_pipe("sentencizer")


Internally, the transformer model will predict over sentences, and the resulting tensor features will be reconstructed to produce document-level annotations.

In order to further improve efficiency and reduce memory requirements, spacy-pytorch-transformers also performs length-based subbatching internally. The subbatching regroups the batched sentences by sequence length, to minimise the amount of padding required. The configuration option words_per_batch controls this behaviour. You can set it to 0 to disable the subbatching, or set it to an integer to require a maximum limit on the number of words (including padding) per subbatch. The default value of 3000 words works reasonably well on a Tesla V100.

Many of the pretrained transformer models have a maximum sequence length. If a sentence is longer than the maximum, it is truncated and the affected ending tokens will receive zeroed vectors.

## 🎛 API

### classPyTT_Language

A subclass of Language that holds a PyTorch-Transformer (PyTT) pipeline. PyTT pipelines work only slightly differently from spaCy's default pipelines. Specifically, we introduce a new pipeline component at the start of the pipeline, PyTT_TokenVectorEncoder. We then modify the nlp.update function to run the PyTT_TokenVectorEncoder before the other pipeline components, and backprop it after the other components are done.

#### staticmethodPyTT_Language.install_extensions

Register the custom extension attributes on the Doc, Span and Token objects. If the extensions have already been registered, spaCy will raise an error. See here for the extension attributes that will be set. You shouldn't have to call this method yourself – it already runs when you import the package.

#### methodPyTT_Language.__init__

See Language.__init__. Expects either a pytt_name setting in the meta or as a keyword argument, specifying the pretrained model name. This is used to set up the model-specific tokenizer.

#### methodPyTT_Language.update

Update the models in the pipeline.

Name Type Description
docs iterable A batch of Doc objects or unicode. If unicode, a Doc object will be created from the text.
golds iterable A batch of GoldParse objects or dictionaries. Dictionaries will be used to create GoldParse objects.
drop float The dropout rate.
sgd callable An optimizer.
losses dict Dictionary to update with the loss, keyed by pipeline component.
component_cfg dict Config parameters for specific pipeline components, keyed by component name.

### classPyTT_WordPiecer

spaCy pipeline component to assign PyTorch-Transformers wordpiece tokenization to the Doc, which can then be used by the token vector encoder. Note that this component doesn't modify spaCy's tokenization. It only sets extension attributes pytt_word_pieces_, pytt_word_pieces and pytt_alignment (alignment between wordpiece tokens and spaCy tokens).

The component is available as pytt_wordpiecer and registered via an entry point, so it can also be created using nlp.create_pipe:

wordpiecer = nlp.create_pipe("pytt_wordpiecer")


#### Config

The component can be configured with the following settings, usually passed in as the **cfg.

Name Type Description
pytt_name unicode Name of pretrained model, e.g. "bert-base-uncased".

#### classmethodPyTT_WordPiecer.from_nlp

Factory to add to Language.factories via entry point.

Name Type Description
nlp spacy.language.Language The nlp object the component is created with.
**cfg - Optional config parameters.
RETURNS PyTT_WordPiecer The wordpiecer.

#### methodPyTT_WordPiecer.__init__

Initialize the component.

Name Type Description
vocab spacy.vocab.Vocab The spaCy vocab to use.
name unicode Name of pretrained model, e.g. "bert-base-uncased".
**cfg - Optional config parameters.
RETURNS PyTT_WordPiecer The wordpiecer.

#### methodPyTT_WordPiecer.predict

Run the wordpiece tokenizer on a batch of docs and return the extracted strings.

Name Type Description
docs iterable A batch of Docs to process.
RETURNS tuple A (strings, None) tuple. The strings are lists of strings, one list per Doc.

#### methodPyTT_WordPiecer.set_annotations

Assign the extracted tokens and IDs to the Doc objects.

Name Type Description
docs iterable A batch of Doc objects.
outputs iterable A batch of outputs.

### classPyTT_TokenVectorEncoder

spaCy pipeline component to use PyTorch-Transformers models. The component assigns the output of the transformer to extension attributes. We also calculate an alignment between the wordpiece tokens and the spaCy tokenization, so that we can use the last hidden states to set the doc.tensor attribute. When multiple wordpiece tokens align to the same spaCy token, the spaCy token receives the sum of their values.

The component is available as pytt_tok2vec and registered via an entry point, so it can also be created using nlp.create_pipe:

tok2vec = nlp.create_pipe("pytt_tok2vec")


#### Config

The component can be configured with the following settings, usually passed in as the **cfg.

Name Type Description
pytt_name unicode Name of pretrained model, e.g. "bert-base-uncased".
words_per_batch int Group sentences into subbatches of max words_per_batch in size. For instance, a batch with one 100 word sentence and one 10 word sentence will have size 200 (due to padding). Set to 0 to disable. Defaults to 2000.

#### classmethodPyTT_TokenVectorEncoder.from_nlp

Factory to add to Language.factories via entry point.

Name Type Description
nlp spacy.language.Language The nlp object the component is created with.
**cfg - Optional config parameters.
RETURNS PyTT_TokenVectorEncoder The token vector encoder.

#### classmethodPyTT_TokenVectorEncoder.from_pretrained

Create a PyTT_TokenVectorEncoder instance using pretrained weights from a PyTorch-Transformers model, even if it's not installed as a spaCy package.

from spacy_pytorch_transformers import PyTT_TokenVectorEncoder
from spacy.tokens import Vocab
tok2vec = PyTT_TokenVectorEncoder.from_pretrained(Vocab(), "bert-base-uncased")

Name Type Description
vocab spacy.vocab.Vocab The spaCy vocab to use.
name unicode Name of pretrained model, e.g. "bert-base-uncased".
**cfg - Optional config parameters.
RETURNS PyTT_TokenVectorEncoder The token vector encoder.

#### classmethodPyTT_TokenVectorEncoder.Model

Create an instance of PyTT_Wrapper, which holds the PyTorch-Transformers model.

Name Type Description
name unicode Name of pretrained model, e.g. "bert-base-uncased".
**cfg - Optional config parameters.
RETURNS thinc.neural.Model The wrapped model.

#### methodPyTT_TokenVectorEncoder.__init__

Initialize the component.

Name Type Description
vocab spacy.vocab.Vocab The spaCy vocab to use.
model thinc.neural.Model / True The component's model or True if not initialized yet.
**cfg - Optional config parameters.
RETURNS PyTT_TokenVectorEncoder The token vector encoder.

#### methodPyTT_TokenVectorEncoder.__call__

Process a Doc and assign the extracted features.

Name Type Description
doc spacy.tokens.Doc The Doc to process.
RETURNS spacy.tokens.Doc The processed Doc.

#### methodPyTT_TokenVectorEncoder.pipe

Process Doc objects as a stream and assign the extracted features.

Name Type Description
stream iterable A stream of Doc objects.
batch_size int The number of texts to buffer. Defaults to 128.
YIELDS spacy.tokens.Doc Processed Docs in order.

#### methodPyTT_TokenVectorEncoder.predict

Run the transformer model on a batch of docs and return the extracted features.

Name Type Description
docs iterable A batch of Docs to process.
RETURNS namedtuple Named tuple containing the outputs.

#### methodPyTT_TokenVectorEncoder.set_annotations

Assign the extracted features to the Doc objects and overwrite the vector and similarity hooks.

Name Type Description
docs iterable A batch of Doc objects.
outputs iterable A batch of outputs.

### classPyTT_TextCategorizer

Subclass of spaCy's built-in TextCategorizer component that supports using the features assigned by the PyTorch-Transformers models via the token vector encoder. It requires the PyTT_TokenVectorEncoder to run before it in the pipeline.

The component is available as pytt_textcat and registered via an entry point, so it can also be created using nlp.create_pipe:

textcat = nlp.create_pipe("pytt_textcat")


#### classmethodPyTT_TextCategorizer.from_nlp

Factory to add to Language.factories via entry point.

Name Type Description
nlp spacy.language.Language The nlp object the component is created with.
**cfg - Optional config parameters.
RETURNS PyTT_TextCategorizer The text categorizer.

#### classmethodPyTT_TextCategorizer.Model

Create a text classification model using a PyTorch-Transformers model for token vector encoding.

Name Type Description
nr_class int Number of classes.
width int The width of the tensors being assigned.
exclusive_classes bool Make categories mutually exclusive. Defaults to False.
**cfg - Optional config parameters.
RETURNS thinc.neural.Model The model.

### dataclassActivations

Dataclass to hold the features produced by PyTorch-Transformers.

Attribute Type Description
last_hidden_state object
pooler_output object
all_hidden_states object
all_attentions object
is_grad bool

### Entry points

This package exposes several entry points that tell spaCy how to initialize its components. If spacy-pytorch-transformers and spaCy are installed in the same environment, you'll be able to run the following and it'll work as expected:

tok2vec = nlp.create_pipe("pytt_tok2vec")


This also means that your custom models can ship a pytt_tok2vec component and define "pytt_tok2vec" in their pipelines, and spaCy will know how to create those components when you deserialize the model. The following entry points are set:

Name Target Type Description
pytt_wordpiecer PyTT_WordPiecer spacy_factories Factory to create the component.
pytt_tok2vec PyTT_TokenVectorEncoder spacy_factories Factory to create the component.
pytt_textcat PyTT_TextCategorizer spacy_factories Factory to create the component.
pytt PyTT_Language spacy_languages Custom Language subclass.

## Project details

Uploaded source
Uploaded py3