Pipeline

Transformer

classv3
String name:transformerBase class:TrainablePipe
Pipeline component for multi-task learning with transformer models

This pipeline component lets you use transformer models in your pipeline. It supports all models that are available via the HuggingFace transformers library. Usually you will connect subsequent components to the shared transformer using the TransformerListener layer. This works similarly to spaCy’s Tok2Vec component and Tok2VecListener sublayer.

The component assigns the output of the transformer to the Doc’s extension attributes. We also calculate an alignment between the word-piece tokens and the spaCy tokenization, so that we can use the last hidden states to set the Doc.tensor attribute. When multiple word-piece tokens align to the same spaCy token, the spaCy token receives the sum of their values. To access the values, you can use the custom Doc._.trf_data attribute. The package also adds the function registries @span_getters and @annotation_setters with several built-in registered functions. For more details, see the usage documentation.

Config and implementation

The default config is defined by the pipeline component factory and describes how the component should be configured. You can override its settings via the config argument on nlp.add_pipe or in your config.cfg for training. See the model architectures documentation for details on the transformer architectures and their arguments and hyperparameters.

SettingDescription
max_batch_itemsMaximum size of a padded batch. Defaults to 4096. int
set_extra_annotationsFunction that takes a batch of Doc objects and transformer outputs to set additional annotations on the Doc. The Doc._.trf_data attribute is set prior to calling the callback. Defaults to null_annotation_setter (no additional annotations). Callable[[List[Doc], FullTransformerBatch], None]
modelThe Thinc Model wrapping the transformer. Defaults to TransformerModel. Model[List[Doc], FullTransformerBatch]
explosion/spacy-transformers/master/spacy_transformers/pipeline_component.py
Can't fetch code example from GitHub :( Please use the link below to view the example. If you've come across a broken link, we always appreciate a pull request to the repository, or a report on the issue tracker. Thanks!

Transformer.__init__ method

Construct a Transformer component. One or more subsequent spaCy components can use the transformer outputs as features in its model, with gradients backpropagated to the single shared weights. The activations from the transformer are saved in the Doc._.trf_data extension attribute. You can also provide a callback to set additional annotations. In your application, you would normally use a shortcut for this and instantiate the component using its string name and nlp.add_pipe.

NameDescription
vocabThe shared vocabulary. Vocab
modelThe Thinc Model wrapping the transformer. Usually you will want to use the TransformerModel layer for this. Model[List[Doc], FullTransformerBatch]
set_extra_annotationsFunction that takes a batch of Doc objects and transformer outputs and stores the annotations on the Doc. The Doc._.trf_data attribute is set prior to calling the callback. By default, no additional annotations are set. Callable[[List[Doc], FullTransformerBatch], None]
keyword-only
nameString name of the component instance. Used to add entries to the losses during training. str
max_batch_itemsMaximum size of a padded batch. Defaults to 128*32. int

Transformer.__call__ method

Apply the pipe to one document. The document is modified in place, and returned. This usually happens under the hood when the nlp object is called on a text and all pipeline components are applied to the Doc in order. Both __call__ and pipe delegate to the predict and set_annotations methods.

NameDescription
docThe document to process. Doc

Transformer.pipe method

Apply the pipe to a stream of documents. This usually happens under the hood when the nlp object is called on a text and all pipeline components are applied to the Doc in order. Both __call__ and pipe delegate to the predict and set_annotations methods.

NameDescription
streamA stream of documents. Iterable[Doc]
keyword-only
batch_sizeThe number of documents to buffer. Defaults to 128. int

Transformer.initialize method

Initialize the component for training and return an Optimizer. get_examples should be a function that returns an iterable of Example objects. The data examples are used to initialize the model of the component and can either be the full training data or a representative sample. Initialization includes validating the network, inferring missing shapes and setting up the label scheme based on the data. This method is typically called by Language.initialize.

NameDescription
get_examplesFunction that returns gold-standard annotations in the form of Example objects. Callable[[], Iterable[Example]]
keyword-only
nlpThe current nlp object. Defaults to None. Optional[Language]

Transformer.predict method

Apply the component’s model to a batch of Doc objects without modifying them.

NameDescription
docsThe documents to predict. Iterable[Doc]

Transformer.set_annotations method

Assign the extracted features to the Doc objects. By default, the TransformerData object is written to the Doc._.trf_data attribute. Your set_extra_annotations callback is then called, if provided.

NameDescription
docsThe documents to modify. Iterable[Doc]
scoresThe scores to set, produced by Transformer.predict.

Transformer.update method

Prepare for an update to the transformer. Like the Tok2Vec component, the Transformer component is unusual in that it does not receive “gold standard” annotations to calculate a weight update. The optimal output of the transformer data is unknown – it’s a hidden layer inside the network that is updated by backpropagating from output layers.

The Transformer component therefore does not perform a weight update during its own update method. Instead, it runs its transformer model and communicates the output and the backpropagation callback to any downstream components that have been connected to it via the TransformerListener sublayer. If there are multiple listeners, the last layer will actually backprop to the transformer and call the optimizer, while the others simply increment the gradients.

NameDescription
examplesA batch of Example objects. Only the Example.predicted Doc object is used, the reference Doc is ignored. Iterable[Example]
keyword-only
dropThe dropout rate. float
sgdAn optimizer. Will be created via create_optimizer if not set. Optional[Optimizer]
lossesOptional record of the loss during training. Updated using the component name as the key. Optional[Dict[str, float]]

Transformer.create_optimizer method

Create an optimizer for the pipeline component.

NameDescription

Transformer.use_params methodcontextmanager

Modify the pipe’s model to use the given parameter values. At the end of the context, the original parameters are restored.

NameDescription
paramsThe parameter values to use in the model. dict

Transformer.to_disk method

Serialize the pipe to disk.

NameDescription
pathA path to a directory, which will be created if it doesn’t exist. Paths may be either strings or Path-like objects. Union[str, Path]
keyword-only
excludeString names of serialization fields to exclude. Iterable[str]

Transformer.from_disk method

Load the pipe from disk. Modifies the object in place and returns it.

NameDescription
pathA path to a directory. Paths may be either strings or Path-like objects. Union[str, Path]
keyword-only
excludeString names of serialization fields to exclude. Iterable[str]

Transformer.to_bytes method

Serialize the pipe to a bytestring.

NameDescription
keyword-only
excludeString names of serialization fields to exclude. Iterable[str]

Transformer.from_bytes method

Load the pipe from a bytestring. Modifies the object in place and returns it.

NameDescription
bytes_dataThe data to load from. bytes
keyword-only
excludeString names of serialization fields to exclude. Iterable[str]

Serialization fields

During serialization, spaCy will export several data fields used to restore different aspects of the object. If needed, you can exclude them from serialization by passing in the string names via the exclude argument.

NameDescription
vocabThe shared Vocab.
cfgThe config file. You usually don’t want to exclude this.
modelThe binary model data. You usually don’t want to exclude this.

TransformerData dataclass

Transformer tokens and outputs for one Doc object. The transformer models return tensors that refer to a whole padded batch of documents. These tensors are wrapped into the FullTransformerBatch object. The FullTransformerBatch then splits out the per-document data, which is handled by this class. Instances of this class are typically assigned to the Doc._.trf_data extension attribute.

NameDescription
tokensA slice of the tokens data produced by the tokenizer. This may have several fields, including the token IDs, the texts and the attention mask. See the transformers.BatchEncoding object for details. dict
tensorsThe activations for the Doc from the transformer. Usually the last tensor that is 3-dimensional will be the most important, as that will provide the final hidden state. Generally activations that are 2-dimensional will be attention weights. Details of this variable will differ depending on the underlying transformer model. List[FloatsXd]
alignAlignment from the Doc’s tokenization to the wordpieces. This is a ragged array, where align.lengths[i] indicates the number of wordpiece tokens that token i aligns against. The actual indices are provided at align[i].dataXd. Ragged
widthThe width of the last hidden layer. int

TransformerData.empty classmethod

Create an empty TransformerData container.

NameDescription

FullTransformerBatch dataclass

Holds a batch of input and output objects for a transformer model. The data can then be split to a list of TransformerData objects to associate the outputs to each Doc in the batch.

NameDescription
spansThe batch of input spans. The outer list refers to the Doc objects in the batch, and the inner list are the spans for that Doc. Note that spans are allowed to overlap or exclude tokens, but each Span can only refer to one Doc (by definition). This means that within a Doc, the regions of the output tensors that correspond to each Span may overlap or have gaps, but for each Doc, there is a non-overlapping contiguous slice of the outputs. List[List[Span]]
tokensThe output of the tokenizer. transformers.BatchEncoding
tensorsThe output of the transformer model. List[torch.Tensor]
alignAlignment from the spaCy tokenization to the wordpieces. This is a ragged array, where align.lengths[i] indicates the number of wordpiece tokens that token i aligns against. The actual indices are provided at align[i].dataXd. Ragged
doc_dataThe outputs, split per Doc object. List[TransformerData]

FullTransformerBatch.unsplit_by_doc method

Return a new FullTransformerBatch from a split batch of activations, using the current object’s spans, tokens and alignment. This is used during the backward pass, in order to construct the gradients to pass back into the transformer model.

NameDescription
arraysThe split batch of activations. List[List[Floats3d]]

FullTransformerBatch.split_by_doc method

Split a TransformerData object that represents a batch into a list with one TransformerData per Doc.

NameDescription

Span getters

Span getters are functions that take a batch of Doc objects and return a lists of Span objects for each doc to be processed by the transformer. This is used to manage long documents by cutting them into smaller sequences before running the transformer. The spans are allowed to overlap, and you can also omit sections of the Doc if they are not relevant.

Span getters can be referenced in the [components.transformer.model.get_spans] block of the config to customize the sequences processed by the transformer. You can also register custom span getters using the @spacy.registry.span_getters decorator.

NameDescription
docsA batch of Doc objects. Iterable[Doc]

doc_spans.v1 registered function

Create a span getter that uses the whole document as its spans. This is the best approach if your Doc objects already refer to relatively short texts.

sent_spans.v1 registered function

Create a span getter that uses sentence boundary markers to extract the spans. This requires sentence boundaries to be set (e.g. by the Sentencizer), and may result in somewhat uneven batches, depending on the sentence lengths. However, it does provide the transformer with more meaningful windows to attend over.

strided_spans.v1 registered function

Create a span getter for strided spans. If you set the window and stride to the same value, the spans will cover each token once. Setting stride lower than window will allow for an overlap, so that some tokens are counted twice. This can be desirable, because it allows all tokens to have both a left and right context.

NameDescription
windowThe window size. int
strideThe stride size. int

Annotation setters registered functions

Annotation setters are functions that take a batch of Doc objects and a FullTransformerBatch and can set additional annotations on the Doc, e.g. to set custom or built-in attributes. You can register custom annotation setters using the @registry.annotation_setters decorator.

NameDescription
docsA batch of Doc objects. List[Doc]
trf_dataThe transformers data for the batch. FullTransformerBatch

The following built-in functions are available:

NameDescription
spacy-transformers.null_annotation_setter.v1Don’t set any additional annotations.

Custom attributes

The component sets the following custom extension attributes:

NameDescription
Doc._.trf_dataTransformer tokens and outputs for the Doc object. TransformerData