---
title: CuratedTransformer
teaser:
  Pipeline component for multi-task learning with Curated Transformer models
tag: class
source: github.com/explosion/spacy-curated-transformers/blob/main/spacy_curated_transformers/pipeline/transformer.py
version: 3.7
api_base_class: /api/pipe
api_string_name: curated_transformer
---

<Infobox title="Important note" variant="warning">

This component is available via the extension package
[`spacy-curated-transformers`](https://github.com/explosion/spacy-curated-transformers).
It exposes the component via entry points, so if you have the package installed,
using `factory = "curated_transformer"` in your
[training config](/usage/training#config) will work out-of-the-box.

</Infobox>

This pipeline component lets you use a curated set of transformer models in your
pipeline. spaCy Curated Transformers currently supports the following model
types:

- ALBERT
- BERT
- CamemBERT
- RoBERTa
- XLM-RoBERT

If you want to use another type of model, use
[spacy-transformers](/api/spacy-transformers), which allows you to use all
Hugging Face transformer models with spaCy.

You will usually connect downstream components to a shared Curated Transformer
pipe using one of the Curated Transformer listener layers. This works similarly
to spaCy's [Tok2Vec](/api/tok2vec), and the
[Tok2VecListener](/api/architectures/#Tok2VecListener) sublayer. The component
assigns the output of the transformer to the `Doc`'s extension attributes. To
access the values, you can use the custom
[`Doc._.trf_data`](#assigned-attributes) attribute.

For more details, see the [usage documentation](/usage/embeddings-transformers).

## Assigned Attributes {id="assigned-attributes"}

The component sets the following
[custom extension attribute](/usage/processing-pipeline#custom-components-attributes):

| Location         | Value                                                                      |
| ---------------- | -------------------------------------------------------------------------- |
| `Doc._.trf_data` | Curated Transformer outputs for the `Doc` object. ~~DocTransformerOutput~~ |

## Config and Implementation {id="config"}

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`](/api/language#add_pipe) or in your
[`config.cfg` for training](/usage/training#config). See the
[model architectures](/api/architectures#curated-trf) documentation for details
on the curated transformer architectures and their arguments and
hyperparameters.

> #### Example
>
> ```python
> from spacy_curated_transformers.pipeline.transformer import DEFAULT_CONFIG
>
> nlp.add_pipe("curated_transformer", config=DEFAULT_CONFIG)
> ```

| Setting             | Description                                                                                                                                                                                                                                        |
| ------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `model`             | The Thinc [`Model`](https://thinc.ai/docs/api-model) wrapping the transformer. Defaults to [`XlmrTransformer`](/api/architectures#curated-trf). ~~Model~~                                                                                          |
| `frozen`            | If `True`, the model's weights are frozen and no backpropagation is performed. ~~bool~~                                                                                                                                                            |
| `all_layer_outputs` | If `True`, the model returns the outputs of all the layers. Otherwise, only the output of the last layer is returned. This must be set to `True` if any of the pipe's downstream listeners require the outputs of all transformer layers. ~~bool~~ |

```python
https://github.com/explosion/spacy-curated-transformers/blob/main/spacy_curated_transformers/pipeline/transformer.py
```

## CuratedTransformer.\_\_init\_\_ {id="init",tag="method"}

> #### Example
>
> ```python
> # Construction via add_pipe with default model
> trf = nlp.add_pipe("curated_transformer")
>
> # Construction via add_pipe with custom config
> config = {
>     "model": {
>         "@architectures": "spacy-curated-transformers.XlmrTransformer.v1",
>         "vocab_size": 250002,
>         "num_hidden_layers": 12,
>         "hidden_width": 768,
>         "piece_encoder": {
>             "@architectures": "spacy-curated-transformers.XlmrSentencepieceEncoder.v1"
>         }
>     }
> }
> trf = nlp.add_pipe("curated_transformer", config=config)
>
> # Construction from class
> from spacy_curated_transformers import CuratedTransformer
> trf = CuratedTransformer(nlp.vocab, model)
> ```

Construct a `CuratedTransformer` 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`](#assigned-attributes) 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`](/api/language#create_pipe).

| Name                | Description                                                                                                                                                                                                                                        |
| ------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `vocab`             | The shared vocabulary. ~~Vocab~~                                                                                                                                                                                                                   |
| `model`             | One of the supported pre-trained transformer models. ~~Model~~                                                                                                                                                                                     |
| _keyword-only_      |                                                                                                                                                                                                                                                    |
| `name`              | The component instance name. ~~str~~                                                                                                                                                                                                               |
| `frozen`            | If `True`, the model's weights are frozen and no backpropagation is performed. ~~bool~~                                                                                                                                                            |
| `all_layer_outputs` | If `True`, the model returns the outputs of all the layers. Otherwise, only the output of the last layer is returned. This must be set to `True` if any of the pipe's downstream listeners require the outputs of all transformer layers. ~~bool~~ |

## CuratedTransformer.\_\_call\_\_ {id="call",tag="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__`](/api/curatedtransformer#call) and
[`pipe`](/api/curatedtransformer#pipe) delegate to the
[`predict`](/api/curatedtransformer#predict) and
[`set_annotations`](/api/curatedtransformer#set_annotations) methods.

> #### Example
>
> ```python
> doc = nlp("This is a sentence.")
> trf = nlp.add_pipe("curated_transformer")
> # This usually happens under the hood
> processed = trf(doc)
> ```

| Name        | Description                      |
| ----------- | -------------------------------- |
| `doc`       | The document to process. ~~Doc~~ |
| **RETURNS** | The processed document. ~~Doc~~  |

## CuratedTransformer.pipe {id="pipe",tag="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__`](/api/curatedtransformer#call)
and [`pipe`](/api/curatedtransformer#pipe) delegate to the
[`predict`](/api/curatedtransformer#predict) and
[`set_annotations`](/api/curatedtransformer#set_annotations) methods.

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> for doc in trf.pipe(docs, batch_size=50):
>     pass
> ```

| Name           | Description                                                   |
| -------------- | ------------------------------------------------------------- |
| `stream`       | A stream of documents. ~~Iterable[Doc]~~                      |
| _keyword-only_ |                                                               |
| `batch_size`   | The number of documents to buffer. Defaults to `128`. ~~int~~ |
| **YIELDS**     | The processed documents in order. ~~Doc~~                     |

## CuratedTransformer.initialize {id="initialize",tag="method"}

Initialize the component for training and return an
[`Optimizer`](https://thinc.ai/docs/api-optimizers). `get_examples` should be a
function that returns an iterable of [`Example`](/api/example) objects. **At
least one example should be supplied.** 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](https://thinc.ai/docs/usage-models#validation) and
setting up the label scheme based on the data. This method is typically called
by [`Language.initialize`](/api/language#initialize).

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> trf.initialize(lambda: examples, nlp=nlp)
> ```

| Name             | Description                                                                                                                                                                |
| ---------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `get_examples`   | Function that returns gold-standard annotations in the form of [`Example`](/api/example) objects. Must contain at least one `Example`. ~~Callable[[], Iterable[Example]]~~ |
| _keyword-only_   |                                                                                                                                                                            |
| `nlp`            | The current `nlp` object. Defaults to `None`. ~~Optional[Language]~~                                                                                                       |
| `encoder_loader` | Initialization callback for the transformer model. ~~Optional[Callable]~~                                                                                                  |
| `piece_loader`   | Initialization callback for the input piece encoder. ~~Optional[Callable]~~                                                                                                |

## CuratedTransformer.predict {id="predict",tag="method"}

Apply the component's model to a batch of [`Doc`](/api/doc) objects without
modifying them.

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> scores = trf.predict([doc1, doc2])
> ```

| Name        | Description                                 |
| ----------- | ------------------------------------------- |
| `docs`      | The documents to predict. ~~Iterable[Doc]~~ |
| **RETURNS** | The model's prediction for each document.   |

## CuratedTransformer.set_annotations {id="set_annotations",tag="method"}

Assign the extracted features to the `Doc` objects. By default, the
[`DocTransformerOutput`](/api/curatedtransformer#doctransformeroutput) object is
written to the [`Doc._.trf_data`](#assigned-attributes) attribute. Your
`set_extra_annotations` callback is then called, if provided.

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> scores = trf.predict(docs)
> trf.set_annotations(docs, scores)
> ```

| Name     | Description                                                  |
| -------- | ------------------------------------------------------------ |
| `docs`   | The documents to modify. ~~Iterable[Doc]~~                   |
| `scores` | The scores to set, produced by `CuratedTransformer.predict`. |

## CuratedTransformer.update {id="update",tag="method"}

Prepare for an update to the transformer.

Like the [`Tok2Vec`](api/tok2vec) component, the `CuratedTransformer` 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 `CuratedTransformer` 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 transformer listener 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.

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> optimizer = nlp.initialize()
> losses = trf.update(examples, sgd=optimizer)
> ```

| Name           | Description                                                                                                                                                                      |
| -------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `examples`     | A batch of [`Example`](/api/example) objects. Only the [`Example.predicted`](/api/example#predicted) `Doc` object is used, the reference `Doc` is ignored. ~~Iterable[Example]~~ |
| _keyword-only_ |                                                                                                                                                                                  |
| `drop`         | The dropout rate. ~~float~~                                                                                                                                                      |
| `sgd`          | An optimizer. Will be created via [`create_optimizer`](#create_optimizer) if not set. ~~Optional[Optimizer]~~                                                                    |
| `losses`       | Optional record of the loss during training. Updated using the component name as the key. ~~Optional[Dict[str, float]]~~                                                         |
| **RETURNS**    | The updated `losses` dictionary. ~~Dict[str, float]~~                                                                                                                            |

## CuratedTransformer.create_optimizer {id="create_optimizer",tag="method"}

Create an optimizer for the pipeline component.

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> optimizer = trf.create_optimizer()
> ```

| Name        | Description                  |
| ----------- | ---------------------------- |
| **RETURNS** | The optimizer. ~~Optimizer~~ |

## CuratedTransformer.use_params {id="use_params",tag="method, contextmanager"}

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

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> with trf.use_params(optimizer.averages):
>     trf.to_disk("/best_model")
> ```

| Name     | Description                                        |
| -------- | -------------------------------------------------- |
| `params` | The parameter values to use in the model. ~~dict~~ |

## CuratedTransformer.to_disk {id="to_disk",tag="method"}

Serialize the pipe to disk.

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> trf.to_disk("/path/to/transformer")
> ```

| Name           | Description                                                                                                                                |
| -------------- | ------------------------------------------------------------------------------------------------------------------------------------------ |
| `path`         | A 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_ |                                                                                                                                            |
| `exclude`      | String names of [serialization fields](#serialization-fields) to exclude. ~~Iterable[str]~~                                                |

## CuratedTransformer.from_disk {id="from_disk",tag="method"}

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

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> trf.from_disk("/path/to/transformer")
> ```

| Name           | Description                                                                                     |
| -------------- | ----------------------------------------------------------------------------------------------- |
| `path`         | A path to a directory. Paths may be either strings or `Path`-like objects. ~~Union[str, Path]~~ |
| _keyword-only_ |                                                                                                 |
| `exclude`      | String names of [serialization fields](#serialization-fields) to exclude. ~~Iterable[str]~~     |
| **RETURNS**    | The modified `CuratedTransformer` object. ~~CuratedTransformer~~                                |

## CuratedTransformer.to_bytes {id="to_bytes",tag="method"}

> #### Example
>
> ```python
> trf = nlp.add_pipe("curated_transformer")
> trf_bytes = trf.to_bytes()
> ```

Serialize the pipe to a bytestring.

| Name           | Description                                                                                 |
| -------------- | ------------------------------------------------------------------------------------------- |
| _keyword-only_ |                                                                                             |
| `exclude`      | String names of [serialization fields](#serialization-fields) to exclude. ~~Iterable[str]~~ |
| **RETURNS**    | The serialized form of the `CuratedTransformer` object. ~~bytes~~                           |

## CuratedTransformer.from_bytes {id="from_bytes",tag="method"}

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

> #### Example
>
> ```python
> trf_bytes = trf.to_bytes()
> trf = nlp.add_pipe("curated_transformer")
> trf.from_bytes(trf_bytes)
> ```

| Name           | Description                                                                                 |
| -------------- | ------------------------------------------------------------------------------------------- |
| `bytes_data`   | The data to load from. ~~bytes~~                                                            |
| _keyword-only_ |                                                                                             |
| `exclude`      | String names of [serialization fields](#serialization-fields) to exclude. ~~Iterable[str]~~ |
| **RETURNS**    | The `CuratedTransformer` object. ~~CuratedTransformer~~                                     |

## Serialization Fields {id="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.

> #### Example
>
> ```python
> data = trf.to_disk("/path", exclude=["vocab"])
> ```

| Name    | Description                                                    |
| ------- | -------------------------------------------------------------- |
| `vocab` | The shared [`Vocab`](/api/vocab).                              |
| `cfg`   | The config file. You usually don't want to exclude this.       |
| `model` | The binary model data. You usually don't want to exclude this. |

## DocTransformerOutput {id="doctransformeroutput",tag="dataclass"}

Curated Transformer outputs for one `Doc` object. Stores the dense
representations generated by the transformer for each piece identifier. Piece
identifiers are grouped by token. Instances of this class are typically assigned
to the [`Doc._.trf_data`](/api/curatedtransformer#assigned-attributes) extension
attribute.

> #### Example
>
> ```python
> # Get the last hidden layer output for "is" (token index 1)
> doc = nlp("This is a text.")
> tensors = doc._.trf_data.last_hidden_layer_state[1]
> ```

| Name              | Description                                                                                                                                                                        |
| ----------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `all_outputs`     | List of `Ragged` tensors that correspends to outputs of the different transformer layers. Each tensor element corresponds to a piece identifier's representation. ~~List[Ragged]~~ |
| `last_layer_only` | If only the last transformer layer's outputs are preserved. ~~bool~~                                                                                                               |

### DocTransformerOutput.embedding_layer {id="doctransformeroutput-embeddinglayer",tag="property"}

Return the output of the transformer's embedding layer or `None` if
`last_layer_only` is `True`.

| Name        | Description                                  |
| ----------- | -------------------------------------------- |
| **RETURNS** | Embedding layer output. ~~Optional[Ragged]~~ |

### DocTransformerOutput.last_hidden_layer_state {id="doctransformeroutput-lasthiddenlayerstate",tag="property"}

Return the output of the transformer's last hidden layer.

| Name        | Description                          |
| ----------- | ------------------------------------ |
| **RETURNS** | Last hidden layer output. ~~Ragged~~ |

### DocTransformerOutput.all_hidden_layer_states {id="doctransformeroutput-allhiddenlayerstates",tag="property"}

Return the outputs of all transformer layers (excluding the embedding layer).

| Name        | Description                            |
| ----------- | -------------------------------------- |
| **RETURNS** | Hidden layer outputs. ~~List[Ragged]~~ |

### DocTransformerOutput.num_outputs {id="doctransformeroutput-numoutputs",tag="property"}

Return the number of layer outputs stored in the `DocTransformerOutput` instance
(including the embedding layer).

| Name        | Description                |
| ----------- | -------------------------- |
| **RETURNS** | Numbef of outputs. ~~int~~ |

## Span Getters {id="span_getters",source="github.com/explosion/spacy-transformers/blob/master/spacy_curated_transformers/span_getters.py"}

Span getters are functions that take a batch of [`Doc`](/api/doc) objects and
return a lists of [`Span`](/api/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.with_spans]` block of the config to customize the
sequences processed by the transformer.

| Name        | Description                                                   |
| ----------- | ------------------------------------------------------------- |
| `docs`      | A batch of `Doc` objects. ~~Iterable[Doc]~~                   |
| **RETURNS** | The spans to process by the transformer. ~~List[List[Span]]~~ |

### WithStridedSpans.v1 {id="strided_spans",tag="registered function"}

> #### Example config
>
> ```ini
> [transformer.model.with_spans]
> @architectures = "spacy-curated-transformers.WithStridedSpans.v1"
> stride = 96
> window = 128
> ```

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.

| Name     | Description              |
| -------- | ------------------------ |
| `window` | The window size. ~~int~~ |
| `stride` | The stride size. ~~int~~ |

## Model Loaders

[Curated Transformer models](/api/architectures#curated-trf) are constructed
with default hyperparameters and randomized weights when the pipeline is
created. To load the weights of an existing pre-trained model into the pipeline,
one of the following loader callbacks can be used. The pre-trained model must
have the same hyperparameters as the model used by the pipeline.

### HFTransformerEncoderLoader.v1 {id="hf_trfencoder_loader",tag="registered_function"}

Construct a callback that initializes a supported transformer model with weights
from a corresponding HuggingFace model.

| Name       | Description                                |
| ---------- | ------------------------------------------ |
| `name`     | Name of the HuggingFace model. ~~str~~     |
| `revision` | Name of the model revision/branch. ~~str~~ |

### PyTorchCheckpointLoader.v1 {id="pytorch_checkpoint_loader",tag="registered_function"}

Construct a callback that initializes a supported transformer model with weights
from a PyTorch checkpoint.

| Name   | Description                              |
| ------ | ---------------------------------------- |
| `path` | Path to the PyTorch checkpoint. ~~Path~~ |

## Tokenizer Loaders

[Curated Transformer models](/api/architectures#curated-trf) must be paired with
a matching tokenizer (piece encoder) model in a spaCy pipeline. As with the
transformer models, tokenizers are constructed with an empty vocabulary during
pipeline creation - They need to be initialized with an appropriate loader
before use in training/inference.

### ByteBPELoader.v1 {id="bytebpe_loader",tag="registered_function"}

Construct a callback that initializes a Byte-BPE piece encoder model.

| Name          | Description                           |
| ------------- | ------------------------------------- |
| `vocab_path`  | Path to the vocabulary file. ~~Path~~ |
| `merges_path` | Path to the merges file. ~~Path~~     |

### CharEncoderLoader.v1 {id="charencoder_loader",tag="registered_function"}

Construct a callback that initializes a character piece encoder model.

| Name        | Description                                                                 |
| ----------- | --------------------------------------------------------------------------- |
| `path`      | Path to the serialized character model. ~~Path~~                            |
| `bos_piece` | Piece used as a beginning-of-sentence token. Defaults to `"[BOS]"`. ~~str~~ |
| `eos_piece` | Piece used as a end-of-sentence token. Defaults to `"[EOS]"`. ~~str~~       |
| `unk_piece` | Piece used as a stand-in for unknown tokens. Defaults to `"[UNK]"`. ~~str~~ |
| `normalize` | Unicode normalization form to use. Defaults to `"NFKC"`. ~~str~~            |

### HFPieceEncoderLoader.v1 {id="hf_pieceencoder_loader",tag="registered_function"}

Construct a callback that initializes a HuggingFace piece encoder model. Used in
conjunction with the HuggingFace model loader.

| Name       | Description                                |
| ---------- | ------------------------------------------ |
| `name`     | Name of the HuggingFace model. ~~str~~     |
| `revision` | Name of the model revision/branch. ~~str~~ |

### SentencepieceLoader.v1 {id="sentencepiece_loader",tag="registered_function"}

Construct a callback that initializes a SentencePiece piece encoder model.

| Name   | Description                                          |
| ------ | ---------------------------------------------------- |
| `path` | Path to the serialized SentencePiece model. ~~Path~~ |

### WordpieceLoader.v1 {id="wordpiece_loader",tag="registered_function"}

Construct a callback that initializes a WordPiece piece encoder model.

| Name   | Description                                      |
| ------ | ------------------------------------------------ |
| `path` | Path to the serialized WordPiece model. ~~Path~~ |

## Callbacks

### gradual_transformer_unfreezing.v1 {id="gradual_transformer_unfreezing",tag="registered_function"}

Construct a callback that can be used to gradually unfreeze the weights of one
or more Transformer components during training. This can be used to prevent
catastrophic forgetting during fine-tuning.

| Name           | Description                                                                                                                                                                  |
| -------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `target_pipes` | A dictionary whose keys and values correspond to the names of Transformer components and the training step at which they should be unfrozen respectively. ~~Dict[str, int]~~ |