spaCy/website/docs/usage/layers-architectures.md

title	teaser	menu	next
Layers and Model Architectures	Power spaCy components with custom neural networks	Type Signatures type-sigs Swapping Architectures swap-architectures PyTorch & TensorFlow frameworks Custom Thinc Models thinc Trainable Components components	/usage/projects

Example

from thinc.api import Model, chain

@spacy.registry.architectures.register("model.v1")
def build_model(width: int, classes: int) -> Model:
    tok2vec = build_tok2vec(width)
    output_layer = build_output_layer(width, classes)
    model = chain(tok2vec, output_layer)
    return model

A model architecture is a function that wires up a Thinc Model instance. It describes the neural network that is run internally as part of a component in a spaCy pipeline. To define the actual architecture, you can implement your logic in Thinc directly, or you can use Thinc as a thin wrapper around frameworks such as PyTorch, TensorFlow and MXNet. Each Model can also be used as a sublayer of a larger network, allowing you to freely combine implementations from different frameworks into a single model.

spaCy's built-in components require a Model instance to be passed to them via the config system. To change the model architecture of an existing component, you just need to update the config so that it refers to a different registered function. Once the component has been created from this config, you won't be able to change it anymore. The architecture is like a recipe for the network, and you can't change the recipe once the dish has already been prepared. You have to make a new one.

### config.cfg (excerpt)
[components.tagger]
factory = "tagger"

[components.tagger.model]
@architectures = "model.v1"
width = 512
classes = 16

Type signatures

Example

from typing import List
from thinc.api import Model, chain
from thinc.types import Floats2d
def chain_model(
    tok2vec: Model[List[Doc], List[Floats2d]],
    layer1: Model[List[Floats2d], Floats2d],
    layer2: Model[Floats2d, Floats2d]
) -> Model[List[Doc], Floats2d]:
    model = chain(tok2vec, layer1, layer2)
    return model

The Thinc Model class is a generic type that can specify its input and output types. Python uses a square-bracket notation for this, so the type Model[List, Dict] says that each batch of inputs to the model will be a list, and the outputs will be a dictionary. You can be even more specific and write for instanceModel[List[Doc], Dict[str, float]] to specify that the model expects a list of Doc objects as input, and returns a dictionary mapping of strings to floats. Some of the most common types you'll see are: ​

Type	Description
List[Doc]	A batch of Doc objects. Most components expect their models to take this as input.
Floats2d	A two-dimensional numpy or cupy array of floats. Usually 32-bit.
Ints2d	A two-dimensional numpy or cupy array of integers. Common dtypes include uint64, int32 and int8.
List[Floats2d]	A list of two-dimensional arrays, generally with one array per Doc and one row per token.
Ragged	A container to handle variable-length sequence data in an unpadded contiguous array.
Padded	A container to handle variable-length sequence data in a padded contiguous array.

See the Thinc type reference for details. The model type signatures help you figure out which model architectures and components can fit together. For instance, the TextCategorizer class expects a model typed Model[List[Doc], Floats2d], because the model will predict one row of category probabilities per Doc. In contrast, the Tagger class expects a model typed Model[List[Doc], List[Floats2d]], because it needs to predict one row of probabilities per token.

There's no guarantee that two models with the same type signature can be used interchangeably. There are many other ways they could be incompatible. However, if the types don't match, they almost surely won't be compatible. This little bit of validation goes a long way, especially if you configure your editor or other tools to highlight these errors early. The config file is also validated at the beginning of training, to verify that all the types match correctly.

If you're using a modern editor like Visual Studio Code, you can set up mypy with the custom Thinc plugin and get live feedback about mismatched types as you write code.

Swapping model architectures

If no model is specified for the TextCategorizer, the TextCatEnsemble architecture is used by default. This architecture combines a simple bag-of-words model with a neural network, usually resulting in the most accurate results, but at the cost of speed. The config file for this model would look something like this:

### config.cfg (excerpt)
[components.textcat]
factory = "textcat"
labels = []

[components.textcat.model]
@architectures = "spacy.TextCatEnsemble.v1"
exclusive_classes = false
pretrained_vectors = null
width = 64
conv_depth = 2
embed_size = 2000
window_size = 1
ngram_size = 1
dropout = 0
nO = null

spaCy has two additional built-in textcat architectures, and you can easily use those by swapping out the definition of the textcat's model. For instance, to use the simple and fast bag-of-words model TextCatBOW, you can change the config to:

### config.cfg (excerpt) {highlight="6-10"}
[components.textcat]
factory = "textcat"
labels = []

[components.textcat.model]
@architectures = "spacy.TextCatBOW.v1"
exclusive_classes = false
ngram_size = 1
no_output_layer = false
nO = null

For details on all pre-defined architectures shipped with spaCy and how to configure them, check out the model architectures documentation.

Defining sublayers

Model architecture functions often accept sublayers as arguments, so that you can try substituting a different layer into the network. Depending on how the architecture function is structured, you might be able to define your network structure entirely through the config system, using layers that have already been defined. ​

In most neural network models for NLP, the most important parts of the network are what we refer to as the embed and encode steps. These steps together compute dense, context-sensitive representations of the tokens, and their combination forms a typical Tok2Vec layer:

### config.cfg (excerpt)
[components.tok2vec]
factory = "tok2vec"

[components.tok2vec.model]
@architectures = "spacy.Tok2Vec.v1"

[components.tok2vec.model.embed]
@architectures = "spacy.MultiHashEmbed.v1"
# ...

[components.tok2vec.model.encode]
@architectures = "spacy.MaxoutWindowEncoder.v1"
# ...

By defining these sublayers specifically, it becomes straightforward to swap out a sublayer for another one, for instance changing the first sublayer to a character embedding with the CharacterEmbed architecture:

### config.cfg (excerpt)
[components.tok2vec.model.embed]
@architectures = "spacy.CharacterEmbed.v1"
# ...

[components.tok2vec.model.encode]
@architectures = "spacy.MaxoutWindowEncoder.v1"
# ...

Most of spaCy's default architectures accept a tok2vec layer as a sublayer within the larger task-specific neural network. This makes it easy to switch between transformer, CNN, BiLSTM or other feature extraction approaches. The transformers documentation section shows an example of swapping out a model's standard tok2vec layer with a transformer. And if you want to define your own solution, all you need to do is register a Model[List[Doc], List[Floats2d]] architecture function, and you'll be able to try it out in any of the spaCy components. ​

Wrapping PyTorch, TensorFlow and other frameworks

Thinc allows you to wrap models written in other machine learning frameworks like PyTorch, TensorFlow and MXNet using a unified Model API. This makes it easy to use a model implemented in a different framework to power a component in your spaCy pipeline. For example, to wrap a PyTorch model as a Thinc Model, you can use Thinc's PyTorchWrapper:

from thinc.api import PyTorchWrapper

wrapped_pt_model = PyTorchWrapper(torch_model)

Let's use PyTorch to define a very simple neural network consisting of two hidden Linear layers with ReLU activation and dropout, and a softmax-activated output layer:

### PyTorch model
from torch import nn

torch_model = nn.Sequential(
    nn.Linear(width, hidden_width),
    nn.ReLU(),
    nn.Dropout2d(dropout),
    nn.Linear(hidden_width, nO),
    nn.ReLU(),
    nn.Dropout2d(dropout),
    nn.Softmax(dim=1)
)

The resulting wrapped Model can be used as a custom architecture as such, or can be a subcomponent of a larger model. For instance, we can use Thinc's chain combinator, which works like Sequential in PyTorch, to combine the wrapped model with other components in a larger network. This effectively means that you can easily wrap different components from different frameworks, and "glue" them together with Thinc:

from thinc.api import chain, with_array, PyTorchWrapper
from spacy.ml import CharacterEmbed

wrapped_pt_model = PyTorchWrapper(torch_model)
char_embed = CharacterEmbed(width, embed_size, nM, nC)
model = chain(char_embed, with_array(wrapped_pt_model))

In the above example, we have combined our custom PyTorch model with a character embedding layer defined by spaCy. CharacterEmbed returns a Model that takes a List[Doc] as input, and outputs a List[Floats2d]. To make sure that the wrapped PyTorch model receives valid inputs, we use Thinc's with_array helper.

You could also implement a model that only uses PyTorch for the transformer layers, and "native" Thinc layers to do fiddly input and output transformations and add on task-specific "heads", as efficiency is less of a consideration for those parts of the network.

Using wrapped models

To use our custom model including the PyTorch subnetwork, all we need to do is register the architecture using the architectures registry. This assigns the architecture a name so spaCy knows how to find it, and allows passing in arguments like hyperparameters via the config. The full example then becomes:

### Registering the architecture {highlight="9"}
from typing import List
from thinc.types import Floats2d
from thinc.api import Model, PyTorchWrapper, chain, with_array
import spacy
from spacy.tokens.doc import Doc
from spacy.ml import CharacterEmbed
from torch import nn

@spacy.registry.architectures("CustomTorchModel.v1")
def create_torch_model(
    nO: int,
    width: int,
    hidden_width: int,
    embed_size: int,
    nM: int,
    nC: int,
    dropout: float,
) -> Model[List[Doc], List[Floats2d]]:
    char_embed = CharacterEmbed(width, embed_size, nM, nC)
    torch_model = nn.Sequential(
        nn.Linear(width, hidden_width),
        nn.ReLU(),
        nn.Dropout2d(dropout),
        nn.Linear(hidden_width, nO),
        nn.ReLU(),
        nn.Dropout2d(dropout),
        nn.Softmax(dim=1)
    )
    wrapped_pt_model = PyTorchWrapper(torch_model)
    model = chain(char_embed, with_array(wrapped_pt_model))
    return model

The model definition can now be used in any existing trainable spaCy component, by specifying it in the config file. In this configuration, all required parameters for the various subcomponents of the custom architecture are passed in as settings via the config.

### config.cfg (excerpt) {highlight="5-5"}
[components.tagger]
factory = "tagger"

[components.tagger.model]
@architectures = "CustomTorchModel.v1"
nO = 50
width = 96
hidden_width = 48
embed_size = 2000
nM = 64
nC = 8
dropout = 0.2

Remember that it is best not to rely on any (hidden) default values, to ensure that training configs are complete and experiments fully reproducible.

Note that when using a PyTorch or Tensorflow model, it is recommended to set the GPU memory allocator accordingly. When gpu_allocator is set to "pytorch" or "tensorflow" in the training config, cupy will allocate memory via those respective libraries, preventing OOM errors when there's available memory sitting in the other library's pool.

### config.cfg (excerpt)
[training]
gpu_allocator = "pytorch"

Custom models with Thinc

Of course it's also possible to define the Model from the previous section entirely in Thinc. The Thinc documentation provides details on the various layers and helper functions available. Combinators can be used to overload operators and a common usage pattern is to bind chain to >>. The "native" Thinc version of our simple neural network would then become:

from thinc.api import chain, with_array, Model, Relu, Dropout, Softmax
from spacy.ml import CharacterEmbed

char_embed = CharacterEmbed(width, embed_size, nM, nC)
with Model.define_operators({">>": chain}):
    layers = (
        Relu(hidden_width, width)
        >> Dropout(dropout)
        >> Relu(hidden_width, hidden_width)
        >> Dropout(dropout)
        >> Softmax(nO, hidden_width)
    )
    model = char_embed >> with_array(layers)

Note that Thinc layers define the output dimension (nO) as the first argument, followed (optionally) by the input dimension (nI). This is in contrast to how the PyTorch layers are defined, where in_features precedes out_features.

Shape inference in Thinc

It is not strictly necessary to define all the input and output dimensions for each layer, as Thinc can perform shape inference between sequential layers by matching up the output dimensionality of one layer to the input dimensionality of the next. This means that we can simplify the layers definition:

Diff

layers = (
    Relu(hidden_width, width)
    >> Dropout(dropout)
-   >> Relu(hidden_width, hidden_width)
+    >> Relu(hidden_width)
    >> Dropout(dropout)
-   >> Softmax(nO, hidden_width)
+   >> Softmax(nO)
)

with Model.define_operators({">>": chain}):
    layers = (
        Relu(hidden_width, width)
        >> Dropout(dropout)
        >> Relu(hidden_width)
        >> Dropout(dropout)
        >> Softmax(nO)
    )

Thinc can even go one step further and deduce the correct input dimension of the first layer, and output dimension of the last. To enable this functionality, you have to call Model.initialize with an input sample X and an output sample Y with the correct dimensions:

### Shape inference with initialization {highlight="3,7,10"}
with Model.define_operators({">>": chain}):
    layers = (
        Relu(hidden_width)
        >> Dropout(dropout)
        >> Relu(hidden_width)
        >> Dropout(dropout)
        >> Softmax()
    )
    model = char_embed >> with_array(layers)
    model.initialize(X=input_sample, Y=output_sample)

The built-in pipeline components in spaCy ensure that their internal models are always initialized with appropriate sample data. In this case, X is typically a List[Doc], while Y is typically a List[Array1d] or List[Array2d], depending on the specific task. This functionality is triggered when nlp.initialize is called.

Dropout and normalization in Thinc

Many of the available Thinc layers allow you to define a dropout argument that will result in "chaining" an additional Dropout layer. Optionally, you can often specify whether or not you want to add layer normalization, which would result in an additional LayerNorm layer. That means that the following layers definition is equivalent to the previous:

with Model.define_operators({">>": chain}):
    layers = (
        Relu(hidden_width, dropout=dropout, normalize=False)
        >> Relu(hidden_width, dropout=dropout, normalize=False)
        >> Softmax()
    )
    model = char_embed >> with_array(layers)
    model.initialize(X=input_sample, Y=output_sample)

Create new trainable components

In addition to swapping out default models in built-in components, you can also implement an entirely new, trainable pipeline component from scratch. This can be done by creating a new class inheriting from Pipe, and linking it up to your custom model implementation.

For details on how to implement pipeline components, check out the usage guide on custom components and the overview of the Pipe methods used by trainable components.

Example: Entity elation extraction component

This section outlines an example use-case of implementing a novel relation extraction component from scratch. We'll implement a binary relation extraction method that determines whether or not two entities in a document are related, and if so, what type of relation. We'll allow multiple types of relations between two such entities (multi-label setting). There are two major steps required:

1. Implement a machine learning model specific to this task. It will have to extract candidates from a Doc and predict a relation for the available candidate pairs.
2. Implement a custom pipeline component powered by the machine learning model that sets annotations on the Doc passing through the pipeline.

Step 1: Implementing the Model

We need to implement a Model that takes a list of documents (List[Doc]) as input, and outputs a two-dimensional matrix (Floats2d) of predictions:

Model type annotations

The Model class is a generic type that can specify its input and output types, e.g. Model[List[Doc], Floats2d]. Type hints are used for static type checks and validation. See the section on type signatures for details.

### Register the model architecture
@registry.architectures.register("rel_model.v1")
def create_relation_model(...) -> Model[List[Doc], Floats2d]:
    model = ...  # 👈 model will go here
    return model

The first layer in this model will typically be an embedding layer such as a Tok2Vec component or a Transformer. This layer is assumed to be of type Model[List[Doc], List[Floats2d]] as it transforms each document into a list of tokens, with each token being represented by its embedding in the vector space.

Next, we need a method that generates pairs of entities that we want to classify as being related or not. As these candidate pairs are typically formed within one document, this function takes a Doc as input and outputs a List of Span tuples. For instance, a very straightforward implementation would be to just take any two entities from the same document:

### Simple candiate generation
def get_candidates(doc: Doc) -> List[Tuple[Span, Span]]:
    candidates = []
    for ent1 in doc.ents:
        for ent2 in doc.ents:
            candidates.append((ent1, ent2))
    return candidates

But we could also refine this further by excluding relations of an entity with itself, and posing a maximum distance (in number of tokens) between two entities. We register this function in the @misc registry so we can refer to it from the config, and easily swap it out for any other candidate generation function.

config.cfg (excerpt)

[model]
@architectures = "rel_model.v1"

[model.tok2vec]
# ...

[model.get_candidates]
@misc = "rel_cand_generator.v1"
max_length = 20

### Extended candidate generation {highlight="1,2,7,8"}
@registry.misc.register("rel_cand_generator.v1")
def create_candidate_indices(max_length: int) -> Callable[[Doc], List[Tuple[Span, Span]]]:
    def get_candidates(doc: "Doc") -> List[Tuple[Span, Span]]:
        candidates = []
        for ent1 in doc.ents:
            for ent2 in doc.ents:
                if ent1 != ent2:
                    if max_length and abs(ent2.start - ent1.start) <= max_length:
                        candidates.append((ent1, ent2))
        return candidates
    return get_candidates

Finally, we require a method that transforms the candidate entity pairs into a 2D tensor using the specified Tok2Vec or Transformer. The resulting Floats2 object will then be processed by a final output_layer of the network. Putting all this together, we can define our relation model in a config file as such:

### config.cfg
[model]
@architectures = "rel_model.v1"
# ...

[model.tok2vec]
# ...

[model.get_candidates]
@misc = "rel_cand_generator.v2"
max_length = 20

[model.create_candidate_tensor]
@misc = "rel_cand_tensor.v1"

[model.output_layer]
@architectures = "rel_output_layer.v1"
# ...

When creating this model, we store the custom functions as attributes and the sublayers as references, so we can access them easily:

tok2vec_layer = model.get_ref("tok2vec")
output_layer = model.get_ref("output_layer")
create_candidate_tensor = model.attrs["create_candidate_tensor"]
get_candidates = model.attrs["get_candidates"]

Step 2: Implementing the pipeline component

To use our new relation extraction model as part of a custom trainable component, we create a subclass of Pipe that holds the model.

### Pipeline component skeleton
from spacy.pipeline import Pipe

class RelationExtractor(Pipe):
     def __init__(self, vocab, model, name="rel"):
        """Create a component instance."""
        self.model = model
        self.vocab = vocab
        self.name = name

    def update(self, examples, drop=0.0, set_annotations=False, sgd=None, losses=None):
        """Learn from a batch of Example objects."""
        ...

    def predict(self, docs):
        """Apply the model to a batch of Doc objects."""
        ...

    def set_annotations(self, docs, predictions):
        """Modify a batch of Doc objects using the predictions."""
         ...

    def initialize(self, get_examples, nlp=None, labels=None):
        """Initialize the model before training."""
        ...

    def add_label(self, label):
        """Add a label to the component."""
        ...

Before the model can be used, it needs to be initialized. This function receives a callback to access the full training data set, or a representative sample. This data set can be used to deduce all relevant labels. Alternatively, a list of labels can be provided to initialize, or you can call the RelationExtractoradd_label directly. The number of labels defines the output dimensionality of the network, and will be used to do shape inference throughout the layers of the neural network. This is triggered by calling Model.initialize.

### The initialize method {highlight="12,18,22"}
from itertools import islice

def initialize(
    self,
    get_examples: Callable[[], Iterable[Example]],
    *,
    nlp: Language = None,
    labels: Optional[List[str]] = None,
):
    if labels is not None:
        for label in labels:
            self.add_label(label)
    else:
        for example in get_examples():
            relations = example.reference._.rel
            for indices, label_dict in relations.items():
                for label in label_dict.keys():
                    self.add_label(label)
    subbatch = list(islice(get_examples(), 10))
    doc_sample = [eg.reference for eg in subbatch]
    label_sample = self._examples_to_truth(subbatch)
    self.model.initialize(X=doc_sample, Y=label_sample)

The initialize method is triggered whenever this component is part of an nlp pipeline, and nlp.initialize is invoked. Typically, this happens when the pipeline is set up before training in spacy train. After initialization, the pipeline component and its internal model can be trained and used to make predictions.

During training, the function update is invoked which delegates to Model.begin_update and a get_loss function that calculate the loss for a batch of examples, as well as the gradient of loss that will be used to update the weights of the model layers. Thinc provides several loss functions that can be used for the implementation of the get_loss function.

### The update method {highlight="12-14"}
def update(
    self,
    examples: Iterable[Example],
    *,
    drop: float = 0.0,
    set_annotations: bool = False,
    sgd: Optional[Optimizer] = None,
    losses: Optional[Dict[str, float]] = None,
) -> Dict[str, float]:
    ...
    docs = [ex.predicted for ex in examples]
    predictions, backprop = self.model.begin_update(docs)
    loss, gradient = self.get_loss(examples, predictions)
    backprop(gradient)
    losses[self.name] += loss
    ...
    return losses

When the internal model is trained, the component can be used to make novel predictions. The predict function needs to be implemented for each subclass of Pipe. In our case, we can simply delegate to the internal model's predict function that takes a batch of Doc objects and returns a Floats2d array:

### The predict method
def predict(self, docs: Iterable[Doc]) -> Floats2d:
    predictions = self.model.predict(docs)
    return self.model.ops.asarray(predictions)

The final method that needs to be implemented, is set_annotations. This function takes the predictions, and modifies the given Doc object in place to store them. For our relation extraction component, we store the data as a dictionary in a custom extension attribute doc._.rel. As keys, we represent the candidate pair by the start offsets of each entity, as this defines an entity pair uniquely within one document.

To interpret the scores predicted by the relation extraction model correctly, we need to refer to the model's get_candidates function that defined which pairs of entities were relevant candidates, so that the predictions can be linked to those exact entities:

Example output

doc = nlp("Amsterdam is the capital of the Netherlands.")
print("spans", [(e.start, e.text, e.label_) for e in doc.ents])
for value, rel_dict in doc._.rel.items():
    print(f"{value}: {rel_dict}")

# spans [(0, 'Amsterdam', 'LOC'), (6, 'Netherlands', 'LOC')]
# (0, 6): {'CAPITAL_OF': 0.89, 'LOCATED_IN': 0.75, 'UNRELATED': 0.002}
# (6, 0): {'CAPITAL_OF': 0.01, 'LOCATED_IN': 0.13, 'UNRELATED': 0.017}

### Registering the extension attribute
from spacy.tokens import Doc
Doc.set_extension("rel", default={})

### The set_annotations method {highlight="5-6,10"}
def set_annotations(self, docs: Iterable[Doc], predictions: Floats2d):
    c = 0
    get_candidates = self.model.attrs["get_candidates"]
    for doc in docs:
        for (e1, e2) in get_candidates(doc):
            offset = (e1.start, e2.start)
            if offset not in doc._.rel:
                doc._.rel[offset] = {}
            for j, label in enumerate(self.labels):
                doc._.rel[offset][label] = predictions[c, j]
            c += 1

Under the hood, when the pipe is applied to a document, it delegates to the predict and set_annotations methods:

### The __call__ method
def __call__(self, Doc doc):
    predictions = self.predict([doc])
    self.set_annotations([doc], predictions)
    return doc

Once our Pipe subclass is fully implemented, we can register the component with the @Language.factory decorator. This assigns it a name and lets you create the component with nlp.add_pipe and via the config.

config.cfg (excerpt)

[components.relation_extractor]
factory = "relation_extractor"

[components.relation_extractor.model]
@architectures = "rel_model.v1"

[components.relation_extractor.model.tok2vec]
# ...

[components.relation_extractor.model.get_candidates]
@misc = "rel_cand_generator.v1"
max_length = 20

### Registering the pipeline component
from spacy.language import Language

@Language.factory("relation_extractor")
def make_relation_extractor(nlp, name, model):
    return RelationExtractor(nlp.vocab, model, name)