spaCy/website/docs/usage/large-language-models.mdx

---
title: Large Language Models
teaser: Integrating LLMs into structured NLP pipelines
menu:
  - ['Motivation', 'motivation']
  - ['Install', 'install']
  - ['Usage', 'usage']
  - ['Logging', 'logging']
  - ['API', 'api']
  - ['Tasks', 'tasks']
  - ['Models', 'models']
---

[The spacy-llm package](https://github.com/explosion/spacy-llm) integrates Large
Language Models (LLMs) into spaCy pipelines, featuring a modular system for
**fast prototyping** and **prompting**, and turning unstructured responses into
**robust outputs** for various NLP tasks, **no training data** required.

- Serializable `llm` **component** to integrate prompts into your pipeline
- **Modular functions** to define the [**task**](#tasks) (prompting and parsing)
  and [**model**](#models) (model to use)
- Support for **hosted APIs** and self-hosted **open-source models**
- Integration with [`LangChain`](https://github.com/hwchase17/langchain)
- Access to
  **[OpenAI API](https://platform.openai.com/docs/api-reference/introduction)**,
  including GPT-4 and various GPT-3 models
- Built-in support for various **open-source** models hosted on
  [Hugging Face](https://huggingface.co/)
- Usage examples for standard NLP tasks such as **Named Entity Recognition** and
  **Text Classification**
- Easy implementation of **your own functions** via the
  [registry](/api/top-level#registry) for custom prompting, parsing and model
  integrations

## Motivation {id="motivation"}

Large Language Models (LLMs) feature powerful natural language understanding
capabilities. With only a few (and sometimes no) examples, an LLM can be
prompted to perform custom NLP tasks such as text categorization, named entity
recognition, coreference resolution, information extraction and more.

Supervised learning is much worse than LLM prompting for prototyping, but for
many tasks it's much better for production. A transformer model that runs
comfortably on a single GPU is extremely powerful, and it's likely to be a
better choice for any task for which you have a well-defined output. You train
the model with anything from a few hundred to a few thousand labelled examples,
and it will learn to do exactly that. Efficiency, reliability and control are
all better with supervised learning, and accuracy will generally be higher than
LLM prompting as well.

`spacy-llm` lets you have **the best of both worlds**. You can quickly
initialize a pipeline with components powered by LLM prompts, and freely mix in
components powered by other approaches. As your project progresses, you can look
at replacing some or all of the LLM-powered components as you require.

Of course, there can be components in your system for which the power of an LLM
is fully justified. If you want a system that can synthesize information from
multiple documents in subtle ways and generate a nuanced summary for you, bigger
is better. However, even if your production system needs an LLM for some of the
task, that doesn't mean you need an LLM for all of it. Maybe you want to use a
cheap text classification model to help you find the texts to summarize, or
maybe you want to add a rule-based system to sanity check the output of the
summary. These before-and-after tasks are much easier with a mature and
well-thought-out library, which is exactly what spaCy provides.

## Install {id="install"}

`spacy-llm` will be installed automatically in future spaCy versions. For now,
you can run the following in the same virtual environment where you already have
`spacy` [installed](/usage).

> ⚠️ This package is still experimental and it is possible that changes made to
> the interface will be breaking in minor version updates.

```bash
python -m pip install spacy-llm
```

## Usage {id="usage"}

The task and the model have to be supplied to the `llm` pipeline component using
the [config system](/api/data-formats#config). This package provides various
built-in functionality, as detailed in the [API](#-api) documentation.

### Example 1: Add a text classifier using a GPT-3 model from OpenAI {id="example-1"}

Create a new API key from openai.com or fetch an existing one, and ensure the
keys are set as environmental variables. For more background information, see
the [OpenAI](/api/large-language-models#gpt-3-5) section.

Create a config file `config.cfg` containing at least the following (or see the
full example
[here](https://github.com/explosion/spacy-llm/tree/main/usage_examples/textcat_openai)):

```ini
[nlp]
lang = "en"
pipeline = ["llm"]

[components]

[components.llm]
factory = "llm"

[components.llm.task]
@llm_tasks = "spacy.TextCat.v2"
labels = ["COMPLIMENT", "INSULT"]

[components.llm.model]
@llm_models = "spacy.GPT-3-5.v1"
config = {"temperature": 0.0}
```

Now run:

```python
from spacy_llm.util import assemble

nlp = assemble("config.cfg")
doc = nlp("You look gorgeous!")
print(doc.cats)
```

### Example 2: Add NER using an open-source model through Hugging Face {id="example-2"}

To run this example, ensure that you have a GPU enabled, and `transformers`,
`torch` and CUDA installed. For more background information, see the
[DollyHF](/api/large-language-models#dolly) section.

Create a config file `config.cfg` containing at least the following (or see the
full example
[here](https://github.com/explosion/spacy-llm/tree/main/usage_examples/ner_dolly)):

```ini
[nlp]
lang = "en"
pipeline = ["llm"]

[components]

[components.llm]
factory = "llm"

[components.llm.task]
@llm_tasks = "spacy.NER.v3"
labels = ["PERSON", "ORGANISATION", "LOCATION"]

[components.llm.model]
@llm_models = "spacy.Dolly.v1"
# For better performance, use dolly-v2-12b instead
name = "dolly-v2-3b"
```

Now run:

```python
from spacy_llm.util import assemble

nlp = assemble("config.cfg")
doc = nlp("Jack and Jill rode up the hill in Les Deux Alpes")
print([(ent.text, ent.label_) for ent in doc.ents])
```

Note that Hugging Face will download the `"databricks/dolly-v2-3b"` model the
first time you use it. You can
[define the cached directory](https://huggingface.co/docs/huggingface_hub/main/en/guides/manage-cache)
by setting the environmental variable `HF_HOME`. Also, you can upgrade the model
to be `"databricks/dolly-v2-12b"` for better performance.

### Example 3: Create the component directly in Python {id="example-3"}

The `llm` component behaves as any other component does, and there are
[task-specific components](/api/large-language-models#config) defined to help
you hit the ground running with a reasonable built-in task implementation.

```python
import spacy

nlp = spacy.blank("en")
llm_ner = nlp.add_pipe("llm_ner")
llm_ner.add_label("PERSON")
llm_ner.add_label("LOCATION")
nlp.initialize()
doc = nlp("Jack and Jill rode up the hill in Les Deux Alpes")
print([(ent.text, ent.label_) for ent in doc.ents])
```

Note that for efficient usage of resources, typically you would use
[`nlp.pipe(docs)`](/api/language#pipe) with a batch, instead of calling
`nlp(doc)` with a single document.

### Example 4: Implement your own custom task {id="example-4"}

To write a [`task`](#tasks), you need to implement two functions:
`generate_prompts` that takes a list of [`Doc`](/api/doc) objects and transforms
them into a list of prompts, and `parse_responses` that transforms the LLM
outputs into annotations on the [`Doc`](/api/doc), e.g. entity spans, text
categories and more.

To register your custom task, decorate a factory function using the
`spacy_llm.registry.llm_tasks` decorator with a custom name that you can refer
to in your config.

> 📖 For more details, see the
> [**usage example on writing your own task**](https://github.com/explosion/spacy-llm/tree/main/usage_examples#writing-your-own-task)

```python
from typing import Iterable, List
from spacy.tokens import Doc
from spacy_llm.registry import registry
from spacy_llm.util import split_labels


@registry.llm_tasks("my_namespace.MyTask.v1")
def make_my_task(labels: str, my_other_config_val: float) -> "MyTask":
    labels_list = split_labels(labels)
    return MyTask(labels=labels_list, my_other_config_val=my_other_config_val)


class MyTask:
    def __init__(self, labels: List[str], my_other_config_val: float):
        ...

    def generate_prompts(self, docs: Iterable[Doc]) -> Iterable[str]:
        ...

    def parse_responses(
        self, docs: Iterable[Doc], responses: Iterable[str]
    ) -> Iterable[Doc]:
        ...
```

```ini
# config.cfg (excerpt)
[components.llm.task]
@llm_tasks = "my_namespace.MyTask.v1"
labels = LABEL1,LABEL2,LABEL3
my_other_config_val = 0.3
```

## Logging {id="logging"}

spacy-llm has a built-in logger that can log the prompt sent to the LLM as well
as its raw response. This logger uses the debug level and by default has a
`logging.NullHandler()` configured.

In order to use this logger, you can setup a simple handler like this:

```python
import logging
import spacy_llm


spacy_llm.logger.addHandler(logging.StreamHandler())
spacy_llm.logger.setLevel(logging.DEBUG)
```

> NOTE: Any `logging` handler will work here so you probably want to use some
> sort of rotating `FileHandler` as the generated prompts can be quite long,
> especially for tasks with few-shot examples.

Then when using the pipeline you'll be able to view the prompt and response.

E.g. with the config and code from [Example 1](#example-1) above:

```python
from spacy_llm.util import assemble


nlp = assemble("config.cfg")
doc = nlp("You look gorgeous!")
print(doc.cats)
```

You will see `logging` output similar to:

```
Generated prompt for doc: You look gorgeous!

You are an expert Text Classification system. Your task is to accept Text as input
and provide a category for the text based on the predefined labels.

Classify the text below to any of the following labels: COMPLIMENT, INSULT
The task is non-exclusive, so you can provide more than one label as long as
they're comma-delimited. For example: Label1, Label2, Label3.
Do not put any other text in your answer, only one or more of the provided labels with nothing before or after.
If the text cannot be classified into any of the provided labels, answer `==NONE==`.

Here is the text that needs classification


Text:
'''
You look gorgeous!
'''

Model response for doc: You look gorgeous!
COMPLIMENT
```

`print(doc.cats)` to standard output should look like:

```
{'COMPLIMENT': 1.0, 'INSULT': 0.0}
```

## API {id="api"}

`spacy-llm` exposes an `llm` factory with
[configurable settings](/api/large-language-models#config).

An `llm` component is defined by two main settings:

- A [**task**](#tasks), defining the prompt to send to the LLM as well as the
  functionality to parse the resulting response back into structured fields on
  the [Doc](/api/doc) objects.
- A [**model**](#models) defining the model to use and how to connect to it.
  Note that `spacy-llm` supports both access to external APIs (such as OpenAI)
  as well as access to self-hosted open-source LLMs (such as using Dolly through
  Hugging Face).

Moreover, `spacy-llm` exposes a customizable [**caching**](#cache) functionality
to avoid running the same document through an LLM service (be it local or
through a REST API) more than once.

Finally, you can choose to save a stringified version of LLM prompts/responses
within the `Doc.user_data["llm_io"]` attribute by setting `save_io` to `True`.
`Doc.user_data["llm_io"]` is a dictionary containing one entry for every LLM
component within the `nlp` pipeline. Each entry is itself a dictionary, with two
keys: `prompt` and `response`.

A note on `validate_types`: by default, `spacy-llm` checks whether the
signatures of the `model` and `task` callables are consistent with each other
and emits a warning if they don't. `validate_types` can be set to `False` if you
want to disable this behavior.

### Tasks {id="tasks"}

A _task_ defines an NLP problem or question, that will be sent to the LLM via a
prompt. Further, the task defines how to parse the LLM's responses back into
structured information. All tasks are registered in the `llm_tasks` registry.

Practically speaking, a task should adhere to the `Protocol` named `LLMTask`
defined in
[`ty.py`](https://github.com/explosion/spacy-llm/blob/main/spacy_llm/ty.py). It
needs to define a `generate_prompts` function and a `parse_responses` function.

Tasks may support prompt sharding (for more info see the API docs on
[sharding](/api/large-language-models#task-sharding) and
[non-sharding](/api/large-language-models#task-nonsharding) tasks). The function
signatures for `generate_prompts` and `parse_responses` depend on whether they
do.

For tasks **not supporting** sharding:

| Task                                                                                    | Description                                                                                                                                                  |     |
| --------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------ | --- |
| [`task.generate_prompts`](/api/large-language-models#task-nonsharding-generate-prompts) | Takes a collection of documents, and returns a collection of prompts, which can be of type `Any`.                                                            |
| [`task.parse_responses`](/api/large-language-models#task-nonsharding-parse-responses)   | Takes a collection of LLM responses and the original documents, parses the responses into structured information, and sets the annotations on the documents. |

For tasks **supporting** sharding:

| Task                                                                                 | Description                                                                                                                                                                                                                                                  |     |
| ------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | --- |
| [`task.generate_prompts`](/api/large-language-models#task-sharding-generate-prompts) | Takes a collection of documents, and returns a collection of collections of prompt shards, which can be of type `Any`.                                                                                                                                       |
| [`task.parse_responses`](/api/large-language-models#task-sharding-parse-responses)   | Takes a collection of collections of LLM responses (one per prompt shard) and the original documents, parses the responses into structured information, sets the annotations on the doc shards, and merges those doc shards back into a single doc instance. |

Moreover, the task may define an optional [`scorer` method](/api/scorer#score).
It should accept an iterable of `Example` objects as input and return a score
dictionary. If the `scorer` method is defined, `spacy-llm` will call it to
evaluate the component.

| Component                                                               | Description                                                                                                       |
| ----------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------- |
| [`spacy.EntityLinker.v1`](/api/large-language-models#el-v1)             | The entity linking task prompts the model to link all entities in a given text to entries in a knowledge base.    |
| [`spacy.Summarization.v1`](/api/large-language-models#summarization-v1) | The summarization task prompts the model for a concise summary of the provided text.                              |
| [`spacy.NER.v3`](/api/large-language-models#ner-v3)                     | Implements Chain-of-Thought reasoning for NER extraction - obtains higher accuracy than v1 or v2.                 |
| [`spacy.NER.v2`](/api/large-language-models#ner-v2)                     | Builds on v1 and additionally supports defining the provided labels with explicit descriptions.                   |
| [`spacy.NER.v1`](/api/large-language-models#ner-v1)                     | The original version of the built-in NER task supports both zero-shot and few-shot prompting.                     |
| [`spacy.SpanCat.v3`](/api/large-language-models#spancat-v3)             | Adaptation of the v3 NER task to support overlapping entities and store its annotations in `doc.spans`.           |
| [`spacy.SpanCat.v2`](/api/large-language-models#spancat-v2)             | Adaptation of the v2 NER task to support overlapping entities and store its annotations in `doc.spans`.           |
| [`spacy.SpanCat.v1`](/api/large-language-models#spancat-v1)             | Adaptation of the v1 NER task to support overlapping entities and store its annotations in `doc.spans`.           |
| [`spacy.REL.v1`](/api/large-language-models#rel-v1)                     | Relation Extraction task supporting both zero-shot and few-shot prompting.                                        |
| [`spacy.TextCat.v3`](/api/large-language-models#textcat-v3)             | Version 3 builds on v2 and allows setting definitions of labels.                                                  |
| [`spacy.TextCat.v2`](/api/large-language-models#textcat-v2)             | Version 2 builds on v1 and includes an improved prompt template.                                                  |
| [`spacy.TextCat.v1`](/api/large-language-models#textcat-v1)             | Version 1 of the built-in TextCat task supports both zero-shot and few-shot prompting.                            |
| [`spacy.Lemma.v1`](/api/large-language-models#lemma-v1)                 | Lemmatizes the provided text and updates the `lemma_` attribute of the tokens accordingly.                        |
| [`spacy.Raw.v1`](/api/large-language-models#raw-v1)                     | Executes raw doc content as prompt to LLM.                                                                        |
| [`spacy.Sentiment.v1`](/api/large-language-models#sentiment-v1)         | Performs sentiment analysis on provided texts.                                                                    |
| [`spacy.Translation.v1`](/api/large-language-models#translation-v1)     | Translates doc content into the specified target language.                                                        |
| [`spacy.NoOp.v1`](/api/large-language-models#noop-v1)                   | This task is only useful for testing - it tells the LLM to do nothing, and does not set any fields on the `docs`. |

#### Providing examples for few-shot prompts {id="few-shot-prompts"}

All built-in tasks support few-shot prompts, i. e. including examples in a
prompt. Examples can be supplied in two ways: (1) as a separate file containing
only examples or (2) by initializing `llm` with a `get_examples()` callback
(like any other pipeline component).

##### (1) Few-shot example file

A file containing examples for few-shot prompting can be configured like this:

```ini
[components.llm.task]
@llm_tasks = "spacy.NER.v2"
labels = PERSON,ORGANISATION,LOCATION
[components.llm.task.examples]
@misc = "spacy.FewShotReader.v1"
path = "ner_examples.yml"
```

The supplied file has to conform to the format expected by the required task
(see the task documentation further down).

##### (2) Initializing the `llm` component with a `get_examples()` callback

Alternatively, you can initialize your `nlp` pipeline by providing a
`get_examples` callback for [`nlp.initialize`](/api/language#initialize) and
setting `n_prompt_examples` to a positive number to automatically fetch a few
examples for few-shot learning. Set `n_prompt_examples` to `-1` to use all
examples as part of the few-shot learning prompt.

```ini
[initialize.components.llm]
n_prompt_examples = 3
```

### Model {id="models"}

A _model_ defines which LLM model to query, and how to query it. It can be a
simple function taking a collection of prompts (consistent with the output type
of `task.generate_prompts()`) and returning a collection of responses
(consistent with the expected input of `parse_responses`). Generally speaking,
it's a function of type `Callable[[Iterable[Any]], Iterable[Any]]`, but specific
implementations can have other signatures, like
`Callable[[Iterable[str]], Iterable[str]]`.

All built-in models are registered in `llm_models`. If no model is specified,
the repo currently connects to the `OpenAI` API by default using REST, and
accesses the `"gpt-3.5-turbo"` model.

Currently three different approaches to use LLMs are supported:

1. `spacy-llm`s native REST interface. This is the default for all hosted models
   (e. g. OpenAI, Cohere, Anthropic, ...).
2. A HuggingFace integration that allows to run a limited set of HF models
   locally.
3. A LangChain integration that allows to run any model supported by LangChain
   (hosted or locally).

Approaches 1. and 2 are the default for hosted model and local models,
respectively. Alternatively you can use LangChain to access hosted or local
models by specifying one of the models registered with the `langchain.` prefix.

<Infobox>
_Why LangChain if there are also are native REST and HuggingFace interfaces? When should I use what?_

Third-party libraries like `langchain` focus on prompt management, integration
of many different LLM APIs, and other related features such as conversational
memory or agents. `spacy-llm` on the other hand emphasizes features we consider
useful in the context of NLP pipelines utilizing LLMs to process documents
(mostly) independent from each other. It makes sense that the feature sets of
such third-party libraries and `spacy-llm` aren't identical - and users might
want to take advantage of features not available in `spacy-llm`.

The advantage of implementing our own REST and HuggingFace integrations is that
we can ensure a larger degree of stability and robustness, as we can guarantee
backwards-compatibility and more smoothly integrated error handling.

If however there are features or APIs not natively covered by `spacy-llm`, it's
trivial to utilize LangChain to cover this - and easy to customize the prompting
mechanism, if so required.

</Infobox>

<Infobox variant="warning">
Note that when using hosted services, you have to ensure that the [proper API
keys](/api/large-language-models#api-keys) are set as environment variables as described by the corresponding
provider's documentation.

</Infobox>

| Model                                                                   | Description                                    |
| ----------------------------------------------------------------------- | ---------------------------------------------- |
| [`spacy.GPT-4.v2`](/api/large-language-models#models-rest)              | OpenAI’s `gpt-4` model family.                 |
| [`spacy.GPT-3-5.v2`](/api/large-language-models#models-rest)            | OpenAI’s `gpt-3-5` model family.               |
| [`spacy.Text-Davinci.v2`](/api/large-language-models#models-rest)       | OpenAI’s `text-davinci` model family.          |
| [`spacy.Code-Davinci.v2`](/api/large-language-models#models-rest)       | OpenAI’s `code-davinci` model family.          |
| [`spacy.Text-Curie.v2`](/api/large-language-models#models-rest)         | OpenAI’s `text-curie` model family.            |
| [`spacy.Text-Babbage.v2`](/api/large-language-models#models-rest)       | OpenAI’s `text-babbage` model family.          |
| [`spacy.Text-Ada.v2`](/api/large-language-models#models-rest)           | OpenAI’s `text-ada` model family.              |
| [`spacy.Davinci.v2`](/api/large-language-models#models-rest)            | OpenAI’s `davinci` model family.               |
| [`spacy.Curie.v2`](/api/large-language-models#models-rest)              | OpenAI’s `curie` model family.                 |
| [`spacy.Babbage.v2`](/api/large-language-models#models-rest)            | OpenAI’s `babbage` model family.               |
| [`spacy.Ada.v2`](/api/large-language-models#models-rest)                | OpenAI’s `ada` model family.                   |
| [`spacy.Azure.v1`](/api/large-language-models#models-rest)              | Azure's OpenAI models.                         |
| [`spacy.Command.v1`](/api/large-language-models#models-rest)            | Cohere’s `command` model family.               |
| [`spacy.Claude-2.v1`](/api/large-language-models#models-rest)           | Anthropic’s `claude-2` model family.           |
| [`spacy.Claude-1.v1`](/api/large-language-models#models-rest)           | Anthropic’s `claude-1` model family.           |
| [`spacy.Claude-instant-1.v1`](/api/large-language-models#models-rest)   | Anthropic’s `claude-instant-1` model family.   |
| [`spacy.Claude-instant-1-1.v1`](/api/large-language-models#models-rest) | Anthropic’s `claude-instant-1.1` model family. |
| [`spacy.Claude-1-0.v1`](/api/large-language-models#models-rest)         | Anthropic’s `claude-1.0` model family.         |
| [`spacy.Claude-1-2.v1`](/api/large-language-models#models-rest)         | Anthropic’s `claude-1.2` model family.         |
| [`spacy.Claude-1-3.v1`](/api/large-language-models#models-rest)         | Anthropic’s `claude-1.3` model family.         |
| [`spacy.PaLM.v1`](/api/large-language-models#models-rest)               | Google’s `PaLM` model family.                  |
| [`spacy.Dolly.v1`](/api/large-language-models#models-hf)                | Dolly models through HuggingFace.              |
| [`spacy.Falcon.v1`](/api/large-language-models#models-hf)               | Falcon models through HuggingFace.             |
| [`spacy.Mistral.v1`](/api/large-language-models#models-hf)              | Mistral models through HuggingFace.            |
| [`spacy.Llama2.v1`](/api/large-language-models#models-hf)               | Llama2 models through HuggingFace.             |
| [`spacy.StableLM.v1`](/api/large-language-models#models-hf)             | StableLM models through HuggingFace.           |
| [`spacy.OpenLLaMA.v1`](/api/large-language-models#models-hf)            | OpenLLaMA models through HuggingFace.          |
| [LangChain models](/api/large-language-models#langchain-models)         | LangChain models for API retrieval.            |

Note that the chat models variants of Llama 2 are currently not supported. This
is because they need a particular prompting setup and don't add any discernible
benefits in the use case of `spacy-llm` (i. e. no interactive chat) compared to
the completion model variants.

### Cache {id="cache"}

Interacting with LLMs, either through an external API or a local instance, is
costly. Since developing an NLP pipeline generally means a lot of exploration
and prototyping, `spacy-llm` implements a built-in
[cache](/api/large-language-models#cache) to avoid reprocessing the same
documents at each run that keeps batches of documents stored on disk.

### Various functions {id="various-functions"}

| Function                                                                | Description                                                                                                                                                                                                                                                                          |
| ----------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| [`spacy.FewShotReader.v1`](/api/large-language-models#fewshotreader-v1) | This function is registered in spaCy's `misc` registry, and reads in examples from a `.yml`, `.yaml`, `.json` or `.jsonl` file. It uses [`srsly`](https://github.com/explosion/srsly) to read in these files and parses them depending on the file extension.                        |
| [`spacy.FileReader.v1`](/api/large-language-models#filereader-v1)       | This function is registered in spaCy's `misc` registry, and reads a file provided to the `path` to return a `str` representation of its contents. This function is typically used to read [Jinja](https://jinja.palletsprojects.com/en/3.1.x/) files containing the prompt template. |
| [Normalizer functions](/api/large-language-models#normalizer-functions) | These functions provide simple normalizations for string comparisons, e.g. between a list of specified labels and a label given in the raw text of the LLM response.                                                                                                                 |