spacy-llm — LLMs in spaCy NLP Pipelines

spacy-llm

spacy-llm brings large language models into spaCy's production NLP pipeline system. It provides drop-in components that use LLMs for tasks like NER, text classification, lemmatisation, and relation extraction — with support for zero-shot and few-shot prompting, Jinja2 templates, and custom task definitions.

Key Features

• spaCy components: llm_ner, llm_textcat, llm_lemmatizer, llm_rel, llm_spancat
• Backends: OpenAI, Anthropic, Cohere, HuggingFace, Ollama, llama.cpp, custom
• Few-shot: add examples to your config YAML — no code changes
• Structured output parsing: maps LLM JSON/text responses to spaCy spans
• Prompt versioning: Jinja2 templates tracked in configs
• Seamlessly composes with other spaCy components (tok2vec, transformers, …)

Quick Start

import spacy

# config.cfg defines the LLM-NER pipeline
nlp = spacy.load("config.cfg")

doc = nlp("Elon Musk founded SpaceX in Hawthorne, California.")
for ent in doc.ents:
    print(ent.text, ent.label_)  # "Elon Musk" PERSON, "SpaceX" ORG, …

# Create a starter config
python -m spacy init config --lang en --pipeline llm > config.cfg

Install via ai-supply

npx ai-supply add spacy-llm-nlp-pipeline-integration

Curated mirror of the open-source spacy-llm (MIT). Get it from the source.