Skip to content

Transformations#

A transformation is something that takes a list of nodes as an input, and returns a list of nodes. Each component that implements the Transformation base class has both a synchronous __call__() definition and an async acall() definition.

Currently, the following components are Transformation objects:

Usage Pattern#

While transformations are best used with with an IngestionPipeline, they can also be used directly.

from llama_index.core.node_parser import SentenceSplitter
from llama_index.core.extractors import TitleExtractor

node_parser = SentenceSplitter(chunk_size=512)
extractor = TitleExtractor()

# use transforms directly
nodes = node_parser(documents)

# or use a transformation in async
nodes = await extractor.acall(nodes)

Combining with An Index#

Transformations can be passed into an index or overall global settings, and will be used when calling from_documents() or insert() on an index.

from llama_index.core import VectorStoreIndex
from llama_index.core.extractors import (
    TitleExtractor,
    QuestionsAnsweredExtractor,
)
from llama_index.core.ingestion import IngestionPipeline
from llama_index.core.node_parser import TokenTextSplitter

transformations = [
    TokenTextSplitter(chunk_size=512, chunk_overlap=128),
    TitleExtractor(nodes=5),
    QuestionsAnsweredExtractor(questions=3),
]

# global
from llama_index.core import Settings

Settings.transformations = [text_splitter, title_extractor, qa_extractor]

# per-index
index = VectorStoreIndex.from_documents(
    documents, transformations=transformations
)

Custom Transformations#

You can implement any transformation yourself by implementing the base class.

The following custom transformation will remove any special characters or punctutaion in text.

import re
from llama_index.core import Document
from llama_index.embeddings.openai import OpenAIEmbedding
from llama_index.core.node_parser import SentenceSplitter
from llama_index.core.ingestion import IngestionPipeline
from llama_index.core.schema import TransformComponent


class TextCleaner(TransformComponent):
    def __call__(self, nodes, **kwargs):
        for node in nodes:
            node.text = re.sub(r"[^0-9A-Za-z ]", "", node.text)
        return nodes

These can then be used directly or in any IngestionPipeline.

# use in a pipeline
pipeline = IngestionPipeline(
    transformations=[
        SentenceSplitter(chunk_size=25, chunk_overlap=0),
        TextCleaner(),
        OpenAIEmbedding(),
    ],
)

nodes = pipeline.run(documents=[Document.example()])