LlamaIndex
Overview¶
LlamaIndex, formerly GPT Index, is a Python data framework designed to manage and structure LLM-based applications, with a particular emphasis on storage, indexing and retrieval of data.
LlamaIndex provides a complete set of tools to automate tasks such as data ingestion from heterogeneous sources (PDF files, Web pages, ...) and retrieval-augmented generation (RAG); it also features a rich ecosystem of plugins that make it possible to connect with third-party components, from vector stores to data readers.
Reference documentation:
Reasons for using LamaIndex¶
When working with LLMs, one often needs to augment the power of a model (which comes equipped with general knowledge already) by supplying domain-specific, possibly proprietary data, such as internal reports, a corpus of PDF presentations, and so on.
Effective, reproducible management of this domain-specific data needed to augment the LLMs requires a fair amount of machinery, which is precisely what LlamaIndex offers: the framework supports several indexing and retrieval techniques, ranging from simple to quite sophisticated, and offers plugin for seamless integration with a variety of third-party products.
Astra DB + LlamaIndex¶
One of the important abstractions in LlamaIndex is that of the "vector store", a general interface which allows usage of most vector-capable storage systems within LlamaIndex.
Recently, support for Astra DB and Apache Cassandra®, as one of the available vector stores, was introduced. This makes it possible to run advanced LLM workloads natively in LlamaIndex while at the same time using Astra DB as the storage backend.
The Astra integration for LlamaIndex is powered by the
open-source cassIO library, which provides a set of standardized facilities to interact
with Astra DB (and Cassandra) through the patterns typically needed by ML/LLM applications.
Vector Store¶
Creating a vector store in Astra with the LlamaIndex abstractions is as simple as these few lines of Python code:
from llama_index.vector_stores import CassandraVectorStore
vector_store = CassandraVectorStore(
session=session,
keyspace=keyspace,
table=table_name,
embedding_dimension=vector_dimension,
)
You can then use the store within most LlamaIndex native higher-level abstractions,
such as VectorStoreIndex, and within complex LLM-based pipelines, such as
the "query engines" to run question-answering interactions, in just a few lines
of code. Metadata filtering is also supported for vector-based document retrieval.
For full working examples, and more, please head over to the CassIO pages dedicated to LlamaIndex.
Future integrations¶
Other LlamaIndex integrations with Astra DB / Cassandra are planned, such as a "table reader" for seamless use within the rest of the LlamaIndex ecosystem.
Find out more!¶
You just caught a glimpse of what is possible with LlamaIndex and Astra DB.
Do you want to know more? Visit cassio.org for full tutorials, setup guides and other resources.