Today we’re excited to launch LlamaIndex v0.10.0. It is by far the biggest update to our Python package to date (see this gargantuan PR), and it takes a massive step towards making LlamaIndex a next-generation, production-ready data framework for your LLM applications.
LlamaIndex v0.10 contains some major updates:
- We have created a llama-index-core package, and split all integrations and templates into separate packages: Hundreds of integrations (LLMs, embeddings, vector stores, data loaders, callbacks, agent tools, and more) are now versioned and packaged as a separate PyPI packages, while preserving namespace imports: for example, you can still usefrom llama_index.llms.openai import OpenAI for a LLM.
- LlamaHub will be the central hub for all integrations: the former llama-hub repo itself is consolidated into the main llama_index repo. Instead of integrations being split between the core library and LlamaHub, every integration will be listed on LlamaHub. We are actively working on updating the site, stay tuned!
- ServiceContext is deprecated: Every LlamaIndex user is familiar with ServiceContext, which over time has become a clunky, unneeded abstraction for managing LLMs, embeddings, chunk sizes, callbacks, and more. As a result we are completely deprecating it; you can now either directly specify arguments or set a default.
Upgrading your codebase to LlamaIndex v0.10 may lead to some breakages, primarily around our integrations/packaging changes, but fortunately we’ve included some scripts to make it as easy as possible to migrate your codebase to use LlamaIndex v0.10.