Pipeline Refactoring and Module Interfaces

We release the refactored code for STORM pipeline to make it easier to run/customize/develop STORM. Interfaces for pipeline modules and data classes have been defined, allowing for better modularity and extensibility.

Language Model Customization Support

STORM now supports the customization of language models across various families and clients, including:

• GPT family
• Claude family
• VLLM client
• TGI client
• Together Client

Examples demonstrating language model customization are provided in the examples/ directory. For more details, please refer to the README.

Retriever Module Customization Support

STORM introduces the ability to customize the Retriever module. As a knowledge curation engine, STORM consumes information provided by the Retriever module. More information about retriever module customization can be found in the README.

This release focuses on enhancing the modularity and flexibility of the STORM pipeline, enabling users to seamlessly integrate and customize language models and retriever modules according to their specific requirements. Contributions to integrating more language model support, search engine/retrieval model support are highly appreciated!