Comprehensive tools, projects and frameworks for developing foundation models tailored to recommendation systems, developped by the USTCLLM group at USTC, leaded by Prof. Defu Lian.
RecFM is a collection of tools and frameworks designed to facilitate the development of robust and efficient recommendation systems. It includes a suite of modular libraries and technologies that cater to various aspects of recommendation system development, from data ingestion to model deployment.
-
RecStudio: A modular recommendation system algorithm library that enables the rapid construction and efficient training of recommendation system models. It also features a visualization platform for better understanding and debugging of models.
-
RecStudio4Industry: The industrial version of RecStudio, designed for high-efficiency model construction and training. It supports data reading from industry-friendly data interfaces such as HDFS and offers quick model deployment and inference capabilities, ideal for building multi-stage recommendation service frameworks.
-
CELA: A cost-effective text embedding model alignment technology for recommendation systems. It converts any text embedding model into a text encoding model that is friendly to recommendation models.
-
GRE: A universal recommendation text embedding model that has been deeply trained and aligned on multiple public datasets across different recommendation domains. It serves multi-domain recommendation systems for text representation extraction.
-
Nexus: Nexus is the first Pytorch-based information retrieval development toolkit aimed at industrial internet applications such as recommendation system and document retrieval.
For detailed usage instructions, please refer to the documentation in each subdir.
If you'd like to contribute to RecFM, please follow the contribution guidelines.
Apache 2.0
For any questions or inquiries, you can reach out to us at [email protected].