datawhalechina/fun-rec

Covers the complete technical evolution from traditional cascading architectures (collaborative filtering, vector/sequence retrieval, feature crossing, multi-task/multi-scenario modeling) to generative paradigms (LLM-based generation, diffusion models, chain-of-thought reasoning). Includes production-level system implementation with tokenization strategies, Scaling Law architecture design, end-to-end generative modeling, and hardware-aware optimization techniques across recommendation retrieval, ranking, and reranking stages.

6,830 stars. Actively maintained with 6 commits in the last 30 days.