First tagged release of MUSS: Multilevel Subset Selection for Relevance and Diversity (UAI 2026).
- Core MUSS algorithm (
src/methods/muss.py): partition → select-clusters → select-within, up to 80× faster than MMR with a constant-factor approximation guarantee - RAG and candidate-retrieval experiment runners with DPP, MMR, DGDS, clustering, and random baselines
- One-click Colab demo and local quick-start notebook
- Paper: https://arxiv.org/abs/2503.11126