LLM-guided Hierarchical Retrieval
LATTICE turns retrieval into an LLM-driven navigation problem over a semantic scaffold for computational tractability needed for large corpora.
Llm
Scalable In-context Ranking with Generative Models
BlockRank imposes blockwise sparse attention and leverages query-token attention signals for efficient in-context ranking
Exploring Design Choices for Building Language-Specific LLMs
This paper examines how adapting LLMs with vocabulary extension and pretraining improves efficiency and performance across languages