fate-ubw/RAGLAB

Integrates dense (ColBERT, Contriever) and sparse retrievers with support for parallel caching via local API, while implementing 6 RAG algorithms (Self-RAG, etc.) across dual Interact and Evaluation modes for quick prototyping versus rigorous benchmarking. Provides standardized evaluation across 10 datasets using metrics like ALCE and FactScore, with pre-trained Llama3 8B/70B generators and LoRA adapters for efficient fine-tuning of retrieval and generation components.

310 stars. No commits in the last 6 months.