What AI says about RGB

RGB is an implementation for benchmarking large language models in Retrieval-Augmented Generation (RAG). It provides datasets (en.json, zh.json and refined en_refine.json / zh_refine.json) and evaluation tooling to assess how LLMs retrieve and integrate information. The repository includes an evaluation script (evalue.py) and instructions to run experiments (e.g., evaluating ChatGPT) across English and Chinese datasets, focusing on information integration and counterfactual robustness.