RAG-Instruct is a RAG dataset designed to comprehensively enhance LLM RAG capabilities, synthesized using GPT-4o. This dataset is based on the Wikipedia corpus and This dataset is based on the Wikipedia corpus and offers the advantages of query-document scenario diversity and task diversity.
The RAG-Instruct dataset can significantly enhance the RAG ability of LLMs and make remarkable improvements in RAG performance across various tasks.
Model
WQA (acc)
PQA (acc)… See the full description on the dataset page:
https://huggingface.co/datasets/FreedomIntelligence/RAG-Instruct.
