In a crucial step towards improving the retrieval systems, eRAG provides a unique evaluation method that significantly correlates with the performance of Retrieval-Augmented Generations (RAG). The innovation lies in using the output from the language model for each retrieved document as the basis for gauging its relevance. Key components of this evaluation method include:

• Individual document assessment: Each document’s suitability for the task is verified by its performance on the end task.
• Improved evaluation metrics: eRAG leverages ground truth labels to offer precise performance insights.
• Resource efficiency: Optimizes computational resources by decreasing reliance on extensive computational evaluations. This methodology offers a promising direction for enhancing the efficiency and accuracy of retrieval systems in language model applications.