Oz Huly

Predicting RAG Performance for Text Completion

2025

We address the challenge of predicting the performance of using retrieval augmented generation (RAG) in large language models (LLMs) for the task of text completion; specifically, we predict the perplexity gain attained by applying RAG. …