Studying on-device RAG with Mistral models

Running a RAG pipeline on-device means every model choice influences answer quality as well as system performance and load. This analysis focuses on small Mistral-family models — Ministral 3 (3B and 8