Research

Papers from the Kafila project, newest first.

PAPERS

  1. UNDER REVIEWSeptember 2026

    Kafila: Serving Large Language Models on a Trusted Set of Heterogeneous Commodity Machines

    Murtaza Rangwala, Richard Sinnott, Rajkumar Buyya

    A bounded session of trusted, unequal machines serves one model. Dividing it by what each device can actually serve shortens the slowest stage by 4.2× against uniform division, measured in three countries.