This repository contains a multi-agent debate and consensus framework evaluated on the MMLU benchmark.
The system runs multiple agents using the same base LLM with different temperatures, iteratively sharing reasoning across rounds and producing a consensus answer via majority vote.
