Create README.md

README.md

@@ -0,0 +1,27 @@
+# Multi-Agent MMLU Review Board
+
+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.
+
+## Features
+- Configurable number of agents and debate rounds
+- Supports Hugging Face Transformers pipelines
+- Works with gated models (e.g. LLaMA)
+- Outputs per-agent responses and consensus accuracy
+
+## Usage (Colab)
+1. Open `multiagent_mmlu.ipynb`
+2. Install dependencies
+3. Login to Hugging Face
+4. Configure model, rounds, and questions
+5. Run evaluation
+
+## Notes
+- This repo does not contain model weights
+- Users must accept gated model licenses and provide a Hugging Face token
+- MMLU data is downloaded from the official source at runtime
+
+## Credits
+Based on the multi-agent debate idea from:
+https://github.com/composable-models/llm_multiagent_debate