Results published on FLASK benchmark official website or research papers

Mixture-of-Agents Approach Enhances Large Language Model Capabilities, Scores 65.1% on AlpacaEval 2.0

Jun 10, 2024, 07:45 PM

Recent advancements in large language models (LLMs) have demonstrated significant improvements in natural language understanding and generation tasks. A new approach, termed Mixture-of-Agents (MoA), has been introduced by researchers J Wang, J Wang, B Athiwaratkun, C Zhang, and J Zou from Duke University and TogetherAI. This method constructs a layered architecture where each layer comprises multiple LLM agents, with each agent utilizing outputs from the previous layer to enhance its performance. The MoA approach has achieved state-of-the-art performance on benchmarks such as AlpacaEval 2.0, MT-Bench, and FLASK, surpassing the capabilities of GPT-4 Omni. Notably, MoA, using only open-source LLMs, scored 65.1% on AlpacaEval 2.0, significantly higher than GPT-4 Omni's 57.5%.