<ul><li>Large Language Models (LLMs) typically rely on Supervised Fine-Tuning (SFT) to specialize in downstream tasks.</li><li>Cross Entropy (CE) loss leads to reduced diversity in the model's outputs, hindering further development.</li><li>A new game-theoretic formulation for SFT is introduced to address output diversity limitations.</li><li>The proposed approach enhances output diversity without compromising downstream performance.</li></ul>

Preserving Diversity in Supervised Fine-Tuning of Large Language Models

Discover more