Improving Diversity in Language Models: When Temperature Fails, Change the Loss
Abstract
Increasing diversity in language models is a challenging yet essential objective. A common approach is to raise the decoding temperature. In this work, we investigate this approach through a simplistic yet common case to provide insights into why decreasing temperature can improve quality (Precision), while increasing it often fails to boost coverage (Recall). Our analysis reveals that for a model to be effectively tunable through temperature adjustments, it must be trained toward coverage. To address this, we propose rethinking loss functions in language models by leveraging the Precision-Recall framework. Our results demonstrate that this approach achieves a substantially better trade-off between Precision and Recall than merely combining negative log-likelihood training with temperature scaling. These findings offer a pathway toward more versatile and robust language modeling techniques.
Cite
@article{arxiv.2508.09654,
title = {Improving Diversity in Language Models: When Temperature Fails, Change the Loss},
author = {Alexandre Verine and Florian Le Bronnec and Kunhao Zheng and Alexandre Allauzen and Yann Chevaleyre and Benjamin Negrevergne},
journal= {arXiv preprint arXiv:2508.09654},
year = {2025}
}
Comments
Forty-Second International Conference on Machine Learning, ICML2025