Saved in:
Bibliographic Details
Main Authors: Peeperkorn, Max, Kouwenhoven, Tom, Brown, Dan, Jordanous, Anna
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2507.20956
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912505799901184
author Peeperkorn, Max
Kouwenhoven, Tom
Brown, Dan
Jordanous, Anna
author_facet Peeperkorn, Max
Kouwenhoven, Tom
Brown, Dan
Jordanous, Anna
contents Instruction-tuning large language models (LLMs) reduces the diversity of their outputs, which has implications for many tasks, particularly for creative tasks. This paper investigates the ``diversity gap'' for a writing prompt narrative generation task. This gap emerges as measured by current diversity metrics for various open-weight and open-source LLMs. The results show significant decreases in diversity due to instruction-tuning. We explore the diversity loss at each fine-tuning stage for the OLMo and OLMo 2 models to further understand how output diversity is affected. The results indicate that DPO has the most substantial impact on diversity. Motivated by these findings, we present a new decoding strategy, conformative decoding, which guides an instruct model using its more diverse base model to reintroduce output diversity. We show that conformative decoding typically increases diversity and even maintains or improves quality.
format Preprint
id arxiv_https___arxiv_org_abs_2507_20956
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Mind the Gap: Conformative Decoding to Improve Output Diversity of Instruction-Tuned Large Language Models
Peeperkorn, Max
Kouwenhoven, Tom
Brown, Dan
Jordanous, Anna
Computation and Language
Artificial Intelligence
Instruction-tuning large language models (LLMs) reduces the diversity of their outputs, which has implications for many tasks, particularly for creative tasks. This paper investigates the ``diversity gap'' for a writing prompt narrative generation task. This gap emerges as measured by current diversity metrics for various open-weight and open-source LLMs. The results show significant decreases in diversity due to instruction-tuning. We explore the diversity loss at each fine-tuning stage for the OLMo and OLMo 2 models to further understand how output diversity is affected. The results indicate that DPO has the most substantial impact on diversity. Motivated by these findings, we present a new decoding strategy, conformative decoding, which guides an instruct model using its more diverse base model to reintroduce output diversity. We show that conformative decoding typically increases diversity and even maintains or improves quality.
title Mind the Gap: Conformative Decoding to Improve Output Diversity of Instruction-Tuned Large Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2507.20956