The Price of Format: Diversity Collapse in LLMs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yun, Longfei, An, Chenyang, Wang, Zilong, Peng, Letian, Shang, Jingbo
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913857803386880
author Yun, Longfei
An, Chenyang
Wang, Zilong
Peng, Letian
Shang, Jingbo
author_facet Yun, Longfei
An, Chenyang
Wang, Zilong
Peng, Letian
Shang, Jingbo
contents Instruction-tuned large language models (LLMs) employ structured templates, such as role markers and special tokens, to enforce format consistency during inference. However, we identify a critical limitation of such formatting: it induces a phenomenon we term diversity collapse, where the model generates semantically similar outputs for open-ended inputs, undermining creativity and variability. We systematically evaluate this effect across tasks like story completion and free-form generation, finding that (1) diversity collapse persists even under high-temperature sampling, and (2) structural tokens in templates significantly constrain the model's output space. To contextualize these findings, we fine-tune the same model using a range of structured prompts and then evaluate them across three axes: downstream task performance, alignment behavior, and output diversity. Our analysis shows that format consistency between fine-tuning and inference is crucial for structure-sensitive tasks (e.g., GSM8K, IFEval), but has marginal influence on knowledge-heavy tasks (e.g., MMLU, WebQuestions). In contrast, output diversity is primarily governed by the presence or absence of structural tokens, with minimal formatting yielding the most diverse outputs. These findings reveal that current prompting conventions, while beneficial for alignment, may inadvertently suppress output diversity, underscoring the need for diversity-aware prompt design and instruction tuning.
format Preprint
id arxiv_https___arxiv_org_abs_2505_18949
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle The Price of Format: Diversity Collapse in LLMs
Yun, Longfei
An, Chenyang
Wang, Zilong
Peng, Letian
Shang, Jingbo
Computation and Language
Artificial Intelligence
Machine Learning
Instruction-tuned large language models (LLMs) employ structured templates, such as role markers and special tokens, to enforce format consistency during inference. However, we identify a critical limitation of such formatting: it induces a phenomenon we term diversity collapse, where the model generates semantically similar outputs for open-ended inputs, undermining creativity and variability. We systematically evaluate this effect across tasks like story completion and free-form generation, finding that (1) diversity collapse persists even under high-temperature sampling, and (2) structural tokens in templates significantly constrain the model's output space. To contextualize these findings, we fine-tune the same model using a range of structured prompts and then evaluate them across three axes: downstream task performance, alignment behavior, and output diversity. Our analysis shows that format consistency between fine-tuning and inference is crucial for structure-sensitive tasks (e.g., GSM8K, IFEval), but has marginal influence on knowledge-heavy tasks (e.g., MMLU, WebQuestions). In contrast, output diversity is primarily governed by the presence or absence of structural tokens, with minimal formatting yielding the most diverse outputs. These findings reveal that current prompting conventions, while beneficial for alignment, may inadvertently suppress output diversity, underscoring the need for diversity-aware prompt design and instruction tuning.
title The Price of Format: Diversity Collapse in LLMs
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2505.18949