Controlling Summarization Length Through EOS Token Weighting

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Belligoli, Zeno, Stergiadis, Emmanouil, Fainman, Eran, Gusev, Ilya
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915328142868480
author Belligoli, Zeno
Stergiadis, Emmanouil
Fainman, Eran
Gusev, Ilya
author_facet Belligoli, Zeno
Stergiadis, Emmanouil
Fainman, Eran
Gusev, Ilya
contents Controlling the length of generated text can be crucial in various text-generation tasks, including summarization. Existing methods often require complex model alterations, limiting compatibility with pre-trained models. We address these limitations by developing a simple approach for controlling the length of automatic text summaries by increasing the importance of correctly predicting the EOS token in the cross-entropy loss computation. The proposed methodology is agnostic to architecture and decoding algorithms and orthogonal to other inference-time techniques to control the generation length. We tested it with encoder-decoder and modern GPT-style LLMs, and show that this method can control generation length, often without affecting the quality of the summary.
format Preprint
id arxiv_https___arxiv_org_abs_2506_05017
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Controlling Summarization Length Through EOS Token Weighting
Belligoli, Zeno
Stergiadis, Emmanouil
Fainman, Eran
Gusev, Ilya
Computation and Language
Machine Learning
Controlling the length of generated text can be crucial in various text-generation tasks, including summarization. Existing methods often require complex model alterations, limiting compatibility with pre-trained models. We address these limitations by developing a simple approach for controlling the length of automatic text summaries by increasing the importance of correctly predicting the EOS token in the cross-entropy loss computation. The proposed methodology is agnostic to architecture and decoding algorithms and orthogonal to other inference-time techniques to control the generation length. We tested it with encoder-decoder and modern GPT-style LLMs, and show that this method can control generation length, often without affecting the quality of the summary.
title Controlling Summarization Length Through EOS Token Weighting
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2506.05017