Progress Ratio Embeddings: An Impatience Signal for Robust Length Control in Neural Text Generation

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Botcazou, Ivanhoé, Amghar, Tassadit, Lamprier, Sylvain, Saubion, Frédéric
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917458184503296
author Botcazou, Ivanhoé
Amghar, Tassadit
Lamprier, Sylvain
Saubion, Frédéric
author_facet Botcazou, Ivanhoé
Amghar, Tassadit
Lamprier, Sylvain
Saubion, Frédéric
contents Modern neural language models achieve high accuracy in text generation, yet precise control over generation length remains underdeveloped. In this paper, we first investigate a recent length control method based on Reverse Positional Embeddings (RPE) and show its limits when control is requested beyond the training distribution. In particular, using a discrete countdown signal tied to the absolute remaining token count leads to instability. To provide robust length control, we introduce Progress Ratio Embeddings (PRE), as continuous embeddings tied to a trigonometric impatience signal. PRE integrates seamlessly into standard Transformer architectures, providing stable length fidelity without degrading text accuracy under standard evaluation metrics. We further show that PRE generalizes well to unseen target lengths. Experiments on two widely used news-summarization benchmarks validate these findings.
format Preprint
id arxiv_https___arxiv_org_abs_2512_06938
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Progress Ratio Embeddings: An Impatience Signal for Robust Length Control in Neural Text Generation
Botcazou, Ivanhoé
Amghar, Tassadit
Lamprier, Sylvain
Saubion, Frédéric
Computation and Language
Modern neural language models achieve high accuracy in text generation, yet precise control over generation length remains underdeveloped. In this paper, we first investigate a recent length control method based on Reverse Positional Embeddings (RPE) and show its limits when control is requested beyond the training distribution. In particular, using a discrete countdown signal tied to the absolute remaining token count leads to instability. To provide robust length control, we introduce Progress Ratio Embeddings (PRE), as continuous embeddings tied to a trigonometric impatience signal. PRE integrates seamlessly into standard Transformer architectures, providing stable length fidelity without degrading text accuracy under standard evaluation metrics. We further show that PRE generalizes well to unseen target lengths. Experiments on two widely used news-summarization benchmarks validate these findings.
title Progress Ratio Embeddings: An Impatience Signal for Robust Length Control in Neural Text Generation
topic Computation and Language
url https://arxiv.org/abs/2512.06938