Learning to Insert [PAUSE] Tokens for Better Reasoning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kim, Eunki, Kim, Sangryul, Thorne, James
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918120239661056
author Kim, Eunki
Kim, Sangryul
Thorne, James
author_facet Kim, Eunki
Kim, Sangryul
Thorne, James
contents To enhance reasoning capabilities, previous works have explored incorporating special-purpose tokens into the training process. These strategies strengthen the learning mechanism of transformer-based large language models (LLMs). Building on prior research, in which inserting dummy tokens consecutively just before reasoning steps can enhance effectiveness, we introduce a novel approach termed Dynamic Inserting Tokens Training (DIT). Our method identifies positions within sequences where model confidence is lowest according to token log-likelihood. Strategically inserting [PAUSE] tokens on these positions bolsters the model's predictive capabilities for subsequent tokens. Experimental results across diverse datasets and models, from the 2.7B model to the 8B model, demonstrate that DIT consistently outperforms traditional fine-tuning and previous token insertion methods. With this simple yet effective method, we achieve accuracy gains of up to 4.7%p on GSM8K, 3.23%p on AQUA-RAT, and pass@1 improvements of up to 3.4%p on MBPP datasets. Our work shows a model-based, dynamic approach rather than a heuristic one, thereby broadening the scope of research in reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2506_03616
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Learning to Insert [PAUSE] Tokens for Better Reasoning
Kim, Eunki
Kim, Sangryul
Thorne, James
Computation and Language
To enhance reasoning capabilities, previous works have explored incorporating special-purpose tokens into the training process. These strategies strengthen the learning mechanism of transformer-based large language models (LLMs). Building on prior research, in which inserting dummy tokens consecutively just before reasoning steps can enhance effectiveness, we introduce a novel approach termed Dynamic Inserting Tokens Training (DIT). Our method identifies positions within sequences where model confidence is lowest according to token log-likelihood. Strategically inserting [PAUSE] tokens on these positions bolsters the model's predictive capabilities for subsequent tokens. Experimental results across diverse datasets and models, from the 2.7B model to the 8B model, demonstrate that DIT consistently outperforms traditional fine-tuning and previous token insertion methods. With this simple yet effective method, we achieve accuracy gains of up to 4.7%p on GSM8K, 3.23%p on AQUA-RAT, and pass@1 improvements of up to 3.4%p on MBPP datasets. Our work shows a model-based, dynamic approach rather than a heuristic one, thereby broadening the scope of research in reasoning.
title Learning to Insert [PAUSE] Tokens for Better Reasoning
topic Computation and Language
url https://arxiv.org/abs/2506.03616