Learning to Insert [PAUSE] Tokens for Better Reasoning

• URL: http://arxiv.org/abs/2506.03616v1
• Date: Wed, 04 Jun 2025 06:48:41 GMT
• Title: Learning to Insert [PAUSE] Tokens for Better Reasoning
• Authors: Eunki Kim, Sangryul Kim, James Thorne,
• Abstract summary: We introduce a novel approach termed Dynamic Inserting Tokens Training (DIT)<n>Our method identifies positions within sequences where model confidence is lowest according to token log-likelihood.<n>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.
• Score: 6.823521786512908
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: 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.

Related papers

• Instruction-Following Pruning for Large Language Models [58.329978053711024]
  We move beyond the traditional static pruning approach of determining a fixed pruning mask for a model.<n>In our method, the pruning mask is input-dependent and adapts dynamically based on the information described in a user instruction.<n>Our approach, termed "instruction-following pruning", introduces a sparse mask predictor that takes the user instruction as input and dynamically selects the most relevant model parameters for the given task.
  arXiv  Detail & Related papers  (2025-01-03T20:19:14Z)
• Critical Tokens Matter: Token-Level Contrastive Estimation Enhances LLM's Reasoning Capability [53.51560766150442]
  Critical tokens are elements within reasoning trajectories that significantly influence incorrect outcomes.<n>We present a novel framework for identifying these tokens through rollout sampling.<n>We show that identifying and replacing critical tokens significantly improves model accuracy.
  arXiv  Detail & Related papers  (2024-11-29T18:58:22Z)
• Faster Language Models with Better Multi-Token Prediction Using Tensor Decomposition [5.575078692353885]
  We propose a new model for multi-token prediction in transformers, aiming to enhance sampling efficiency without compromising accuracy.<n>By generalizing it to a rank-$r$ canonical probability decomposition, we develop an improved model that predicts multiple tokens simultaneously.
  arXiv  Detail & Related papers  (2024-10-23T11:06:36Z)
• Bidirectional Awareness Induction in Autoregressive Seq2Seq Models [47.82947878753809]
  Bidirectional Awareness Induction (BAI) is a training method that leverages a subset of elements in the network, the Pivots, to perform bidirectional learning without breaking the autoregressive constraints. In particular, we observed an increase of up to 2.4 CIDEr in Image-Captioning, 4.96 BLEU in Neural Machine Translation, and 1.16 ROUGE in Text Summarization compared to the respective baselines.
  arXiv  Detail & Related papers  (2024-08-25T23:46:35Z)
• Monte Carlo Tree Search Boosts Reasoning via Iterative Preference Learning [55.96599486604344]
  We introduce an approach aimed at enhancing the reasoning capabilities of Large Language Models (LLMs) through an iterative preference learning process. We use Monte Carlo Tree Search (MCTS) to iteratively collect preference data, utilizing its look-ahead ability to break down instance-level rewards into more granular step-level signals. The proposed algorithm employs Direct Preference Optimization (DPO) to update the LLM policy using this newly generated step-level preference data.
  arXiv  Detail & Related papers  (2024-05-01T11:10:24Z)
• Exploring the Role of Token in Transformer-based Time Series Forecasting [10.081240480138487]
  Transformer-based methods are a mainstream approach for solving time series forecasting (TSF) Most focus on optimizing the model structure, with few studies paying attention to the role of tokens for predictions. We find that the gradients mainly depend on tokens that contribute to the predicted series, called positive tokens. To utilize T-PE and V-PE, we propose T2B-PE, a Transformer-based dual-branch framework.
  arXiv  Detail & Related papers  (2024-04-16T07:21:39Z)
• StochCA: A Novel Approach for Exploiting Pretrained Models with Cross-Attention [2.66269503676104]
  We introduce a novel fine-tuning method, called cross-attention (StochCA), specific to Transformer architectures. This method modifies the Transformer's self-attention mechanism to selectively utilize knowledge from pretrained models during fine-tuning. Our experimental results show the superiority of StochCA over state-of-the-art approaches in both areas.
  arXiv  Detail & Related papers  (2024-02-25T13:53:49Z)
• Few-Shot Named Entity Recognition: A Comprehensive Study [92.40991050806544]
  We investigate three schemes to improve the model generalization ability for few-shot settings. We perform empirical comparisons on 10 public NER datasets with various proportions of labeled data. We create new state-of-the-art results on both few-shot and training-free settings.
  arXiv  Detail & Related papers  (2020-12-29T23:43:16Z)
• Document Ranking with a Pretrained Sequence-to-Sequence Model [56.44269917346376]
  We show how a sequence-to-sequence model can be trained to generate relevance labels as "target words" Our approach significantly outperforms an encoder-only model in a data-poor regime.
  arXiv  Detail & Related papers  (2020-03-14T22:29:50Z)
• AvgOut: A Simple Output-Probability Measure to Eliminate Dull Responses [97.50616524350123]
  We build dialogue models that are dynamically aware of what utterances or tokens are dull without any feature-engineering. The first model, MinAvgOut, directly maximizes the diversity score through the output distributions of each batch. The second model, Label Fine-Tuning (LFT), prepends to the source sequence a label continuously scaled by the diversity score to control the diversity level. The third model, RL, adopts Reinforcement Learning and treats the diversity score as a reward signal.
  arXiv  Detail & Related papers  (2020-01-15T18:32:06Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.