Steering Information Utility in Key-Value Memory for Language Model Post-Training

• URL: http://arxiv.org/abs/2507.05158v2
• Date: Wed, 29 Oct 2025 04:59:45 GMT
• Title: Steering Information Utility in Key-Value Memory for Language Model Post-Training
• Authors: Chunyuan Deng, Ruidi Chang, Hanjie Chen,
• Abstract summary: We introduce InfoSteer, a lightweight method that encourages parametric information utilization in language models (LMs) during post-training.<n>This simple guidance yields consistent performance improvements across 15 downstream tasks in both in-distribution (ID) and out-of-distribution (OOD) evaluations.<n>Our work underscores that vanilla post-training does not fully exploit the potential gained during pre-training, and that steering LMs in latent representation space offers a promising approach to enhance both performance and interpretability.
• Score: 16.655945393684373
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Recent advancements in language models (LMs) have marked a shift toward the growing importance of post-training. Yet, post-training approaches such as supervised fine-tuning (SFT) do not guarantee the effective use of knowledge acquired during pretraining. We therefore introduce InfoSteer, a lightweight method that encourages parametric information utilization in LMs during post-training. Specifically, InfoSteer treats the feed-forward network (FFN) layer as associate key-value memory and promotes the use of stored memory vectors via forward-pass interventions or regularization during backpropagation. This simple guidance during post-training phase yields consistent performance improvements across diverse model families -- including Qwen, Gemma and Llama -- spanning 15 downstream tasks in both in-distribution (ID) and out-of-distribution (OOD) evaluations. Beyond performance gains, we also find that steered LMs can adaptively allocate information by placing more emphasis on generating semantically meaningful tokens, while using fewer resources on simple transition ones (e.g., `\texttt{,}' or `\texttt{and}'). Our work underscores that vanilla post-training does not fully exploit the potential gained during pre-training, and that steering LMs in latent representation space offers a promising approach to enhance both performance and interpretability. The code is available at: https://github.com/chili-lab/InfoSteer.

Related papers

• SAE as a Crystal Ball: Interpretable Features Predict Cross-domain Transferability of LLMs without Training [70.84726713548099]
  SAE-based Transferability Score (STS) is a new metric to forecast post-training transferability.<n>We show that STS accurately predicts the transferability of supervised fine-tuning, achieving Pearson correlation coefficients above 0.7 with actual performance changes.
  arXiv  Detail & Related papers  (2026-03-03T12:01:09Z)
• Enhancing Cross-task Transfer of Large Language Models via Activation Steering [75.41750053623298]
  Cross-task in-context learning offers a direct solution for transferring knowledge across tasks.<n>We investigate whether cross-task transfer can be achieved via latent space steering without parameter updates or input expansion.<n>We propose a novel Cross-task Activation Steering Transfer framework that enables effective transfer by manipulating the model's internal activation states.
  arXiv  Detail & Related papers  (2025-07-17T15:47:22Z)
• Low-Perplexity LLM-Generated Sequences and Where To Find Them [0.0]
  We introduce a systematic approach centered on analyzing low-perplexity sequences - high-probability text spans generated by the model.<n>Our pipeline reliably extracts such long sequences across diverse topics while avoiding degeneration, then traces them back to their sources in the training data.<n>For those that do match, we quantify the distribution of occurrences across source documents, highlighting the scope and nature of verbatim recall.
  arXiv  Detail & Related papers  (2025-07-02T15:58:51Z)
• LLM Post-Training: A Deep Dive into Reasoning Large Language Models [131.10969986056]
  Large Language Models (LLMs) have transformed the natural language processing landscape and brought to life diverse applications.<n>Post-training methods enable LLMs to refine their knowledge, improve reasoning, enhance factual accuracy, and align more effectively with user intents and ethical considerations.
  arXiv  Detail & Related papers  (2025-02-28T18:59:54Z)
• Aligning Instruction Tuning with Pre-training [81.4748965653345]
  We propose Aligning Instruction Tuning with Pre-training (AITP) to align instruction tuning with pre-training distributions.<n>We show consistent performance improvements with AITP on three fully open large language models (LLMs) across eight benchmarks.
  arXiv  Detail & Related papers  (2025-01-16T08:27:40Z)
• Adaptive Rank, Reduced Forgetting: Knowledge Retention in Continual Learning Vision-Language Models with Dynamic Rank-Selective LoRA [19.982853959240497]
  Pre-trained vision-language embedding models such as CLIP have been widely adopted and validated in Continual Learning (CL)<n>Existing CL methods primarily focus on continual downstream adaptation using components isolated from the pre-trained model (PTM)<n>We propose a universal and efficient CL approach for CLIP based on Dynamic Rank-Selective LoRA (CoDyRA)
  arXiv  Detail & Related papers  (2024-12-01T23:41:42Z)
• Efficient Transfer Learning for Video-language Foundation Models [13.166348605993292]
  We propose a parameter-efficient Multi-modalpatio Ssupervised-Temporal Adapter (MSTA) to enhance alignment between textual and visual representations.<n>We evaluate the effectiveness of our approach across four tasks: zero-shot transfer, few-shot learning, base-to-novel generalization, and fully-Temporal learning.
  arXiv  Detail & Related papers  (2024-11-18T01:25:58Z)
• Enhancing Robustness of Vision-Language Models through Orthogonality Learning and Self-Regularization [77.62516752323207]
  We introduce an orthogonal fine-tuning method for efficiently fine-tuning pretrained weights and enabling enhanced robustness and generalization. A self-regularization strategy is further exploited to maintain the stability in terms of zero-shot generalization of VLMs, dubbed OrthSR. For the first time, we revisit the CLIP and CoOp with our method to effectively improve the model on few-shot image classficiation scenario.
  arXiv  Detail & Related papers  (2024-07-11T10:35:53Z)
• Co-training for Low Resource Scientific Natural Language Inference [65.37685198688538]
  We propose a novel co-training method that assigns weights based on the training dynamics of the classifiers to the distantly supervised labels. By assigning importance weights instead of filtering out examples based on an arbitrary threshold on the predicted confidence, we maximize the usage of automatically labeled data. The proposed method obtains an improvement of 1.5% in Macro F1 over the distant supervision baseline, and substantial improvements over several other strong SSL baselines.
  arXiv  Detail & Related papers  (2024-06-20T18:35:47Z)
• Soft Prompting for Unlearning in Large Language Models [11.504012974208466]
  This work focuses on investigating machine unlearning for Large Language Models motivated by data protection regulations. We propose a framework textbfSoft textbfPrompting for textbfUntextbflearning (SPUL) We conduct a rigorous evaluation of the proposed method and our results indicate that SPUL can significantly improve the trade-off between utility and forgetting.
  arXiv  Detail & Related papers  (2024-06-17T19:11:40Z)
• Instruction Tuning With Loss Over Instructions [42.9106826952674]
  Instruction Modelling (IM) trains LMs by applying a loss function to the instruction and prompt part rather than solely to the output part. We show that, in many scenarios, IM can effectively improve the LM performance on both NLP tasks and open-ended generation benchmarks. Remarkably, in the most advantageous case, IM boosts model performance on AlpacaEval 1.0 by over 100%.
  arXiv  Detail & Related papers  (2024-05-23T10:12:03Z)
• Augmenting Unsupervised Reinforcement Learning with Self-Reference [63.68018737038331]
  Humans possess the ability to draw on past experiences explicitly when learning new tasks. We propose the Self-Reference (SR) approach, an add-on module explicitly designed to leverage historical information. Our approach achieves state-of-the-art results in terms of Interquartile Mean (IQM) performance and Optimality Gap reduction on the Unsupervised Reinforcement Learning Benchmark.
  arXiv  Detail & Related papers  (2023-11-16T09:07:34Z)
• Generation-driven Contrastive Self-training for Zero-shot Text Classification with Instruction-following LLM [31.25193238045053]
  We introduce a novel method, namely GenCo, which leverages the strong generative power of large language models to assist in training a smaller language model. In our method, an LLM plays an important role in the self-training loop of a smaller model in two important ways. It helps crafting additional high-quality training pairs, by rewriting input texts conditioned on predicted labels.
  arXiv  Detail & Related papers  (2023-04-24T07:35:38Z)
• Meet in the Middle: A New Pre-training Paradigm [41.52858444519968]
  Most language models (LMs) are trained and applied in an autoregressive left-to-right fashion. We propose a new pre-training paradigm with techniques that jointly improve the training data efficiency. We show the effectiveness of our pre-training paradigm with extensive experiments on both programming and natural language models.
  arXiv  Detail & Related papers  (2023-03-13T17:17:11Z)
• Task Residual for Tuning Vision-Language Models [69.22958802711017]
  We propose a new efficient tuning approach for vision-language models (VLMs) named Task Residual Tuning (TaskRes) TaskRes explicitly decouples the prior knowledge of the pre-trained models and new knowledge regarding a target task. The proposed TaskRes is simple yet effective, which significantly outperforms previous methods on 11 benchmark datasets.
  arXiv  Detail & Related papers  (2022-11-18T15:09:03Z)
• Learning to Win Lottery Tickets in BERT Transfer via Task-agnostic Mask Training [55.43088293183165]
  Recent studies show that pre-trained language models (PLMs) like BERT contain matchingworks that have similar transfer learning performance as the original PLM. In this paper, we find that the BERTworks have even more potential than these studies have shown. We train binary masks over model weights on the pre-training tasks, with the aim of preserving the universal transferability of the subnetwork.
  arXiv  Detail & Related papers  (2022-04-24T08:42:47Z)
• On the Transferability of Pre-trained Language Models: A Study from Artificial Datasets [74.11825654535895]
  Pre-training language models (LMs) on large-scale unlabeled text data makes the model much easier to achieve exceptional downstream performance. We study what specific traits in the pre-training data, other than the semantics, make a pre-trained LM superior to their counterparts trained from scratch on downstream tasks.
  arXiv  Detail & Related papers  (2021-09-08T10:39:57Z)
• Recall and Learn: Fine-tuning Deep Pretrained Language Models with Less Forgetting [66.45372974713189]
  We propose a recall and learn mechanism, which adopts the idea of multi-task learning and jointly learns pretraining tasks and downstream tasks. Experiments show that our method achieves state-of-the-art performance on the GLUE benchmark. We provide open-source RecAdam, which integrates the proposed mechanisms into Adam to facility the NLP community.
  arXiv  Detail & Related papers  (2020-04-27T08:59:57Z)

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.