Skip to content
Preprint

Data-Efficient Adaptation of LLMs via Attention Head Reweighting

Jul 2026 · 0 citations · 73 references
Computer Science

TL;DR

Experiments on diverse open-source text classification datasets show that AHR can outperform standard baselines like LoRA when learning from limited samples, despite having 200-1000x fewer trainable parameters, as the authors' AHR only modifies ~0.0001% of the model's parameters.

Abstract

Learning effectively from limited data is critical in domains like security where labeled examples are scarce. Large language models (LLMs) have demonstrated some capabilities for data-efficient learning, especially through parameter-efficient adaptation methods, but continue to struggle when faced with few samples for difficult tasks. To meet this challenge, we propose Attention Head Reweighting (AHR), a data-efficient method that adapts LLMs to new text-classification tasks by learning only a single scalar per attention head. This drastically reduces the number of parameters that need to be learned by making use of the functional specialization of individual attention heads. Experiments on diverse open-source text classification datasets show that AHR can outperform standard baselines like LoRA when learning from limited samples, despite having 200-1000x fewer trainable parameters, as our AHR only modifies ~0.0001% of the model's parameters. In addition, our learned weights are easy to interpret and can be analyzed to better understand the mechanisms and attention heads responsible for in-context learning abilities in LLMs.

View source

Similar papers

Preprint Aug 2026

From Specialization to Generalization: Instruction-tuned LLMs for Robust Harmful Content Mitigation

Large language models (LLMs) demonstrate impressive performance across a wide range of general NLP tasks; however, their effectiveness in sensitive domains, such as hate speech detection, remains less clear. Prior studies comparing prompted LLMs with state-of-the-art encoder-based models (e.g., BERT variants (Roy et al., 2023; D\"onmez et al., 2024)) have shown only marginal gains, suggesting that LLMs may not excel in hate speech detection or mitigation. In this work, we revisit this question through the lens of instruction tuning. By thoroughly unifying 36 English hate speech datasets spanning multiple labeling schemes, we fine-tune a generalist LLM, based on Qwen3 (Qwen Team, 2025), specifically for hate speech mitigation. Our results demonstrate not only state-of-the-art performance on in-domain benchmarks but also substantial improvements in cross-domain and cross-lingual generalization--areas where encoder-based specialist classifiers often struggle.

Lukas Edman, Daryna Dementieva, Alexander Fraser · 0 citations
Conference Open access 2026

RoBSA: RoPE-based Blockwise Sparse Multi-head Latent Attention

This paper introduces RoPE-based Block-wise Sparse Attention (RoBSA), a method designed specifically for MLA during the decoding stage of model inference that significantly reduces end-to-end inference latency in the decoding stage by up to 2 .

Xinyu Shi, Kairong Luo, Zhen Zheng et al. · 0 citations
Conference Open access 2026

Soft Orthogonal Low-Rank Adaptation for Knowledge Sharing in Large Language Model Continual Learning

When large language models are used in real-world scenarios, continual learning (CL) becomes a non-trivial problem. In particular, continual learning with modern LLMs is challenged both by the substantial computational costs induced by their massive parameter scale, and by the limitations of current CL meth-ods, which are mainly designed to mitigate catastrophic forgetting while neglecting knowledge sharing across tasks. We further observe that models with stronger performance exhibit stronger inter-task connections. In light of the above challenges and findings, we propose At-tribution Scores-based Soft Orthogonality Low-Rank Adaptation (ASO-LoRA), an effective and efficient framework that simultaneously facilitates knowledge transfer while mitigating catastrophic forgetting. Specifically, ASO-LoRA initially assigns task-specific parameter subspaces for new tasks utilizing multi-LoRA modules, enabling for efficient training and inference without relying on task labels. Then, ASO-LoRA leverages attribution scores to evaluate task similarity and employs soft orthogonality between task-specific subspaces, guiding gradient updates in directions that promote parameter isolation, achieving a balance between knowledge transfer and preservation. Experiments are carried out on both the T5-large and the LLaMA2-7B, showing ASO-LoRA’s superior performance and suitability as a plug-in CL solution for general Transformer-based LLMs. Code is available at https://github. com/736619821/ASO-LORA .

Yitong Wang, Xue Han, Wenchun Gao et al. · 0 citations
Preprint Jul 2026

SAMPLe: SAM-based Optimizer for Prompt Learning in VLMs

SAMPLe (Sharpness-Aware Minimization Prompt Learning), a plug-in sharpness-aware optimizer that enhances prompt generalizability by accounting for loss landscape sharpness, is introduced, establishing itself as a robust, model-agnostic solution for prompt learning.

Hossein Rajoli, Fatemeh Lotfi, Niloufar Alipour Talemi et al. · 0 citations
Conference Open access 2026

LeLoRA: Learnable Low-Rank Adaptation of Large Language Models

Fine-tuning large language models (LLMs) is an effective approach to enhancing their performance on specialized downstream tasks. Among the various techniques, low-rank adaptation has garnered significant attention due to its ability to maintain the full performance of fine-tuning while enhancing computational efficiency. However, existing approaches often rely on manually specified and fixed hyperparameters to identify the trainable components within weight matrices, resulting in suboptimal performance and low parameter efficiency. This paper presents a novel Le arnable Lo w-R ank A daptation (LeLoRA) framework that utilizes dynamically learned fine-tuning strategies to facilitate the effective adaptation of LLMs. Our framework integrates an LLM with a policy network that automatically and adaptively generates matrix-specific adaptation strategies to identify the trainable components of each weight matrix, taking into account their unique characteristics, such as singular values and matrix norms. A reinforcement learning-based optimization algorithm is then employed to iteratively update the LLM and the policy network, ensuring that the generated strategies adapt in real time to the evolving states of the LLM. Extensive experiments have been conducted across various natural language processing tasks. The results across ten different LLMs, ranging from 125M to 70B parameters, provide compelling evidence that LeLoRA consistently outperforms existing baselines in adapting LLMs.

Xiaoling Zhou, Mingjie Zhang, Zhemg Lee et al. · 0 citations
Conference Open access 2026

Towards Fast and Accurate Modeling for Cross-Lingual Label Projection

Information extraction (IE) systems rely on structured data for training, but such annotated data is highly imbalanced across languages, with low-resource languages receiving little attention. Label projection techniques aim to bridge this gap by transferring structured annotations from high-resource to low-resource languages. However, existing methods are either inaccurate or too slow for large-scale use. This work aims to address this problem by developing a more effective method that remains sufficiently efficient for large-scale projection. In particular, we propose to synthesize alignment sequence pairs and fine-tune an encoder model with span alignment objective, while controlling data influence during training. Experimental results across 50+ languages show that our framework consistently outperforms previous state-of-the-art methods while maintaining fast inference speed. In addition, we introduce EXP - the first benchmark for explicit evaluation of label projection, thereby reducing confounders and non-determinism in method assessment.

Thang Le, Huy Huu Nguyen, A. Luu et al. · 0 citations