Cover Image

Few Tokens, Big Leverage: Preserving Safety Alignment by Constraining Safety Tokens during Fine-tuning

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Wang, Guoli, Shi, Haonan, Ouyang, Tu, Wang, An
Format:	Preprint
Published:	2026
Subjects:	Computation and Language Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

EASE: Practical and Efficient Safety Alignment for Small Language Models
by: Shi, Haonan, et al.
Published: (2025)

Navigating the Designs of Privacy-Preserving Fine-tuning for Large Language Models
by: Shi, Haonan, et al.
Published: (2025)

Why LLM Safety Guardrails Collapse After Fine-tuning: A Similarity Analysis Between Alignment and Fine-tuning Datasets
by: Hsiung, Lei, et al.
Published: (2025)

What Makes and Breaks Safety Fine-tuning? A Mechanistic Study
by: Jain, Samyak, et al.
Published: (2024)

Can Perplexity Predict Fine-tuning Performance? An Investigation of Tokenization Effects on Sequential Language Models for Nepali
by: Luitel, Nishant, et al.
Published: (2024)

Benign Samples Matter! Fine-tuning On Outlier Benign Samples Severely Breaks Safety
by: Guan, Zihan, et al.
Published: (2025)

Token-Efficient Leverage Learning in Large Language Models
by: Zeng, Yuanhao, et al.
Published: (2024)

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning
by: Perin, Gabriel J., et al.
Published: (2025)

Multitask Mayhem: Unveiling and Mitigating Safety Gaps in LLMs Fine-tuning
by: Jan, Essa, et al.
Published: (2024)

Unsupervised Morphological Tree Tokenizer
by: Zhu, Qingyang, et al.
Published: (2024)

Critical Tokens Matter: Token-Level Contrastive Estimation Enhances LLM's Reasoning Capability
by: Lin, Zicheng, et al.
Published: (2024)

Fundamental Safety-Capability Trade-offs in Fine-tuning Large Language Models
by: Chen, Pin-Yu, et al.
Published: (2025)

Advancing LLM Safe Alignment with Safety Representation Ranking
by: Du, Tianqi, et al.
Published: (2025)

TokenShapley: Token Level Context Attribution with Shapley Value
by: Xiao, Yingtai, et al.
Published: (2025)

PairAlign: A Framework for Sequence Tokenization via Self-Alignment with Applications to Audio Tokenization
by: Banerjee, Adhiraj, et al.
Published: (2026)

SEAL: Safety-enhanced Aligned LLM Fine-tuning via Bilevel Data Selection
by: Shen, Han, et al.
Published: (2024)

Memento: Fine-tuning LLM Agents without Fine-tuning LLMs
by: Zhou, Huichi, et al.
Published: (2025)

Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection
by: Sun, Guanglong, et al.
Published: (2026)

Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons
by: Chen, Jianhui, et al.
Published: (2024)

Token Buncher: Shielding LLMs from Harmful Reinforcement Learning Fine-Tuning
by: Feng, Weitao, et al.
Published: (2025)

Probabilistic Token Alignment for Large Language Model Fusion
by: Zeng, Runjia, et al.
Published: (2025)

RIFF: Learning to Rephrase Inputs for Few-shot Fine-tuning of Language Models
by: Najafi, Saeed, et al.
Published: (2024)

LanguaShrink: Reducing Token Overhead with Psycholinguistics
by: Liang, Xuechen, et al.
Published: (2024)

Memory-Efficient Fine-Tuning of Transformers via Token Selection
by: Simoulin, Antoine, et al.
Published: (2025)

Bangla Grammatical Error Detection Leveraging Transformer-based Token Classification
by: Islam, Shayekh Bin, et al.
Published: (2024)

Emulated Disalignment: Safety Alignment for Large Language Models May Backfire!
by: Zhou, Zhanhui, et al.
Published: (2024)

LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety
by: Yang, Junxiao, et al.
Published: (2026)

Multilingual Safety Alignment via Self-Distillation
by: Qin, Ruiyang, et al.
Published: (2026)

Revisiting Graph-Tokenizing Large Language Models: A Systematic Evaluation of Graph Token Understanding
by: Zhang, Zhongjian, et al.
Published: (2026)

Financial Sentiment Analysis: Leveraging Actual and Synthetic Data for Supervised Fine-tuning
by: Atsiwo, Abraham
Published: (2024)

Supervised Fine-Tuning Needs to Unlock the Potential of Token Priority
by: Shen, Zhanming, et al.
Published: (2026)

Test-Time Safety Alignment
by: Saglam, Baturay, et al.
Published: (2026)

Lifelong Safety Alignment for Language Models
by: Wang, Haoyu, et al.
Published: (2025)

Small Vocabularies, Big Gains: Pretraining and Tokenization in Time Series Models
by: Roger, Alexis, et al.
Published: (2025)

Not All Tokens Matter: Towards Efficient LLM Reasoning via Token Significance in Reinforcement Learning
by: Liu, Hanbing, et al.
Published: (2025)

Improving LLM Safety Alignment with Dual-Objective Optimization
by: Zhao, Xuandong, et al.
Published: (2025)

Re-Emergent Misalignment: How Narrow Fine-Tuning Erodes Safety Alignment in LLMs
by: Giordani, Jeremiah
Published: (2025)

NeuronTune: Fine-Grained Neuron Modulation for Balanced Safety-Utility Alignment in LLMs
by: Pan, Birong, et al.
Published: (2025)

Lossless Token Sequence Compression via Meta-Tokens
by: Harvill, John, et al.
Published: (2025)

SparsePO: Controlling Preference Alignment of LLMs via Sparse Token Masks
by: Christopoulou, Fenia, et al.
Published: (2024)