arXiv subjects

Yung-Chen Tang

Publications and source records attributed to Yung-Chen Tang.

7 recordsLinked to original sources

arXiv2026

Similar Models Learn Differently: Final-Window Pretraining Shapes Post-Training Beyond SFT

Developers judge a model checkpoint by how it behaves. After supervised fine-tuning (SFT), two checkpoints that perform about the same across relevant benchmarks are treated as interchangeable, equally ready for the next alignment stage, typically preference optimization. We ask whether this judgment misses a pretraining imprint: a difference that no post-SFT benchmark reveals, yet that decides how each checkpoint responds to further training. To find out, we run a controlled experiment on the final window of pretraining, the last data trained on before instruction tuning. Six branches fork from one partially pretrained checkpoint and differ only in this window: 500 million tokens, 0.1% to 1% of the tokens that precede it. Each branch trains its window on a single data source: generic web text, filtered web text, normative discourse, safety text, mathematical text, or synthetic educational text. SFT and post-training are then identical. After SFT the branches behave near-identically, within about one point on instruction following, refusal, and capability, yet the same post-training carries them to very different endpoints, under both a direct preference optimization update and a reinforcement learning update with a verifiable reward. We measure this deviation through refusal of harmful requests: when post-training begins the safety text branch refuses no more than the web text branch, yet by the end it has lost far less of its refusal. The other four branches gain little or no protection, so the effect is selective to what the window contained. The protection requires the safety text to arrive last rather than earlier in pretraining, and it reproduces on a second model family. What a model is pretrained on last shapes how it reacts to alignment. Therefore, a checkpoint should not be evaluated by its post-SFT behavior alone, and what it was trained on last should be reported with it.

cs.AI↗

arXiv2026

Sparse Neuron Ablation Triggers Catastrophic Collapse of the Language Core in Large Vision-Language Models

Large Vision-Language Models (LVLMs) have shown impressive multimodal understanding capabilities, yet the structures that sustain their functionality remain poorly understood from a mechanistic interpretability standpoint. We propose Consistently Activated Neurons (CAN), a progressive neuron ablation method to identify critical neurons whose removal triggers catastrophic collapse, and use it to investigate structural vulnerabilities in representative 7B LVLMs. Experiments reveal that catastrophic collapse can be triggered by ablating as few as four neurons in \texttt{LLaVA-1.5-7b-hf} and a few thousand in \texttt{InstructBLIP-vicuna-7b}, both representing a small fraction of model parameters. Notably, critical neurons are predominantly localized in the language model, particularly in its down-projection layer, rather than in the vision components. We also observe a consistent two-stage collapse pattern: initial expressive degradation followed by sudden, complete collapse. These findings reveal that LVLM functionality depends on a sparse subset of neurons concentrated in the language backbone, offering mechanistic insights into how their functionality is structured and where these models are most vulnerable.

cs.AI↗

arXiv2026

Defining and Evaluating Physical Safety for Large Language Models

Large Language Models (LLMs) are increasingly used to control robotic systems such as drones, but their risks of causing physical threats and harm in real-world applications remain unexplored. Our study addresses the critical gap in evaluating LLM physical safety by developing a comprehensive benchmark for drone control. We classify the physical safety risks of drones into four categories: (1) human-targeted threats, (2) object-targeted threats, (3) infrastructure attacks, and (4) regulatory violations. Our evaluation of mainstream LLMs reveals an undesirable trade-off between utility and safety, with models that excel in code generation often performing poorly in crucial safety aspects. Furthermore, while incorporating advanced prompt engineering techniques such as In-Context Learning and Chain-of-Thought can improve safety, these methods still struggle to identify unintentional attacks. In addition, larger models demonstrate better safety capabilities, particularly in refusing dangerous commands. Our findings and benchmark can facilitate the design and evaluation of physical safety for LLMs. The project page is available at huggingface.co/spaces/TrustSafeAI/LLM-physical-safety.

cs.LG↗

arXiv2025

CarBoN: Calibrated Best-of-N Sampling Improves Test-time Reasoning

Allocating more computation during inference time (test-time scaling) improves language model performance, especially for reasoning tasks. However, popular methods like Best-of-$N$ sampling often show diminishing returns as $N$ increases. To address this inefficiency, we introduce a general test-time calibration framework that adaptively modifies the model toward high-reward reasoning paths, with theoretical guarantees of improving the lower bound of expected reward under finite sampling, all without large language model (LLM) retraining. Within this framework, we propose CarBoN (Calibrated Best-of-$N$), a two-phase method that first explores the solution space and then learns a calibration of the logits via an input-specific temperature $T$ and additive shift vector $δ$, guiding generation toward more reliable reasoning. Experiments on MATH-500 and AIME-2024 show that CarBoN improves efficiency, with up to $4\times$ fewer rollouts to reach the same accuracy, while often achieving higher accuracy under fixed budgets. We also analyze the complementary roles of $T$ and $δ$ in balancing output diversity and correctness, and demonstrate that the framework also generalizes to step-level sampling strategies such as beam search. For more information, please refer to our project page at huggingface.co/spaces/TrustSafeAI/Test-Time-Calibration.

cs.LG↗

arXiv2025

Why LLM Safety Guardrails Collapse After Fine-tuning: A Similarity Analysis Between Alignment and Fine-tuning Datasets

Recent advancements in large language models (LLMs) have underscored their vulnerability to safety alignment jailbreaks, particularly when subjected to downstream fine-tuning. However, existing mitigation strategies primarily focus on reactively addressing jailbreak incidents after safety guardrails have been compromised, removing harmful gradients during fine-tuning, or continuously reinforcing safety alignment throughout fine-tuning. As such, they tend to overlook a critical upstream factor: the role of the original safety-alignment data. This paper therefore investigates the degradation of safety guardrails through the lens of representation similarity between upstream alignment datasets and downstream fine-tuning tasks. Our experiments demonstrate that high similarity between these datasets significantly weakens safety guardrails, making models more susceptible to jailbreaks. Conversely, low similarity between these two types of datasets yields substantially more robust models and thus reduces harmfulness score by up to 10.33%. By highlighting the importance of upstream dataset design in the building of durable safety guardrails and reducing real-world vulnerability to jailbreak attacks, these findings offer actionable insights for fine-tuning service providers.

cs.CR↗

arXiv2024

Neural Clamping: Joint Input Perturbation and Temperature Scaling for Neural Network Calibration

Neural network calibration is an essential task in deep learning to ensure consistency between the confidence of model prediction and the true correctness likelihood. In this paper, we propose a new post-processing calibration method called Neural Clamping, which employs a simple joint input-output transformation on a pre-trained classifier via a learnable universal input perturbation and an output temperature scaling parameter. Moreover, we provide theoretical explanations on why Neural Clamping is provably better than temperature scaling. Evaluated on BloodMNIST, CIFAR-100, and ImageNet image recognition datasets and a variety of deep neural network models, our empirical results show that Neural Clamping significantly outperforms state-of-the-art post-processing calibration methods. The code is available at github.com/yungchentang/NCToolkit, and the demo is available at huggingface.co/spaces/TrustSafeAI/NCTV.

cs.LG↗

arXiv2022

NCTV: Neural Clamping Toolkit and Visualization for Neural Network Calibration

With the advancement of deep learning technology, neural networks have demonstrated their excellent ability to provide accurate predictions in many tasks. However, a lack of consideration for neural network calibration will not gain trust from humans, even for high-accuracy models. In this regard, the gap between the confidence of the model's predictions and the actual correctness likelihood must be bridged to derive a well-calibrated model. In this paper, we introduce the Neural Clamping Toolkit, the first open-source framework designed to help developers employ state-of-the-art model-agnostic calibrated models. Furthermore, we provide animations and interactive sections in the demonstration to familiarize researchers with calibration in neural networks. A Colab tutorial on utilizing our toolkit is also introduced.

cs.LG↗