Spectral Insights into Data-Oblivious Critical Layers in Large Language Models

Xuyuan Liu, Lei Hsiung, Yaoqing Yang, Yujun Yan

Track: Extended Abstract Track

Keywords: Representation Dynamic, LLM Training Behavior

TL;DR: We show how the LLM’s representation space can serve as a signal for predicting its training behavior.

Abstract: Understanding how representations evolve across layers in large language models (LLMs) is key to improving their interpretability and robustness. While recent studies have identified critical layers linked to specific functions or behaviors, these efforts typically rely on data-dependent analyses of fine-tuned models, limiting their use to post-hoc settings. In contrast, we introduce a \textit{data-oblivious} approach to identify intrinsic critical layers in pre-fine-tuned LLMs by analyzing representation dynamics via Centered Kernel Alignment (CKA). We show that layers with significant shifts in representation space are also those most affected during fine-tuning—a pattern that holds consistently across tasks for a given model. Our spectral analysis further reveals that these shifts are driven by changes in the top principal components, which encode semantic transitions from rationales to conclusions. We further apply these findings to two practical scenarios: efficient domain adaptation, where fine-tuning critical layers leads to greater loss reduction compared to non-critical layers; and backdoor defense, where freezing them reduces attack success rates by up to 40\%.

Submission Number: 20