Less is More: Local Intrinsic Dimensions of Contextual Language Models

Bastian Rieck (AIDOS Lab, Department of Computer Science, University of Fribourg, Switzerland) · Benjamin Matthias Ruppik (Heinrich-Heine-Universität Düsseldorf) · Julius von Rohrscheidt (Technische Universität München) · Carel van Niekerk (Heinrich-Heine Universität Düsseldorf) · Michael Heck (Heinrich Heine Universität Düsseldorf) · Renato Vukovic (Heinrich Heine University Düsseldorf) · Shutong Feng (Heinrich-Heine Universität Düsseldorf) · Hsien-chin Lin (Heinrich Heine University Düsseldorf) · Nurul Lubis (NAIST/HHU) · Marcus Zibrowius (Heinrich-Heine-Universität Düsseldorf) · Milica Gasic (Heinrich Heine University Düsseldorf)
adaptabilityarithmetic taskcontextual latent embeddingsdialogue state trackingembedding spacesemotion recognitionfine-tuninggeneralizabilitygeneralization abilitygrokkinginterpretabilitylocal dimensionsoverfittingtraining dynamics

Understanding the internal mechanisms of large language models (LLMs) remains a challenging and complex endeavor. Even fundamental questions, such as how fine-tuning affects model behavior, often require extensive empirical evaluation. In this paper, we introduce a novel perspective based on the geometric properties of contextual latent embeddings to study the effects of training and fine-tuning. To that end, we measure the local dimensions of a contextual language model's latent space and analyze their shifts during training and fine-tuning. We show that the local dimensions provide insights into the model's training dynamics and generalization ability. Specifically, the mean of the local dimensions predicts when the model’s training capabilities are exhausted, as exemplified in a dialogue state tracking task, overfitting, as demonstrated in an emotion recognition task, and grokking, as illustrated with an arithmetic task. Furthermore, our experiments suggest a practical heuristic: reductions in the mean local dimension tend to accompany and predict subsequent performance gains. Through this exploration, we aim to provide practitioners with a deeper understanding of the implications of fine-tuning on embedding spaces, facilitating informed decisions when configuring models for specific applications. The results of this work contribute to the ongoing discourse on the interpretability, adaptability, and generalizability of LLMs by bridging the gap between intrinsic model mechanisms and geometric properties in the respective embeddings.