Emergence Of A High-dimensional Abstraction Phase In Language Transformers

Cheng Emily, Doimo Diego, Kervadec Corentin, Macocco Iuri, Yu Jade, Laio Alessandro, Baroni Marco. Arxiv 2024

[Paper]    
Model Architecture Pretraining Methods Reinforcement Learning Transformer

A language model (LM) is a mapping from a linguistic context to an output token. However, much remains to be known about this mapping, including how its geometric properties relate to its function. We take a high-level geometric approach to its analysis, observing, across five pre-trained transformer-based LMs and three input datasets, a distinct phase characterized by high intrinsic dimensionality. During this phase, representations (1) correspond to the first full linguistic abstraction of the input; (2) are the first to viably transfer to downstream tasks; (3) predict each other across different LMs. Moreover, we find that an earlier onset of the phase strongly predicts better language modelling performance. In short, our results suggest that a central high-dimensionality phase underlies core linguistic processing in many common LM architectures.

Similar Work