Beyond Induction Heads: In-Context Meta Learning Induces Multi-Phase Circuit Emergence

TL;DR

This paper investigates how large language models develop in-context meta-learning abilities through multi-phase circuit emergence, extending understanding beyond induction heads and revealing complex training dynamics.

Contribution

The study introduces an extended copy task to analyze in-context meta-learning, revealing multiple phases and unique circuit emergence during training, which deepens understanding of ICL mechanisms.

Findings

Multiple training phases with distinct circuit emergence.

Circuit development correlates with meta-learning ability.

Insights into the dynamics of transformer training processes.

Abstract

Transformer-based language models exhibit In-Context Learning (ICL), where predictions are made adaptively based on context. While prior work links induction heads to ICL through a sudden jump in accuracy, this can only account for ICL when the answer is included within the context. However, an important property of practical ICL in large language models is the ability to meta-learn how to solve tasks from context, rather than just copying answers from context; how such an ability is obtained during training is largely unexplored. In this paper, we experimentally clarify how such meta-learning ability is acquired by analyzing the dynamics of the model's circuit during training. Specifically, we extend the copy task from previous research into an In-Context Meta Learning setting, where models must infer a task from examples to answer queries. Interestingly, in this setting, we find that there are multiple phases in the process of acquiring such abilities, and that a unique circuit emerges in each phase, contrasting with the single-phases change in induction heads. The emergence of such circuits can be related to several phenomena known in large language models, and our analysis lead to a deeper understanding of the source of the transformer's ICL ability.