Skip to yearly menu bar Skip to main content


Poster
in
Workshop: Actionable Interpretability

Interpreting the repeated token phenomenon in LLMs

Itay Yona · Jamie Hayes · Ilia Shumailov · Federico Barbero · Yossi Gandelsman

[ ] [ Project Page ]
Sat 19 Jul 1 p.m. PDT — 2 p.m. PDT

Abstract:

Large Language Models (LLMs), despite their impressive capabilities, often fail to accurately repeat a single word when prompted to, and instead output unrelated text. This unexplained failure mode represents a vulnerability, allowing even end-users to diverge models away from their intended behavior. We aim to explain the causes for this phenomenon and link it to the concept of ``attention sinks'', an emergent LLM behavior crucial for fluency, in which the initial token receives disproportionately high attention scores. Our investigation identifies the neural circuit responsible for attention sinks and shows how long repetitions disrupt this circuit. We extend this finding to other non-repeating sequences that exhibit similar circuit disruptions. To address this, we propose a targeted patch that effectively resolves the issue without negatively impacting the model's overall performance. This study provides a mechanistic explanation for an LLM vulnerability, demonstrating how interpretability can diagnose and address issues, and offering insights that pave the way for more secure and reliable models.

Chat is not available.