Iteration Head: A Mechanistic Study of Chain-of-Thought - Archive ouverte HAL
Pré-Publication, Document De Travail Année : 2024

Iteration Head: A Mechanistic Study of Chain-of-Thought

Résumé

Chain-of-Thought (CoT) reasoning is known to improve Large Language Models both empirically and in terms of theoretical approximation power. However, our understanding of the inner workings and conditions of apparition of CoT capabilities remains limited. This paper helps fill this gap by demonstrating how CoT reasoning emerges in transformers in a controlled and interpretable setting. In particular, we observe the appearance of a specialized attention mechanism dedicated to iterative reasoning, which we coined "iteration heads". We track both the emergence and the precise working of these iteration heads down to the attention level, and measure the transferability of the CoT skills to which they give rise between tasks.
Fichier principal
Vignette du fichier
2406.02128v1.pdf (13.23 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04620726 , version 1 (21-06-2024)

Identifiants

Citer

Vivien Cabannes, Charles Arnal, Wassim Bouaziz, Alice Yang, Francois Charton, et al.. Iteration Head: A Mechanistic Study of Chain-of-Thought. 2024. ⟨hal-04620726⟩
11 Consultations
7 Téléchargements

Altmetric

Partager

More