Saved in:
Bibliographic Details
Main Authors: He, Tianyu, Doshi, Darshil, Das, Aritra, Gromov, Andrey
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2406.02550
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910683147272192
author He, Tianyu
Doshi, Darshil
Das, Aritra
Gromov, Andrey
author_facet He, Tianyu
Doshi, Darshil
Das, Aritra
Gromov, Andrey
contents Large language models can solve tasks that were not present in the training set. This capability is believed to be due to in-context learning and skill composition. In this work, we study the emergence of in-context learning and skill composition in a collection of modular arithmetic tasks. Specifically, we consider a finite collection of linear modular functions $z = a \, x + b \, y \;\mathrm{mod}\; p$ labeled by the vector $(a, b) \in \mathbb{Z}_p^2$. We use some of these tasks for pre-training and the rest for out-of-distribution testing. We empirically show that a GPT-style transformer exhibits a transition from in-distribution to out-of-distribution generalization as the number of pre-training tasks increases. We find that the smallest model capable of out-of-distribution generalization requires two transformer blocks, while for deeper models, the out-of-distribution generalization phase is \emph{transient}, necessitating early stopping. Finally, we perform an interpretability study of the pre-trained models, revealing highly structured representations in both attention heads and MLPs; and discuss the learned algorithms. Notably, we find an algorithmic shift in deeper models, as we go from few to many in-context examples.
format Preprint
id arxiv_https___arxiv_org_abs_2406_02550
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Learning to grok: Emergence of in-context learning and skill composition in modular arithmetic tasks
He, Tianyu
Doshi, Darshil
Das, Aritra
Gromov, Andrey
Machine Learning
Disordered Systems and Neural Networks
High Energy Physics - Theory
Large language models can solve tasks that were not present in the training set. This capability is believed to be due to in-context learning and skill composition. In this work, we study the emergence of in-context learning and skill composition in a collection of modular arithmetic tasks. Specifically, we consider a finite collection of linear modular functions $z = a \, x + b \, y \;\mathrm{mod}\; p$ labeled by the vector $(a, b) \in \mathbb{Z}_p^2$. We use some of these tasks for pre-training and the rest for out-of-distribution testing. We empirically show that a GPT-style transformer exhibits a transition from in-distribution to out-of-distribution generalization as the number of pre-training tasks increases. We find that the smallest model capable of out-of-distribution generalization requires two transformer blocks, while for deeper models, the out-of-distribution generalization phase is \emph{transient}, necessitating early stopping. Finally, we perform an interpretability study of the pre-trained models, revealing highly structured representations in both attention heads and MLPs; and discuss the learned algorithms. Notably, we find an algorithmic shift in deeper models, as we go from few to many in-context examples.
title Learning to grok: Emergence of in-context learning and skill composition in modular arithmetic tasks
topic Machine Learning
Disordered Systems and Neural Networks
High Energy Physics - Theory
url https://arxiv.org/abs/2406.02550