Saved in:
Bibliographic Details
Main Authors: Yang, Liu, Lin, Ziqian, Lee, Kangwook, Papailiopoulos, Dimitris, Nowak, Robert
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.09240
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915105325711360
author Yang, Liu
Lin, Ziqian
Lee, Kangwook
Papailiopoulos, Dimitris
Nowak, Robert
author_facet Yang, Liu
Lin, Ziqian
Lee, Kangwook
Papailiopoulos, Dimitris
Nowak, Robert
contents In-context learning is a remarkable capability of transformers, referring to their ability to adapt to specific tasks based on a short history or context. Previous research has found that task-specific information is locally encoded within models, though their emergence and functionality remain unclear due to opaque pre-training processes. In this work, we investigate the formation of task vectors in a controlled setting, using models trained from scratch on synthetic datasets. Our findings confirm that task vectors naturally emerge under certain conditions, but the tasks may be relatively weakly and/or non-locally encoded within the model. To promote strong task vectors encoded at a prescribed location within the model, we propose an auxiliary training mechanism based on a task vector prompting loss (TVP-loss). This method eliminates the need to search for task-correlated encodings within the trained model and demonstrably improves robustness and generalization.
format Preprint
id arxiv_https___arxiv_org_abs_2501_09240
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Task Vectors in In-Context Learning: Emergence, Formation, and Benefit
Yang, Liu
Lin, Ziqian
Lee, Kangwook
Papailiopoulos, Dimitris
Nowak, Robert
Machine Learning
In-context learning is a remarkable capability of transformers, referring to their ability to adapt to specific tasks based on a short history or context. Previous research has found that task-specific information is locally encoded within models, though their emergence and functionality remain unclear due to opaque pre-training processes. In this work, we investigate the formation of task vectors in a controlled setting, using models trained from scratch on synthetic datasets. Our findings confirm that task vectors naturally emerge under certain conditions, but the tasks may be relatively weakly and/or non-locally encoded within the model. To promote strong task vectors encoded at a prescribed location within the model, we propose an auxiliary training mechanism based on a task vector prompting loss (TVP-loss). This method eliminates the need to search for task-correlated encodings within the trained model and demonstrably improves robustness and generalization.
title Task Vectors in In-Context Learning: Emergence, Formation, and Benefit
topic Machine Learning
url https://arxiv.org/abs/2501.09240