Saved in:
Bibliographic Details
Main Author: Zhou, Andy
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2503.10617
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910919213187072
author Zhou, Andy
author_facet Zhou, Andy
contents Adapting large language models to multiple tasks can cause cross-skill interference, where improvements for one skill degrade another. While methods such as LoRA impose orthogonality constraints at the weight level, they do not fully address interference in hidden-state representations. We propose Compositional Subspace Representation Fine-tuning (CS-ReFT), a novel representation-based approach that learns multiple orthonormal subspace transformations, each specializing in a distinct skill, and composes them via a lightweight router. By isolating these subspace edits in the hidden state, rather than weight matrices, CS-ReFT prevents cross-task conflicts more effectively. On the AlpacaEval benchmark, applying CS-ReFT to Llama-2-7B achieves a 93.94% win rate, surpassing GPT-3.5 Turbo (86.30%) while requiring only 0.0098% of model parameters. These findings show that specialized representation edits, composed via a simple router, significantly enhance multi-task instruction following with minimal overhead.
format Preprint
id arxiv_https___arxiv_org_abs_2503_10617
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Compositional Subspace Representation Fine-tuning for Adaptive Large Language Models
Zhou, Andy
Computation and Language
Artificial Intelligence
Adapting large language models to multiple tasks can cause cross-skill interference, where improvements for one skill degrade another. While methods such as LoRA impose orthogonality constraints at the weight level, they do not fully address interference in hidden-state representations. We propose Compositional Subspace Representation Fine-tuning (CS-ReFT), a novel representation-based approach that learns multiple orthonormal subspace transformations, each specializing in a distinct skill, and composes them via a lightweight router. By isolating these subspace edits in the hidden state, rather than weight matrices, CS-ReFT prevents cross-task conflicts more effectively. On the AlpacaEval benchmark, applying CS-ReFT to Llama-2-7B achieves a 93.94% win rate, surpassing GPT-3.5 Turbo (86.30%) while requiring only 0.0098% of model parameters. These findings show that specialized representation edits, composed via a simple router, significantly enhance multi-task instruction following with minimal overhead.
title Compositional Subspace Representation Fine-tuning for Adaptive Large Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2503.10617