Saved in:
Bibliographic Details
Main Authors: Zindari, Ali, Jiang, Xiaowen, Mulayoff, Rotem, Stich, Sebastian U.
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.19028
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916024966709248
author Zindari, Ali
Jiang, Xiaowen
Mulayoff, Rotem
Stich, Sebastian U.
author_facet Zindari, Ali
Jiang, Xiaowen
Mulayoff, Rotem
Stich, Sebastian U.
contents Low-rank adaptation (LoRA) is a widely used parameter-efficient fine-tuning method, yet its learned correction is static: the same low-rank update is applied to every input. This input-agnostic approach creates an inevitable compromise between adapting to the fine-tuning distribution and preserving pre-trained behavior on inputs outside that distribution, contributing to catastrophic forgetting. We introduce DISeL (Dynamic Input-Sensitive LoRA), which augments LoRA modules with lightweight input-dependent gates over individual rank-one components. The gating mechanism is designed to preserve the pre-trained model's behavior by default, while training learns to activate selected components that reduce the fine-tuning loss. DISeL adds only a small number of parameters and preserves the low-rank structure. Across RoBERTa on GLUE, and Llama and Mistral models fine-tuned for mathematical reasoning and code generation, DISeL reduces forgetting relative to LoRA and related variants while maintaining competitive fine-tuning accuracy. In addition, the learned gate activations provide an interpretable diagnostic view of which layers and rank components are most activated during fine-tuning, giving insight into where task-specific adaptation is concentrated. Code available at https://github.com/alizindari/DISeL .
format Preprint
id arxiv_https___arxiv_org_abs_2605_19028
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Learning When to Adapt
Zindari, Ali
Jiang, Xiaowen
Mulayoff, Rotem
Stich, Sebastian U.
Machine Learning
Low-rank adaptation (LoRA) is a widely used parameter-efficient fine-tuning method, yet its learned correction is static: the same low-rank update is applied to every input. This input-agnostic approach creates an inevitable compromise between adapting to the fine-tuning distribution and preserving pre-trained behavior on inputs outside that distribution, contributing to catastrophic forgetting. We introduce DISeL (Dynamic Input-Sensitive LoRA), which augments LoRA modules with lightweight input-dependent gates over individual rank-one components. The gating mechanism is designed to preserve the pre-trained model's behavior by default, while training learns to activate selected components that reduce the fine-tuning loss. DISeL adds only a small number of parameters and preserves the low-rank structure. Across RoBERTa on GLUE, and Llama and Mistral models fine-tuned for mathematical reasoning and code generation, DISeL reduces forgetting relative to LoRA and related variants while maintaining competitive fine-tuning accuracy. In addition, the learned gate activations provide an interpretable diagnostic view of which layers and rank components are most activated during fine-tuning, giving insight into where task-specific adaptation is concentrated. Code available at https://github.com/alizindari/DISeL .
title Learning When to Adapt
topic Machine Learning
url https://arxiv.org/abs/2605.19028