Mechanisms are Transferable: Data-Efficient Low-Resource Adaptation via Circuit-Targeted Supervised Fine-Tuning

TL;DR

This paper introduces CT-SFT, a method for low-resource language adaptation of large language models that selectively fine-tunes task-relevant attention heads, improving cross-lingual accuracy and reducing catastrophic forgetting.

Contribution

The paper proposes Circuit-Targeted Supervised Fine-Tuning (CT-SFT), a novel head-level parameter updating method that enhances low-resource language adaptation while preserving source language knowledge.

Findings

CT-SFT outperforms full fine-tuning in cross-lingual tasks.

Selective head updating reduces catastrophic forgetting.

Trade-off observed between circuit head editing and low-relevance head preservation.

Abstract

Adapting LLMs to low-resource languages is difficult: labeled data is scarce, full-model fine-tuning is unstable, and continued cross-lingual tuning can cause catastrophic forgetting. We propose Circuit-Targeted Supervised Fine-Tuning (CT-SFT): a counterfactual-free adaptation of CD-T (Contextual Decomposition Transformer) that uses a label-balanced mean baseline and task-directional relevance scoring to identify a sparse set of task-relevant attention heads in a proxy-language checkpoint, then transfer learns to a target language by updating only those heads (plus LayerNorm) via head-level gradient masking. Across NusaX-Senti and XNLI, CT-SFT improves cross-lingual accuracy over continued full fine-tuning while updating only a small subset of model parameters. We find an editing-preserving trade-off: harder transfers favor editing circuit heads, while easier transfers often favor near-zero (i.e., low-relevance heads) updates, preserving the source mechanism. CT-SFT also substantially reduces catastrophic forgetting, preserving proxy/source-language competence during transfer.