GenKnowSub: Improving Modularity and Reusability of LLMs through General Knowledge Subtraction

TL;DR

This paper introduces GenKnowSub, a modular framework that disentangles general knowledge from task-specific adaptations in LLMs, leading to improved zero-shot generalization and performance across languages and models.

Contribution

It proposes a novel general knowledge subtraction method using residual LoRA modules, enhancing modularity and reusability of LLMs without additional training.

Findings

Consistent performance gains across multiple benchmarks.

Effective in monolingual and cross-lingual settings.

Generalizes to weaker LLMs.

Abstract

Large language models often struggle with zero-shot generalization, and several modular approaches have been proposed to address this challenge. Yet, we hypothesize that a key limitation remains: the entanglement of general knowledge and task-specific adaptations. To overcome this, we propose a modular framework that disentangles these components by constructing a library of task-specific LoRA modules alongside a general-domain LoRA. By subtracting this general knowledge component from each task-specific module, we obtain residual modules that focus more exclusively on task-relevant information, a method we call general knowledge subtraction (GenKnowSub). Leveraging the refined task-specific modules and the Arrow routing algorithm \citep{ostapenko2024towards}, we dynamically select and combine modules for new inputs without additional training. Our studies on the Phi-3 model and standard Arrow as baselines reveal that using general knowledge LoRAs derived from diverse languages, including English, French, and German, yields consistent performance gains in both monolingual and cross-lingual settings across a wide set of benchmarks. Further experiments on Phi-2 demonstrate how GenKnowSub generalizes to weaker LLMs. The complete code and data are available at https://github.com/saharsamr/Modular-LLM.