More Than Catastrophic Forgetting: Integrating General Capabilities For Domain-Specific LLMs

TL;DR

This paper addresses the challenge of integrating general capabilities with domain-specific knowledge in LLMs, introducing a new framework called GCI and a method named ALoRA to improve performance on domain tasks.

Contribution

It introduces the concept of General Capabilities Integration (GCI) for harmonizing general and domain-specific skills in LLMs, along with the ALoRA method for dynamic information transfer.

Findings

GCI effectively combines general and domain-specific capabilities.

ALoRA improves information transfer and task performance.

Experiments validate the significance and effectiveness of the proposed approach.

Abstract

The performance on general tasks decreases after Large Language Models (LLMs) are fine-tuned on domain-specific tasks, the phenomenon is known as Catastrophic Forgetting (CF). However, this paper presents a further challenge for real application of domain-specific LLMs beyond CF, called General Capabilities Integration (GCI), which necessitates the integration of both the general capabilities and domain knowledge within a single instance. The objective of GCI is not merely to retain previously acquired general capabilities alongside new domain knowledge, but to harmonize and utilize both sets of skills in a cohesive manner to enhance performance on domain-specific tasks. Taking legal domain as an example, we carefully design three groups of training and testing tasks without lacking practicability, and construct the corresponding datasets. To better incorporate general capabilities across domain-specific scenarios, we introduce ALoRA, which utilizes a multi-head attention module upon LoRA, facilitating direct information transfer from preceding tokens to the current one. This enhancement permits the representation to dynamically switch between domain-specific knowledge and general competencies according to the attention. Extensive experiments are conducted on the proposed tasks. The results exhibit the significance of our setting, and the effectiveness of our method.