ORBIT: Preserving Foundational Language Capabilities in GenRetrieval via Origin-Regulated Merging

TL;DR

This paper introduces ORBIT, a method to prevent catastrophic forgetting in large language models during generative retrieval fine-tuning by regulating model weight drift.

Contribution

ORBIT actively tracks and constrains model weight changes to preserve foundational language capabilities during task-specific fine-tuning.

Findings

ORBIT outperforms baseline continual learning methods.

ORBIT maintains higher retrieval and language reasoning performance.

Model weight distance correlates with forgetting severity.

Abstract

Despite the rapid advancements in large language model (LLM) development, fine-tuning them for specific tasks often results in the catastrophic forgetting of their general, language-based reasoning abilities. This work investigates and addresses this challenge in the context of the Generative Retrieval (GenRetrieval) task. During GenRetrieval fine-tuning, we find this forgetting occurs rapidly and correlates with the distance between the fine-tuned and original model parameters. Given these observations, we propose ORBIT, a novel approach that actively tracks the distance between fine-tuned and initial model weights, and uses a weight averaging strategy to constrain model drift during GenRetrieval fine-tuning when this inter-model distance exceeds a maximum threshold. Our results show that ORBIT retains substantial text and retrieval performance by outperforming both common continual learning baselines and related regularization methods that also employ weight averaging.