MixLoRA-DSI: Dynamically Expandable Mixture-of-LoRA Experts for Rehearsal-Free Generative Retrieval over Dynamic Corpora

TL;DR

MixLoRA-DSI introduces a dynamic, resource-efficient method for updating generative retrieval models with new documents by selectively expanding experts based on out-of-distribution detection, avoiding full retraining.

Contribution

It proposes a novel expandable mixture-of-LoRA experts framework with an OOD-driven expansion strategy for continual retrieval model updates.

Findings

Outperforms full-model update baselines on NQ320k and MS MARCO Passage.

Achieves lower training costs with minimal parameter overhead.

Enables sublinear parameter growth through selective expert expansion.

Abstract

Continually updating model-based indexes in generative retrieval with new documents remains challenging, as full retraining is computationally expensive and impractical under resource constraints. We propose MixLoRA-DSI, a novel framework that combines an expandable mixture of Low-Rank Adaptation experts with a layer-wise out-of-distribution (OOD)-driven expansion strategy. Instead of allocating new experts for each new corpus, our proposed expansion strategy enables sublinear parameter growth by selectively introducing new experts only when significant number of OOD documents are detected. Experiments on NQ320k and MS MARCO Passage demonstrate that MixLoRA-DSI outperforms full-model update baselines, with minimal parameter overhead and substantially lower training costs.