Paper Detail

CD-LoRA: Consistency-Driven Low-Rank Adaptation for Multi-Task Fine-Tuning

Qian Zha, Jinda Liu, Yuan Wu, Yi Chang

arxiv Score 5.3

Published 2026-08-22 · First seen 2026-08-26

General AI

Abstract

While Multi-Task Learning (MTL) is essential for adapting Large Language Models (LLMs) to diverse domains, prevailing LoRA-based methods rely on complex routing mechanisms that partition task-specific knowledge. In this work, we reveal that such routing-based designs are prone to a training-inference discrepancy, where stochastic routing decisions under distribution shifts compromise inference stability. Driven by a second-order Taylor analysis that exposes the instability induced by routing variance, we challenge the training-inference discrepancy and propose Consistency-Driven Low-Rank Adaptation (CD-LoRA). By eliminating routers entirely, CD-LoRA employs a consistency-driven alignment mechanism to enforce representation congruence across tasks in a shared low-rank space. This paradigm fosters robust, task-agnostic features without explicit partitioning overhead. Extensive experiments show that CD-LoRA consistently outperforms state-of-the-art multi-adapter baselines, offering a simpler, router-free, and more stable solution for multi-task PEFT. The code is available at the anonymous link https://github.com/zhaqian21/CD-LoRA.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
later
Vote
Not set.
Saved
no
Collections
Not filed yet.
Next action
Not filled yet.

Reading Brief

No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.

Why It Surfaced

No ranking explanation is available yet.

Tags

No tags.

BibTeX

@article{zha2026cd,
  title = {CD-LoRA: Consistency-Driven Low-Rank Adaptation for Multi-Task Fine-Tuning},
  author = {Qian Zha and Jinda Liu and Yuan Wu and Yi Chang},
  year = {2026},
  abstract = {While Multi-Task Learning (MTL) is essential for adapting Large Language Models (LLMs) to diverse domains, prevailing LoRA-based methods rely on complex routing mechanisms that partition task-specific knowledge. In this work, we reveal that such routing-based designs are prone to a training-inference discrepancy, where stochastic routing decisions under distribution shifts compromise inference stability. Driven by a second-order Taylor analysis that exposes the instability induced by routing var},
  url = {https://arxiv.org/abs/2608.21909},
  keywords = {cs.LG},
  eprint = {2608.21909},
  archiveprefix = {arXiv},
}

Metadata

{}