Learning to Select, Not Relearn: Hard-Routed Mixtures of Reasoning LoRAs

• URL: http://arxiv.org/abs/2606.31413v1
• Date: Tue, 30 Jun 2026 09:40:38 GMT
• Title: Learning to Select, Not Relearn: Hard-Routed Mixtures of Reasoning LoRAs
• Abstract summary: We propose a framework for composing frozen reasoning LoRA experts through unit-scale hard selection.<n>Experiments show that Hard-Routed MoR-LoRA preserves expert behavior while requiring substantially fewer trainable parameters than soft-routing mixture baselines.
• Score: 27.71932231964013
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Composing independently trained LoRA adapters into a single large language model is useful for multi-domain adaptation, especially when the original training data cannot be shared. A common approach is to use MoE-style routing over LoRA experts, but for frozen pretrained adapters, soft weighted combinations can change the unit-scale additive update under which each LoRA module was originally trained. We propose \textbf{Hard-Routed MoR-LoRA}, a two-stage framework for composing frozen reasoning LoRA experts through unit-scale hard selection. First, domain-specific LoRA adapters are trained independently using reinforcement learning from verifiable feedback to obtain reasoning experts. Then, all experts are frozen, reasoning traces are distilled from them, and only a lightweight shared router together with a small attention LoRA is trained for integration. The router selects exactly one expert per token using hard top-1 routing, while a straight-through estimator enables gradient-based training. Experiments across five benchmarks, multiple model scales, and additional model families show that Hard-Routed MoR-LoRA preserves expert behavior while requiring substantially fewer trainable parameters than soft-routing mixture baselines. Our analysis further shows that normalized soft mixtures often concentrate most routing mass on a single expert, suggesting that hard unit-scale routing provides a simple and efficient abstraction for frozen LoRA expert composition.

Related papers

• MoE$^2$-LoRA: When MoE Models Meet MoE-style Low-Rank Adaptation [23.49321952252126]
  Mixture-of-Experts (MoE) architectures have been widely adopted in large language models, yet parameter-efficient fine-tuning (PEFT) for MoE models remains underexplored.<n>We make the first attempt to fine-tune MoE models with MoE-style low-rank adaptation: our method, entitled MoE$2$-LoRA, deeply couples the pretrained expert specialization with task-specific adaptivity.
  arXiv  Detail & Related papers  (2026-07-24T04:50:39Z)
• ReMix: Reinforcement routing for mixtures of LoRAs in LLM finetuning [85.39146836671897]
  Low-rank adapters (LoRAs) are a parameter-efficient finetuning technique that injects trainable low-rank matrices into pretrained models to adapt them to new tasks.<n>Existing Mixture-of-LoRAs routers assign a learned routing weight to each LoRA to enable end-to-end training of the router.<n>We propose a new router designed that we call Reinforcement Routing for Mixture-of-LoRAs (ReMix)
  arXiv  Detail & Related papers  (2026-03-10T18:51:27Z)
• CoMoL: Efficient Mixture of LoRA Experts via Dynamic Core Space Merging [49.87105462292961]
  Core Space Mixture of LoRA (bfCoMoL) is a novel MoE-LoRA framework that incorporates expert diversity, parameter efficiency, and fine-grained adaptation.<n>CoMoL consistently outperforms existing methods across multiple tasks.
  arXiv  Detail & Related papers  (2026-02-28T09:40:11Z)
• FURINA: Free from Unmergeable Router via LINear Aggregation of mixed experts [17.056585698418587]
  Mixture of Experts (MoE) has been successfully integrated into Low-Rank Adaptation (LoRA) for parameter-efficient fine-tuning.<n>A key limitation of existing MoE-LoRA methods is their reliance on a discrete router.<n>We propose FURINA, a novel Free from Unmergeable Router framework based on the LINear Aggregation of experts.
  arXiv  Detail & Related papers  (2025-09-18T12:22:32Z)
• LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing [17.171872354057694]
  We propose the LoRA-Mixer, a modular and lightweight MoE framework that integrates LoRA experts.<n>Our core innovation lies in replacing the projection matrices of the attention module's input/output linear layers with task-specific LoRA experts.<n>LoRA-Mixer achieves significant improvements on datasets such as GSM8K, HumanEval, and MedQA.
  arXiv  Detail & Related papers  (2025-06-17T14:58:54Z)
• Each Rank Could be an Expert: Single-Ranked Mixture of Experts LoRA for Multi-Task Learning [53.053604713064544]
  Low-Rank Adaptation (LoRA) is widely used for adapting large language models (LLMs) to specific domains due to its efficiency and modularity.<n>Recent works adopt Mixture of Experts (MoE) by treating each LoRA module as an expert, thereby mitigating task interference through multiple specialized LoRA modules.<n>While effective, these methods often isolate knowledge within individual tasks, failing to fully exploit the shared knowledge across related tasks.<n>We propose Single-ranked Mixture of Experts LoRA (textbfSMoRA), which embeds MoE into LoRA by textittreating each rank as an
  arXiv  Detail & Related papers  (2025-01-25T06:56:39Z)
• MiLoRA: Efficient Mixture of Low-Rank Adaptation for Large Language Models Fine-tuning [9.91790333647256]
  Low-rank adaptation (LoRA) and its mixture-of-experts (MOE) variants are highly effective parameter-efficient fine-tuning (PEFT) methods. We propose Mixture of Low-Rank Adaptation (MiLoRA), a novel and efficient LoRA variant. MiLoRA differs from previous MOE-style LoRA methods by considering each LoRA module as an expert and employing a prompt-aware routing mechanism.
  arXiv  Detail & Related papers  (2024-10-23T17:04:40Z)
• Retrieval-Augmented Mixture of LoRA Experts for Uploadable Machine Learning [57.36978335727009]
  Low-Rank Adaptation (LoRA) offers an efficient way to fine-tune large language models (LLMs) In this paper, we propose a framework that adaptively retrieves and composes multiple LoRAs based on input prompts.
  arXiv  Detail & Related papers  (2024-06-24T05:24:41Z)
• MixLoRA: Enhancing Large Language Models Fine-Tuning with LoRA-based Mixture of Experts [3.6301530893494127]
  MixLoRA is an approach to construct a resource-efficient sparse MoE model based on LoRA. Our evaluations show that MixLoRA improves about 9% accuracy compared to state-of-the-art PEFT methods in multi-task learning scenarios.
  arXiv  Detail & Related papers  (2024-04-22T02:15:52Z)
• Mixture of LoRA Experts [87.50120181861362]
  This paper introduces the Mixture of LoRA Experts (MoLE) approach, which harnesses hierarchical control and unfettered branch selection. The MoLE approach achieves superior LoRA fusion performance in comparison to direct arithmetic merging.
  arXiv  Detail & Related papers  (2024-04-21T11:59:53Z)
• LoraRetriever: Input-Aware LoRA Retrieval and Composition for Mixed Tasks in the Wild [76.67343971195267]
  Low-Rank Adaptation (LoRA) provides an efficient solution for fine-tuning large language models (LLM) LoraRetriever is a retrieve-then-compose framework that adaptively retrieves and composes multiple LoRAs according to the input prompts. Experimental results indicate that LoraRetriever consistently outperforms the baselines.
  arXiv  Detail & Related papers  (2024-02-15T15:02:46Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.