ReLope: KL-Regularized LoRA Probes for Multimodal LLM Routing

• URL: http://arxiv.org/abs/2603.24787v1
• Date: Wed, 25 Mar 2026 20:00:57 GMT
• Title: ReLope: KL-Regularized LoRA Probes for Multimodal LLM Routing
• Abstract summary: We introduce two complementary approaches for improving probe routing in large language model (LLM) systems.<n>First, we propose the emphAttention Probe, which aggregates hidden states from the preceding layer based on attention scores to recover distributed correctness signals.<n>Second, we present the emphKL-Regularized LoRA Probe (ReLope), which inserts a lightweight LoRA adapter and applies a KL regularizer to learn routing-aware representations.
• Score: 5.106806600400179
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Routing has emerged as a promising strategy for balancing performance and cost in large language model (LLM) systems that combine lightweight models with powerful but expensive large models. Recent studies show that \emph{probe routing}, which predicts the correctness of a small model using its hidden states, provides an effective solution in text-only LLMs. However, we observe that these probes degrade substantially when applied to multimodal LLMs (MLLMs). Through empirical analysis, we find that the presence of visual inputs weakens the separability of correctness signals in hidden states, making them harder to extract using standard probe designs. To address this challenge, we introduce two complementary approaches for improving probe routing in MLLMs. First, we propose the \emph{Attention Probe}, which aggregates hidden states from the preceding layer based on attention scores to recover distributed correctness signals. Second, we present the \emph{KL-Regularized LoRA Probe (ReLope)}, which inserts a lightweight LoRA adapter and applies a KL regularizer to learn routing-aware representations. Comprehensive experiments show that our methods consistently outperform baselines, suggesting that improving the quality of hidden states is key to effective routing in MLLMs. Our code is available at https://github.com/Spinozaaa/ReLope.

Related papers

• Rethinking LoRA for Privacy-Preserving Federated Learning in Large Models [14.755143405057929]
  Fine-tuning large vision models (LVMs) and large language models (LLMs) under differentially private learning (DPFL) is hindered by a fundamental privacy-utility trade-off.<n>Low-Rank Adaptation (LoRA), a promising parameter-efficient fine-tuning (PEFT) method, reduces computational and communication costs by introducing two trainable low-rank matrices while freezing pre-trained weights.<n>We propose LA-LoRA, a novel approach that decouples gradient interactions and aligns update directions across clients to enhance robustness under stringent privacy constraints.
  arXiv  Detail & Related papers  (2026-02-23T15:05:28Z)
• Memory-Based Advantage Shaping for LLM-Guided Reinforcement Learning [18.215893951726166]
  In environments with sparse or delayed rewards, reinforcement learning incurs high sample complexity.<n>This limitation has motivated the use of large language models (LLMs) for subgoal discovery and trajectory guidance.<n>We address these challenges by constructing a memory graph that encodes subgoals and trajectories from both LLM guidance and the agent's own successful rollouts.
  arXiv  Detail & Related papers  (2026-02-20T01:44:35Z)
• Routing Manifold Alignment Improves Generalization of Mixture-of-Experts LLMs [24.791817951102487]
  We show that aligning the manifold of routing weights with that of task embedding can effectively reduce the gap.<n>In experiments, we finetune routers in OLMoE, DeepSeekMoE, and Qwen3-MoE using RoMA.
  arXiv  Detail & Related papers  (2025-11-10T18:59:53Z)
• GainRAG: Preference Alignment in Retrieval-Augmented Generation through Gain Signal Synthesis [30.185213495829164]
  The Retrieval-Augmented Generation (RAG) framework introduces a retrieval module to dynamically inject retrieved information into the input context of large language models (LLMs)<n>We propose GainRAG, a novel approach that aligns the retriever's and LLM's preferences by defining a new metric, "gain", which measure how well an input passage contributes to correct outputs.<n>The experimental results on 6 datasets verify the effectiveness of GainRAG.
  arXiv  Detail & Related papers  (2025-05-24T14:14:57Z)
• Dynamic Low-Rank Sparse Adaptation for Large Language Models [54.1231638555233]
  Low-rank Sparse Adaptation (LoSA) is a novel method that seamlessly integrates low-rank adaptation into sparse LLM sparsity.<n>LoSA dynamically sparsifies the LoRA outcomes based on the corresponding sparse weights during fine-tuning.<n>LoSA can efficiently boost the efficacy of sparse LLMs within a few hours, without introducing any additional inferential burden.
  arXiv  Detail & Related papers  (2025-02-20T18:37:32Z)
• Universal Model Routing for Efficient LLM Inference [69.86195589350264]
  Model routing is a technique for reducing the inference cost of large language models (LLMs)<n>We propose UniRoute, a new approach to the problem of dynamic routing, where new, previously unobserved LLMs are available at test time.<n>We show that these are estimates of a theoretically optimal routing rule, and quantify their errors via an excess risk bound.
  arXiv  Detail & Related papers  (2025-02-12T20:30:28Z)
• $γ-$MoD: Exploring Mixture-of-Depth Adaptation for Multimodal Large Language Models [87.43596173378913]
  We propose an innovative strategy for existing MLLMs called $gamma$-MoD. In $gamma$-MoD, a novel metric is proposed to guide the deployment of MoDs in the MLLM. Based on ARank, we propose two novel designs to maximize the computational sparsity of MLLM.
  arXiv  Detail & Related papers  (2024-10-17T17:59:53Z)
• Get my drift? Catching LLM Task Drift with Activation Deltas [55.75645403965326]
  Task drift allows attackers to exfiltrate data or influence the LLM's output for other users.<n>We show that a simple linear classifier can detect drift with near-perfect ROC AUC on an out-of-distribution test set.<n>We observe that this approach generalizes surprisingly well to unseen task domains, such as prompt injections, jailbreaks, and malicious instructions.
  arXiv  Detail & Related papers  (2024-06-02T16:53:21Z)
• ResLoRA: Identity Residual Mapping in Low-Rank Adaption [96.59370314485074]
  We propose ResLoRA, an improved framework of low-rank adaptation (LoRA) Our method can achieve better results in fewer training steps without any extra trainable parameters or inference cost compared to LoRA. The experiments on NLG, NLU, and text-to-image tasks demonstrate the effectiveness of our method.
  arXiv  Detail & Related papers  (2024-02-28T04:33:20Z)
• The Unlocking Spell on Base LLMs: Rethinking Alignment via In-Context Learning [61.68787689234622]
  A recent study, LIMA, shows that using merely 1K examples for alignment tuning can achieve significant alignment performance as well. This raises questions about how exactly the alignment tuning transforms a base LLM. We show that the gap between tuning-free and tuning-based alignment methods can be significantly reduced through strategic prompting.
  arXiv  Detail & Related papers  (2023-12-04T00:46:11Z)

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.