Training-Side Stage-Aware Low-Rank Adaptation

Follow this topic Ask about this topic

Momentum

2 papers in the last four weeks, against 1 the four weeks before. 0.0% of all new papers.

Jul 6Week of Sep 21

Latest papers 39

Sep 29, 2026cs.AI

Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It

Pretrained transformers use little of their depth to follow references in context. Thirteen base models reliably follow only 1.4-3.6 lines, and extra pretrained loops add little. A task-trained rank-8 LoRA at one early layer extends this computation with all model weights frozen. Qwen3-8B improves from 15.5% to 99% exact accuracy on 24-line chains; a longer-trained LoRA reaches 50 lines. Ouro-1.4B reaches 60 lines after four loops and at least 160 after eight. The LoRA starts a relay: program lines pass on their chain identity through a short range of middle layers. Frozen heads read progressively further up the chain, and removing parent-line attention stops the relay. A frozen-model measurement locates the last useful intervention layer within tolerance in three of four held-out models. Task-specific LoRAs also improve MuSiQue. Default answers therefore understate the computation accessible through a tiny edit. Code and an interactive demo are available at https://lunamos.github.io/stop-thinking-too-early/

Zehao Jin, Ruixuan Deng, Junran Wang

Georgia Institute of Technology

Transformer Architectures Generative Pretrained Transformers Training-Side Stage-Aware Low-Rank Adaptation

Aug 31, 2026cs.CE

Lightweight Adaptation of EEG Foundation Models for Stroke Motor Imagery Decoding: Domain Shift and Subject-Level Robustness

Motor imagery (MI) electroencephalography (EEG) decoding could support post-stroke rehabilitation, but models developed on healthy cohorts may not transfer reliably to pathological EEG. We evaluated whether Low-Rank Adaptation (LoRA) can efficiently adapt three pretrained EEG foundation models (i.e., LaBraM-base, REVE-base, and REVE-large) for binary left- versus right-hand MI decoding. Frozen-backbone head-only baselines and LoRA adaptation were evaluated using subject-wise five-fold cross-validation on the PhysioNet EEG Motor Movement/Imagery Dataset and a binary subset of the UET175 dataset comprising 30 stroke participants. On EEGMMIDB, LoRA increased accuracy to 0.822 for LaBraM-base and 0.957 for REVE-base. On UET175, all head-only models performed near chance. With LoRA, LaBraM-base remained near chance (0.499

\pm

0.009), whereas REVE-base reached 0.847

\pm

0.194 and outperformed REVE-large (0.806

\pm

0.178), indicating that increased model capacity alone did not improve stroke-domain adaptation. The strongest stroke configuration, REVE-base LoRA, was further evaluated using within-cohort leave-one-subject-out cross-validation (LOOCV), showing 0.952 mean accuracy, but subject-wise accuracy ranged from 0.586 to 1.000, revealing a small low-performing tail. Zero-shot transfer from EEGMMIDB to UET175 remained near chance (0.464

\pm

0.072). These findings show that healthy-benchmark performance does not ensure transfer to stroke EEG. Translation of EEG foundation models to pseudo-online or real-time rehabilitation BCIs should therefore include target-domain adaptation and subject-level assessment of temporal informativeness, spatial sensitivity, and physiological discriminability.

Anh T. Nguyen, Zihua Sun, Michelle J. Johnson

School of Engineering and Applied Science, Department of Bioengineering, University of Pennsylvania, Philadelphia, PA, USA · School of Engineering and Applied Science, Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, PA, USA · Department of Physical Medicine and Rehabilitation, Department of Bioengineering, and Department of Mechanical Engineering and Applied Mechanics, University of Pennsylvania, Philadelphia, PA, USA

Electroencephalography Foundation Models Rehabilitation Electrocardiogram Foundation Models Training-Side Stage-Aware Low-Rank Adaptation

Aug 31, 2026cs.LG

RW-LoRA: Communication-Efficient Decentralized LoRA Fine-Tuning via Random Walks

Parameter-efficient fine-tuning methods such as LoRA have become a standard approach for adapting large foundation models. Adopting fine-tuning to distributed settings faces several challenges. Most existing distributed LoRA methods rely on centralized aggregation, and gossip-based decentralized LoRA requires repeated synchronization among multiple model copies. Both methods incur significant communication overhead and introduce errors due to simultaneous aggregation of multiple model updates. In this paper, we take a different perspective and propose a random-walk-based LoRA fine-tuning scheme. Instead of maintaining multiple model replicas, a single model token traverses the network and is updated sequentially using local fine-tuning objectives. This design eliminates the need for global synchronization, substantially reduces communication and computation costs, and avoids aggregation errors. We provide rigorous convergence guarantees for non-convex objectives under standard assumptions. Through empirical results on multiple NLP tasks and graph topologies, we show that the proposed method achieves competitive task performance with substantially less communication and computation than gossip-based LoRA.

Xingran Chen, Rohit Bhagat, Ghadir Ayache +3

Engineering Systems and Design Pillar, Singapore University of Technology and Design, Singapore 487372 · Department of Electrical and Computer Engineering, Rutgers University, Piscataway Township, NJ 08854, USA · LinkedIn, New York, NY 10118, USA +2

Hip-Lora Parameter-Efficient Fine-Tuning Methods Training-Side Stage-Aware Low-Rank Adaptation Model Fine-Tuning Decentralized Learning Wireless Foundation Models

Aug 10, 2026cs.LG

Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA

Macaron-V1 is an open agent-model family for experiential intelligence: learning from experience in real environments and continuing to learn after deployment. It is organized around two system goals. Adaptation is pursued through recursive improvement of versioned model-harness pairs, where experience from one configuration is evaluated under an external contract and used to construct its successor. Collaboration is pursued via the Mixture-of-LoRA (MoL) architecture that freezes a base model, composes specialist LoRA adapters, and selects one LoRA per user turn. The flagship Macaron-V1-Venti combines a 744B GLM-5.2 base with four LoRAs for chat, agent, coding, and GenUI; the Qwen3.6-based Macaron-V1-Tall (50B) uses the same design for local deployment. This report presents Macaron-V1 as a co-designed system spanning architecture, algorithms, and infrastructure. The MoL architecture supports continual learning through extensible LoRA specialists. The algorithm combines Model-Harness Co-design and recursive self-improvement loop, including the UI4A component-native GenUI harness, a stateful action substrate, versioned HCP contract, and the agentic RL framework MindForge. The supporting infrastructure includes the post-training platform MinT, the long-context RL method LongStraw, and stability techniques for sparse MoE and DSA base models. We evaluate Macaron-V1 on Personal Intelligence, GenUI, and general capability benchmarks against frontier baselines. Our results validate the current system, while compounding gains from continual learning and collective intelligence remain open questions.

Mind Lab, :, Vin Bo +74

Replay-Based Continual Learning Mixture-Of-Agents Continual Learning Multi-Lora Recursive Self-Improvement Training-Side Stage-Aware Low-Rank Adaptation

Jul 30, 2026cs.CL

Understanding Is Done Early: A Depth Division of Labor in Large Language Models and Its Use for Unbounded-Context Memory

Transformer depth is not used uniformly: lower and middle layers build semantic representations, while upper layers increasingly specialize them for prediction. We turn this division of labor into CoMem (Comprehension Memory), which writes each context chunk only through an intermediate layer, retrieves a fixed number of cached residual states, and recomputes the query-conditioned upper layers over the resulting pack. For a fixed retrieval budget, model-side read compute and memory are independent of stored-context length. We evaluate a continued-trained Qwen3-8B base LM under a unified chat-template-free protocol. The backbone is frozen; the flagship trains only a rank-32 self-distillation LoRA on plain PG19, and we report an adapter-free arm separately. CoMem reaches 97.05 on RULER and 38.27 on LoCoMo versus 34.59 for full-context KV-Direct; the dialogue-memory advantage survives conversation-cluster resampling and an independent judge. Results on additional long-context and long-document tasks expose both the benefits of bounded retrieval and its in-window compression tax. Controlled depth sweeps show that deeper caching lowers per-query recomputation but incurs a fidelity loss that self-distillation substantially repairs. In a separate adapter-free efficiency control on an NVIDIA H20 at 128k, CoMem uses 18.26 GB rather than 89.36 GB and achieves a 7.83x prefill speedup. These results show that long-context memory can be organized along the layer axis, not only the token axis.

Hanzuo Liu, Xuan Qi, Chunyu Liu +6

Tsinghua University · Tencent

Efficient Long-Context Inference Large Language Model Memory Conversational Memory Semantic Memory Cache Training-Side Stage-Aware Low-Rank Adaptation

Jul 24, 2026cs.CL

MoE $^2$ -LoRA: When MoE Models Meet MoE-style Low-Rank Adaptation

Mixture-of-Experts (MoE) architectures have been widely adopted in large language models, yet parameter-efficient fine-tuning (PEFT) for MoE models remains underexplored. Existing PEFT methods for MoE either ignore router priors with uniform adapters, reducing efficiency and risking forgetting, or rely on static expert selection, limiting per-token capacity and cross-expert feature learning. In this paper, 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 via a dual-channel Routing-Conditioned Projection (RCP) module, which reuses base router activations to inform LoRA routing. We further introduce a single global LoRA expert pool shared across all layers, enabling model-wide adaptation with emergent layer-wise affinities and balanced expert utilization. MoE

^2

-LoRA simultaneously benefits from the advantages of prior reuse, dynamic adapter routing, and model-wide knowledge sharing. Evaluated on multiple MoE backbones with varying scales and expert granularities, MoE

^2

-LoRA consistently achieves state-of-the-art downstream accuracy while retaining stronger general capabilities.

Qingyu Yang, Haonan He, Minglei Li +4

Shanghai Artificial Intelligence Laboratory · KTH Royal Institute of Technology · University of Science and Technology of China +2

Mixture-Of-Experts Large Language Models Training-Side Stage-Aware Low-Rank Adaptation Mixture-Of-Experts Mixture-Of-Experts Architectures

Jul 21, 2026cs.CL

Selective State-Space Adaptation and Retrieval for Language Model Reasoning

Low-rank adaptation introduces a static learned update applied identically to every input. The update provides task-level adaptation but does not explicitly represent token-level or instance-level state variation. A family of adapters is proposed that introduces selective state-space control at two complementary granularities. At the token level, MaLoRA (Mamba-modulated low-rank adaptation) makes the adapter's scaling factor a dynamic input-dependent function with recurrent state across tokens, in contrast to the stateless modulators of prior work. The token-level adapter improves over low-rank adaptation. On the other hand, it differentiates tokens by structural role but not by contextual relevance, which motivates placing evidence selection at the context level. At the context level, MaRA (Mamba Retrieval Adapter) tracks cross-segment reasoning state and selects the segments most relevant to the query. State-space controlled retrieval of approximately three million parameters exceeds an eight-billion-parameter dense retriever on supporting-paragraph recall. Although base models perform poorly on the task without adaptation (14 to 25 F1), MaRA recovers the evidence relevance latent in their representations. Across three frozen backbones and two multi-hop reasoning benchmarks, the end-to-end family improves reasoning accuracy on every cell of the 3-by-2 grid, by +6.4 F1 (+10.0% relative) on average over the LoRA baseline.

Atahan Dokme, Larry Heck

AI Virtual Assistant (AVA) Lab Georgia Institute of Technology

Large Language Model Adaptation Training-Side Stage-Aware Low-Rank Adaptation Efficient Task Adaptation State Space Models

Jul 8, 2026cs.CV

Stage-Aware Adaptation and Distribution Calibration for Subject-Driven Personalized Text-to-Image Generation

Subject-driven personalized text-to-image generation requires a pretrained diffusion model to acquire a specific subject from a few reference images while preserving subject identity, following novel text prompts, and maintaining sample diversity. Existing optimization-based methods instantiate subject adaptation through full fine-tuning, textual embedding optimization, or low-rank parameter updates; PaRa further constrains personalization from the perspective of parameter rank reduction. However, a uniform low-rank constraint or a uniform adapter strength cannot explicitly distinguish the capacity requirements of different denoising stages. Moreover, inference-time candidate selection driven mainly by identity similarity may compress the selected samples in the visual representation space. We decompose the problem into two complementary components: SPaRa denotes training-side stage-aware low-rank adaptation, DCAL denotes inference-side distribution-calibrated candidate selection, and SPaRa-DCAL denotes the combined framework. Theoretical analysis shows that timestep-dependent scaling controls the effective perturbation magnitude of a low-rank adapter, while identity-biased candidate selection restricts the radius of selected features around the reference center under explicit conditions. Auditable experiments under the SDXL and DreamBooth 30-subject protocol show that DCAL improves 1-LPIPS, CLIP-I, DINO-I, and CLIP-T on a fixed LoRA candidate pool, while revealing a clear trade-off with CLIP/DINO pairwise diversity and pairwise LPIPS. These results indicate that personalized generation should be evaluated through identity consistency, text alignment, and representation diversity rather than identity metrics alone.

Wenyan Xu, Alizer Wong

School of Computer Science, Guangdong University of Technology · School of Computer Science, Peking University · 3ManXis

Text-To-Image Training-Side Stage-Aware Low-Rank Adaptation Distributions

Jul 5, 2026cs.CV

RSLoRA: Training-free Rank Allocation for LoRA via Representational Sensitivity Probing

Low-Rank Adaptation (LoRA) has become a cornerstone of parameter-efficient fine-tuning (PEFT); however, the conventional practice of uniform rank assignment ignores the functional heterogeneity of neural layers. Existing rank allocation methods typically struggle with a trade-off between computational intensity and heuristic simplicity: training-based methods suffer from prohibitive overhead, while pre-allocation methods fail to capture the dynamic task-specific representation manifold. In this paper, we propose RSLoRA (Representational Sensitivity LoRA), a training-free and gradient-free rank allocator driven by activation-space geometry. We identify a "sensitivity regime shift" across layers, observing that static weight analysis and local gradients are insufficient to reflect how updates reshape a model's internal representations. To address this, RSLoRA introduces a virtual representational probing mechanism. By simulating adaptation through structured low-rank noise and measuring the resulting manifold displacement by using Effective Rank and Frechet Distance, we identify high-sensitivity modules that require higher rank capacity. Our framework effectively bridges the gap between expert-crafted heuristics and actual representational impact. Extensive evaluations demonstrate that RSLoRA consistently outperforms state-of-the-art allocators (e.g., AdaLoRA, GoRA) across mainstream benchmarks. By eliminating the need for iterative training-time adjustments and backward gradients, RSLoRA provides a highly efficient, robust, and representation-aware solution for large-scale model adaptation.

Jiaqi Liu, Haidong Kang, Qihui Zhao +1

School of Computer Science and Technology Dalian University of Technology · Hebei Key Laboratory of Marine Perception Network and Data Processing Northeastern University · School of Computer and Communication Engineering Northeastern University +1

Training-Side Stage-Aware Low-Rank Adaptation Parameter-Efficient Adaptation Training-Free

Jul 3, 2026cs.LG

Co-Adaptive Multi-Task LoRA: Transfer-Aware, Label-Free Control of Domain Participation

Fine-tuning a single low-rank adapter on many domains at once is multi-task learning: the domains must be co-learned, and how they share the adapter decides whether they help or hurt one another. Most efficient fine-tuning pipelines ignore this and train on a fixed, uniform mixture, leaving two coupled questions unanswered: how much should each domain participate, and which domains should be co-trained given that some transfer positively and others interfere? We show that both answers can be read off cheaply and without labels. A forward pass of the current shared adapter over a small unlabeled probe yields, per domain, a competence signal whose level tracks remaining headroom and whose trajectory tracks learning speed; the drift of these probe representations yields a signed cross-domain affinity that predicts pairwise transfer. We fold both into CoDA, a co-adaptive controller that solves a small entropy-regularized quadratic program on the simplex to set each domain's participation -- jointly its loss weight and its share of the sampled data -- rewarding high-headroom, still-learning, mutually synergistic domains and damping interfering ones. The controller is forward-only, adds no trainable parameters, and wraps any multi-task LoRA pipeline. Across five heterogeneous domains and two backbones, CoDA improves the average over uniform mixing, learned mixtures, gradient-surgery multi-task optimizers, and online data selection while using half the data, and lowers cross-domain gradient conflict. We prove that the competence signal tracks domain risk, that the participation program has a unique fixed point reached by a contraction, and that its solution performs transfer-aware water-filling; analysis, ablations, and controls corroborate each claim.

Wei Zhang, Lin Tang, Ming Zhao +1

1Sichuan University, Chengdu, China · University of Electronic Science and Technology of China, Chengdu, China

Domain Adaptation Training-Side Stage-Aware Low-Rank Adaptation Low-Rank Adaptation Adapters Adaptive Control Participation Label-Free+1

Jul 3, 2026cs.LG

Stacked LoRA for Subject-Adaptive EEG Foundation Models in Motor Imagery Decoding

Electroencephalography (EEG) decoding for brain-computer interfaces (BCIs) faces a major challenge: substantial inter-subject variability limits effective cross-subject generalization. Consequently, practical systems still rely largely on subject-specific models trained from scratch and requiring individual recalibration. EEG foundation models have recently emerged as a promising alternative; however, even large pretrained models cannot simply be used as fixed feature extractors and still require additional adaptation before they can be reliably applied to downstream tasks. In this work, we address this challenge through targeted adaptation strategies. Building on recent EEG foundation models such as REVE, LaBraM, and LUNA, we examine the impact of different low-rank adaptation strategies on motor imagery classification. We propose a framework that structurally decouples subject-invariant knowledge from subject-specific neural signatures: the low-rank update at each adapted layer is split into a Global adapter, trained jointly across all subjects, and Subject-Specific adapters, each absorbing individual variability. To assess the contribution of each path, we compare three adaptation strategies: (i) subject-specific LoRA (ii) global LoRA and (iii) stacked LoRA, combining both Global and Subject Specific adapters. Experiments on BCI Competition IV-2a, PhysioNet Motor Imagery, and the clinical Zuo2025 benchmark show that Stacked LoRA effectively mitigates inter-subject variability, achieving the best accuracy in the large majority of backbone and dataset combinations. Our analysis further reveals that the optimal balance between the global and subject-specific paths depends on the target population: a shared adapter is sufficient for large, diverse cohorts, whereas subject-specific adaptation is decisive under the high inter-session variability of clinical recordings.

Aymen Sarhane, Fouad Lbakali, Mouad Souissi +2

IMT Atlantique, Lab-STICC, Brest, France

Electroencephalography Foundation Models Brain-Computer Interface Electroencephalography Training-Side Stage-Aware Low-Rank Adaptation

Jul 2, 2026cs.LG

Bayesian Sparse Low-Rank Adaptation for Large Language Model Uncertainty Estimation

Large language models (LLMs) exhibit remarkable reasoning capabilities, but their task-specific fine-tuning is notoriously plagued by overconfidence, severely hindering trustworthy deployment. We propose Data-Adaptive Lower-Rank Adaptation (DALorRA), a simple and effective variational Bayesian sparse framework that shifts the paradigm of uncertainty quantification from the dense parameter space to the lightweight rank level of low-rank adaptation (LoRA). With the insight that LoRA essentially aggregates multiple rank-one components that may provide superfluous model capacity, DALorRA imposes stochastic masking on rank dimensions, enabling Bayesian regularization of model capacity during training and ensemble-like calibration during inference. Extensive experiments demonstrate DALorRA's excellent calibration of LLMs without compromising reasoning accuracy.

Jijie Zhang, Zhe Ren, Quan Zhang +1

School of Artificial Intelligence, Jilin University · Michigan State University

Large Language Model Uncertainty Training-Side Stage-Aware Low-Rank Adaptation Large Language Model Adaptation Uncertainty Bayesian Networks

Jun 26, 2026cs.CV

Monocular Avatar Reconstruction via Cascaded Diffusion Priors and UV-Space Differentiable Shading

Reconstructing high-fidelity, relightable 3D avatars from a single in-the-wild image is a challenging ill-posed problem, primarily hindered by the scarcity of high-quality PBR data and the complexity of disentangling illumination from intrinsic materials. In this paper, we present a data-efficient framework that leverages the robust priors of a unified pre-trained diffusion backbone to sequentially address texture completion, delighting, and material decomposition. Unlike existing methods that rely on fragmented pipelines or extensive proprietary datasets, we utilize cascaded Low-Rank Adaptations (LoRAs) to adapt the strong generative prior of the diffusion model for each sub-task in UV space. Specifically, we first employ an Inpainting LoRA to complete missing UV textures caused by occlusion, leveraging the model's semantic understanding to generate semantically and photometrically coherent details. Subsequently, a Light-Homogenization LoRA and a novel Cross-Intrinsic Attention mechanism are introduced to remove baked-in lighting and collaboratively synthesize pixel-aligned PBR maps (Albedo, Normal, Roughness, Specular, and Displacement). To ensure physical plausibility, we impose a UV-space differentiable BRDF shading loss during the decomposition stage, forcing the generative process to adhere to the rendering equation without the artifacts typical of rasterization-based supervision. Extensive experiments demonstrate that our method, trained on fewer than 100 real 3D scans, generates comprehensive, 4K-resolution PBR assets with superior realism and generalization compared to state-of-the-art methods, and all training code and model weights will be released upon acceptance.

Hong Li, Minqi Meng, Yanjun Liang +10

Avatar Reconstruction Diffusion Priors Texture Raymap-Induced Geometry Training-Side Stage-Aware Low-Rank Adaptation

Jun 26, 2026cs.LG

MixTTA: Low-Rank Cross-Channel Mixing for Reliable Test-Time Adaptation

Test-Time Adaptation (TTA) methods commonly update the affine parameters of normalization layers to adapt deployed models under distribution shifts. However, per-channel affine parameters perform axis-aligned scaling and shifting, making them geometrically incapable of correcting cross-channel structural changes induced by distribution shift. To address this limitation, we propose MixTTA, a lightweight plug-in module that equips normalization layers with a low-rank cross-channel transformation, enabling inter-channel mixing at each layer. To ensure that the low-rank branch captures only cross-channel interactions, we also propose Decoupling Projection that enforces strict separation from the diagonal affine path, along with Spectral Projection that prevents rank-1 collapse under non-stationary test streams. MixTTA can be seamlessly integrated into any existing normalization-based TTA method. Experiments in both standard and wild TTA settings show consistent improvements over strong baselines while mitigating adaptation failure under challenging conditions. The source code is publicly available at https://github.com/delta6189/MixTTA.

Mansoo Jung, Youngwook Kim, Jungwoo Lee

Seoul National University, Seoul, Republic of Korea · Kookmin University, Seoul, Republic of Korea · HodooAI Labs, Seoul, Republic of Korea

Stable Test-Time Adaptation Test-Time Adaptation Training-Side Stage-Aware Low-Rank Adaptation Mix Low-Rank Structure Sequential Scaling

Jun 26, 2026cs.CV

TextDS: Parameter-Efficient Representation Alignment for Scene Text Detection under Distribution Shifts

In real-world deployments, scene text detectors inevitably face distribution shifts beyond the training distribution. Prior work often depends on large-scale scene-text pretraining, yet evaluation under cross-domain changes and real-world imaging degradations remains limited. We propose TextDS, an efficient framework for scene text detection under distribution shifts. First, we propose a data-efficient dual-encoder design with visual foundation models, eliminating the reliance on large-scale scene-text pretraining. Second, we introduce Step-wise LoRA adaptation (SWLoRA), which performs progressive low-rank refinement with a dynamic early-exit mechanism for effective feature adaptation. Third, we propose Common Subspace Fusion (CSF) to align and fuse the two branches in a shared subspace while retaining complementary, shift-robust information. Finally, we construct adverse-condition scene text detection datasets to address the gap in evaluating under imaging degradation. Experiments show that TextDS achieves competitive performance in scene text detection, demonstrating robustness across domains and adverse imaging conditions with only 4.9M trainable parameters.

Boyuan Chen, Zichen Dang, Chuang Yang +2

School of Electrical Engineering, Xi’an Jiaotong University · Department of Electrical and Electronic Engineering, The Hong Kong Polytechnic University

Scene Text Recognition Image-Text Alignment Training-Side Stage-Aware Low-Rank Adaptation Visual Degradation Distribution Shifts Subspace

Jun 11, 2026cs.CV

Prompt2Effect: Training-Free Image-to-Video Model Specialization via LoRA Generation

While personalizing Image-to-Video (I2V) diffusion models with specific visual effects is increasingly demanded for high-end generation, current practice requires training a separate Low-Rank Adaptation (LoRA) module for each effect, incurring substantial data curation and iterative optimization costs that hinder interactive control. We present Prompt2Effect, a weight-driven hypernetwork that amortizes per-effect training by directly synthesizing effect-specific LoRA weights in a single forward pass. Unlike prior hypernetworks that regress adapter weights purely from semantics, Prompt2Effect is explicitly conditioned on the frozen base model weights, grounding prediction in the structural geometry of each layer. Furthermore, instead of predicting raw LoRA matrices, we introduce an SVD-canonicalized parameterization that resolves factorization ambiguity and stabilizes large-scale synthesis. Extensive experiments demonstrate that Prompt2Effect achieves on-par or superior video quality and effect alignment compared to conventional LoRA fine-tuning, while reducing the computational cost from 56 GPU training hours to 3.3 seconds of hypernetwork inference. When used as initialization for subsequent fine-tuning, our predicted weights further improve final performance and accelerate optimization by approximately 10x.

Xiaomeng Yang, Yanyu Li, Gordon Guocheng Qian +7

1Northeastern University, 2Snap Inc.

Training-Side Stage-Aware Low-Rank Adaptation Training-Free Hypernetwork Forward Propagation

Jun 1, 2026cs.CV

Low-Rank Adaptation of Frozen Vision-Language Models for Blind Image Quality Assessment

Blind image quality assessment (BIQA) predicts perceived image quality without access to a pristine reference and is fundamental to applications such as image compression, transmission, and restoration. Recent BIQA methods increasingly rely on large vision-language models (VLMs). Although frozen VLMs provide an efficient alternative to computationally expensive full fine-tuning, it remains unclear how much performance is sacrificed by not adapting the backbone and, more importantly, under what conditions such adaptation is truly beneficial. Answering this question, however, is complicated by the widespread use of image-level splitting on synthetic-distortion benchmarks, where distorted versions of the same reference image can appear in both training and test partitions. This content overlap artificially inflates the apparent performance of frozen representations, masking their true generalization ability and potentially leading to incorrect conclusions about the value of backbone adaptation. We therefore address these two issues jointly. We develop an efficient BIQA framework that fuses a natural-scene-statistics descriptor with frozen SigLIP and CLIP-H embeddings through a lightweight regression head, and then apply parameter-efficient Low-Rank Adaptation (LoRA) to the SigLIP backbone, training only

0.23\%

of its parameters. Evaluating both frozen and adapted models across six datasets under image-level and reference-level protocols, we find that image-level splitting inflates frozen-feature SROCC by up to

0.44

and masks wide variation in true difficulty, which reference-level evaluation reveals. Under this content-independent protocol, LoRA adaptation recovers performance in proportion to the exposed difficulty, with the largest gains where frozen features generalize poorly (up to

+0.357

SROCC on TID2013) and little benefit where they are already strong.

Bishr Omer Adam, Xu Li

School of Electronics and Information, Northwestern Polytechnical University, Xi’an, China

Image Quality Assessment Frozen Vision-Language Models Training-Side Stage-Aware Low-Rank Adaptation Distortion

May 29, 2026cs.LG

Balanced LoRA: Removing Parameter Invariance to Accelerate Convergence

Low-Rank Adaptation (LoRA) is the most widely adopted method for fine-tuning large language models. Notably, LoRA is inherently overparameterized: multiple pairs of low-rank factors can yield the same adapted weight matrix. We show--both theoretically and empirically--that these pairs exhibit significantly different condition numbers. As a result, converging to different loss minimizers directly impacts the convergence rate of LoRA. Building on this observation, we introduce Balanced Low-Rank Adaptation (BaLoRA), a variant of LoRA that projects iterates onto a balanced manifold. This manifold improves the conditioning of the loss landscape while preserving the adapted matrix. The projection step is computationally lightweight and integrates seamlessly into existing fine-tuning pipelines. Empirically, BaLoRA converges faster than standard LoRA and achieves superior performance across a range of fine-tuning tasks.

Valérie Castin, Kimia Nadjahi, Pierre Ablin +1

1 ´Ecole Normale Sup´erieure PSL, Paris, France · 2CNRS · 3Apple, Paris, France.

Training-Side Stage-Aware Low-Rank Adaptation Low-Rank Structure Convergence

May 25, 2026cs.CV

CollectionLoRA: Collecting 50 Effects in 1 LoRA via Multi-Teacher On-Policy Distillation

Customized image editing aims to equip pre-trained diffusion models with specific visual effects using limited paired data, typically via Low-Rank Adaptation (LoRA). As the number of desired effects grows, storing and dynamically loading numerous these effect LoRAs significantly increases deployment overhead. Furthermore, current pipelines typically cascade these effect LoRAs with acceleration modules for fast generation, which triggers severe parameter interference and results in concept bleeding and style degradation. We propose CollectionLoRA, a multi-teacher on-policy distillation framework capable of distilling the concepts of up to 50 different effect LoRAs along with few-step generation capabilities into a single LoRA. This fundamentally resolves the feature interference issue and significantly reduces deployment costs. Specifically, the method introduces (i) a Probabilistic Dual-Stream Routing mechanism that enables the model to randomly switch between data sources during training, effectively enhancing its generalization in unseen scenarios; (ii) an Asymmetric Orthogonal Prompting strategy to achieve concept isolation within the prompt space; (iii) a Coarse-to-Fine Distillation Objective to mitigate the distribution gap between the teacher and student models. Extensive evaluations show that CollectionLoRA distills all customized effects and few-step generation into a single LoRA, reducing deployment overhead while achieving concept fidelity comparable to or better than independently trained teacher models. Code: https://github.com/Qwen-Applications/CollectionLoRA

Fangtai Wu, Hailong Guo, Shijie Huang +7

Zhejiang University · Qwen Applications Business Group of Alibaba · Xi’an Jiaotong University

Multi-Lora Training-Side Stage-Aware Low-Rank Adaptation Few-Step Generation Image Editing Impact Analysis

May 18, 2026cs.LG

Learning When to Adapt

Low-rank adaptation (LoRA) is a widely used parameter-efficient fine-tuning method, yet its learned correction is static: the same low-rank update is applied to every input. This input-agnostic approach creates an inevitable compromise between adapting to the fine-tuning distribution and preserving pre-trained behavior on inputs outside that distribution, contributing to catastrophic forgetting. We introduce DISeL (Dynamic Input-Sensitive LoRA), which augments LoRA modules with lightweight input-dependent gates over individual rank-one components. The gating mechanism is designed to preserve the pre-trained model's behavior by default, while training learns to activate selected components that reduce the fine-tuning loss. DISeL adds only a small number of parameters and preserves the low-rank structure. Across RoBERTa on GLUE, and Llama and Mistral models fine-tuned for mathematical reasoning and code generation, DISeL reduces forgetting relative to LoRA and related variants while maintaining competitive fine-tuning accuracy. In addition, the learned gate activations provide an interpretable diagnostic view of which layers and rank components are most activated during fine-tuning, giving insight into where task-specific adaptation is concentrated. Code available at https://github.com/alizindari/DISeL .

Ali Zindari, Xiaowen Jiang, Rotem Mulayoff +1

1Universität des Saarlandes · 2CISPA Helmholtz Center for Information Security

Training-Side Stage-Aware Low-Rank Adaptation

May 16, 2026cs.CV

GLT-PEFT: Gated Lie-Tucker Parameter-Efficient Fine-Tuning for Alzheimer's Disease Diagnosis with Hippocampal Segmentation Pretraining

Parameter-efficient fine-tuning (PEFT) has emerged as a promising paradigm for adapting pretrained models under limited data conditions. However, most existing PEFT methods are designed for matrix-structured parameters and are not well suited for high-dimensional convolutional kernels in medical imaging models. Moreover, they typically rely on additive updates and lack mechanisms to preserve the geometric structure of pretrained parameters, while multiplicative (geometry-aware) updates are difficult to integrate within a unified framework. To address this issue, this paper proposes GLT-PEFT, a gated Lie-Tucker parameter-efficient fine-tuning framework for Alzheimer's disease (AD) diagnosis. The proposed approach transfers a hippocampal segmentation pretrained model to a downstream classification task. Tucker decomposition enables tensor-aware low-rank adaptation of 3D convolutional kernels, while Lie group-based transformations provide structure-preserving multiplicative updates. A gating mechanism further reconciles additive and multiplicative update forms, resulting in a unified and more stable fine-tuning strategy. Extensive experiments demonstrate that GLT-PEFT achieves effective cross-task transfer while significantly reducing trainable parameters, highlighting its effectiveness for efficient and robust adaptation in medical imaging models.

Guanghua He, Hancan Zhu, Gaohang Yu +1

Department of Mathematics, Hangzhou Dianzi University, Hangzhou, 310018, China · School of Mathematics, Physics and Information, Shaoxing2026 University, Shaoxing, 312000, China · Department of Mathematics, Zhejiang University of Science and Technology, Hangzhou, 310023, China

Parameter-Efficient Fine-Tuning Methods Training-Side Stage-Aware Low-Rank Adaptation Alzheimer Multimodal Pretraining Confabulation

May 15, 2026cs.LG

Strategic Over-Parameterization for Generalizable Low-Rank Adaptation

Adapting large language models (LLMs) to downstream tasks via full fine-tuning is increasingly impractical due to its computational and memory demands. Parameter-efficient fine-tuning (PEFT) approaches such as Low-Rank Adaptation (LoRA) mitigate this by confining updates to a compact set of trainable parameters, but this aggressive reduction often sacrifices generalization, especially under transfer across heterogeneous tasks and domains. We revisit the tension between parameter efficiency and adaptation capacity, and ask whether the two are truly at odds. We answer in the negative by introducing LoRA-Over, a framework grounded in a simple principle: enrich the optimization landscape during training, then collapse the enrichment at inference. LoRA-Over injects auxiliary parameters into the low-rank adapters during training to broaden the effective hypothesis space, and through a decomposition-based reformulation folds them back into a standard low-rank structure with negligible reconstruction error, keeping inference cost identical to vanilla LoRA. Since not all weight matrices benefit equally from added capacity, we further propose two scheduling strategies, one statically predefined and one dynamically determined at runtime, that direct extra capacity where most needed. We evaluate LoRA-Over on language understanding (GLUE, T5-Base), dialogue (MT-Bench), arithmetic reasoning (GSM8K), and code generation (HumanEval), using LLaMA 2-7B and LLaMA 3.1-8B. Across all benchmarks and scales, LoRA-Over consistently outperforms vanilla LoRA, showing that principled over-parameterization designed to vanish at inference is an effective lever for improving PEFT generalization. Code will be released upon acceptance.

Jing Gao, Zhong-Yi Lu, Pan Zhang +1

School of Fundamental Physics and Mathematical Sciences, Hangzhou Institute for Advanced Study, UCAS, Hangzhou 310024, China · School of Physical Sciences, University of Chinese Academy of Sciences, No. 19A Yuquan Road, Beijing 100049, China · CAS Key Laboratory of Theoretical Physics, Institute of Theoretical Physics, Chinese Academy of Sciences, Beijing 100190, China +1

Training-Side Stage-Aware Low-Rank Adaptation Low-Rank Adaptation Adapters Large Language Model Adaptation Overparameterization Rapid Adaptation

May 14, 2026cs.LG

GPart: End-to-End Isometric Fine-Tuning via Global Parameter Partitioning

Low-rank adaptation (LoRA) has become a dominant paradigm for parameter-efficient fine-tuning (PEFT) of large-scale deep learning models. However, its bilinear parameterization induces a parameter-dependent geometry: the mapping from trainable parameters to weight updates is not generally distance-preserving. Related methods that project a low-dimensional vector into LoRA's parameter space, such as Uni-LoRA, improve parameter efficiency, but the subsequent bilinear map breaks end-to-end isometry. We propose GPart (Global Partition fine-tuning), a highly parameter-efficient fine-tuning method that maps a

d

-dimensional trainable vector directly into the full weight space through a sparse, isometric partition matrix. GPart retains a fixed global parameter-sharing prior while removing the additional low-rank reconstruction used by LoRA-based methods. This yields a simple parameterization with a single main hyperparameter (

d

), exact end-to-end isometry, and a minimal checkpoint representation consisting of the trainable vector and a random seed. GPart builds on the premise of effective fine-tuning within random low-dimensional subspaces of the full weight space without requiring a low-rank matrix factorization. Across natural language understanding, computer vision, and mathematical reasoning benchmarks, GPart matches or improves over existing PEFT methods at ultra-low parameter budgets. Beyond offering mathematical tractability and memory efficiency, the direct linear parameterization of GPart streamlines model selection and paves the way for compact adapter composition. Overall, GPart provides an elegant and competitive alternative for fine-tuning under small parameter budgets, with a fixed and predictable geometry between trainable coordinates and weight-space updates.

Paolo Mandica, Michał Brzozowski, Zuzanna Dubanowska +1

Samsung AI Center, Warsaw, Poland · University of Warsaw, Poland

Parameter-Efficient Fine-Tuning Methods Training-Side Stage-Aware Low-Rank Adaptation Model Fine-Tuning Partitioning

May 13, 2026cs.LG

Continual Fine-Tuning of Large Language Models via Program Memory

Parameter-Efficient Fine-Tuning (PEFT), particularly Low-Rank Adaptation (LoRA), has become a standard approach for adapting Large Language Models (LLMs) under limited compute. However, in continual settings where models are updated sequentially with small datasets, conventional LoRA updates struggle to balance rapid adaptation and knowledge retention. Existing methods typically treat the low-rank space as a homogeneous update region, lacking mechanisms to regulate how short-term updates are consolidated over time. We propose a continual LoRA framework with \textbf{Pro}gram memory, inspired by \textbf{C}omplementary \textbf{L}earning Systems in neuroscience. Our approach, dubbed \textbf{ProCL}, organizes LoRA adapters into structured program memory slots that are dynamically retrieved through input-conditioned attention. This enables rapid and localized adaptation, encouraging similar inputs to reuse shared adapter regions while reserving unused capacity for future data. The slots are then combined with the underlying adapter, which maintains a distributed representation that gradually accumulates knowledge across tasks to balance plasticity and stability. Our method operates entirely within the LoRA parameterization and incurs no additional inference cost. Experiments on diverse benchmarks demonstrate improved retention and reduced catastrophic forgetting over other continual LoRA strategies.

Hung Le, Svetha Venkatesh

Deakin Applied AI Initiative, Deakin University, Australia

Training-Side Stage-Aware Low-Rank Adaptation Replay-Based Continual Learning Catastrophic Forgetting Model Fine-Tuning Continual Fine-Tuning Peak Memory+1

May 11, 2026cs.LG

HELLoRA: Hot Experts Layer-Level Low-Rank Adaptation for Mixture-of-Experts Models

Low-Rank Adaptation (LoRA) dominates parameter-efficient fine-tuning of large language models, yet most variants target dense architectures. Mixture-of-Experts (MoE) models scale parameters at near-constant per-token compute, and their sparse activation patterns create untapped opportunities for more efficient adaptation. We propose Hot-Experts Layer-level Low-Rank Adaptation (HELLoRA), which attaches LoRA modules only to the most frequently activated experts at each layer. This simple mechanism reduces trainable parameters and adapter-induced FLOPs while improving downstream performance, an effect we attribute to a form of structured regularization that preserves pretrained expert specialization. To stress-test HELLoRA under extreme parameter budgets, we further compose it with LoRI to form HELLoRI, which freezes the up-projection and sparsifies the down-projection. Across three MoE backbones, namely OlMoE-1B-7B, Mixtral-8x7B, and DeepSeekMoE, and three task families covering mathematical reasoning, code generation, and safety alignment, HELLoRA consistently outperforms strong PEFT baselines. Relative to vanilla LoRA on OlMoE, HELLoRA uses 15.7% of the trainable parameters, reduces adapter FLOPs by 38.7%, achieves 1.9x the training throughput, and improves accuracy by 9.2%. On DeepSeekMoE, HELLoRA outperforms LoRA while using only 23.2% of its trainable parameters. These results demonstrate that activation-aware adapter placement is an effective and practical route to scaling PEFT for MoE language models.

Jia Wei, Zhonghao Zhang, Ping Chen +5

Department of Computer Science and Technlogy Tsinghua University Beijing 100084, China · School of Computer Science and Technlogy Xi’an Jiaotong University · The State Key Laboratory of Blockchain and Data Security, Zhejiang University Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security

Training-Side Stage-Aware Low-Rank Adaptation Mixture-Of-Experts Large Language Models Large Language Model Adaptation Mixture-Of-Experts Experts Deepseek

May 9, 2026cs.CV

Gate-and-Merge: Zero-shot Compositional Personalization of Vision Language Models

This paper tackles compositional personalization of vision-language models (VLMs). In this problem, multiple user-defined concepts must be recognized or described jointly at test time. We introduce Gate-and-Merge, a zero-shot framework that enables compositional personalization without the need for co-occurrence training. During personalization, each concept is learned independently as a lightweight LoRA adapter, paired with a concept token. The base model remains unchanged and concepts are kept disentangled. At inference, we enable composition by merging concept-specific LoRA updates directly in weight space. To suppress irrelevant activations and prevent interference, a gating mechanism is employed to estimate textual and visual cues and select only the modules that contribute to the prediction. We further stabilize composition by combining only the most meaningful and mutually consistent updates, helping preserve each concept's identity. Our quantitative and qualitative analyses show consistent gains in performance across multiple personalization tasks in both single-concept and compositional settings.

Guodong Ding, Angela Yao

National University of Singapore

Personalization Training-Side Stage-Aware Low-Rank Adaptation Zero-Shot Token Co-Occurrence Graphs

May 8, 2026cs.CL

Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation

Recent literature on fine-tuning Large Language Models highlights a fundamental debate. While Full Fine-Tuning (FFT) provides greater representational plasticity, Low-Rank Adaptation (LoRA) can match or surpass FFT performance while constraining updates to a low-rank space and potentially benefiting from additional regularization. Through empirical evaluation across diverse tasks (SQL, Medical QA, and Counterfactual Knowledge) and varying language models (Gemma-3-1B, Qwen2.5-1.5B, and Qwen2.5-3B), we observe both trends and find that the better static architecture depends on the task and model. Spectral and truncation analyses further show that endpoint compressibility alone does not explain these task differences, suggesting task-score sensitivity and constrained optimization trajectories as possible explanations. To address this challenge, we propose a Mixture of LoRA and Full (MoLF) Fine-Tuning, a unified framework that enables continuous navigation between both training regimes. MoLF dynamically routes updates between FFT and LoRA at the optimizer level to ensure that exact gradient signals are available to both experts throughout training, while only selected experts update their weights. For memory-constrained environments, we also introduce MoLF-Efficient, which freezes base weights and only routes updates among a pair of LoRA experts of potentially varying rank. Our evaluations show that MoLF either improves on or stays within

1.5

percentage points of the better of FFT and LoRA across the nine tested settings, while MoLF-Efficient outperforms both AdaLoRA and AdaMix in eight of nine settings, with gains over the stronger baseline of up to

11.70

percentage points on Fact,

3.13

on Med, and

2.98

on SQL.

Haozhan Tang, Xiuqi Zhu, Xinyin Zhang +3

Carnegie Mellon University · Tsinghua University · Infinigence AI

Training-Side Stage-Aware Low-Rank Adaptation Large Language Model Adaptation Model Fine-Tuning

May 5, 2026cs.LG

Ortho-Hydra: Orthogonalized Experts for DiT LoRA

LoRA fine-tuning of diffusion transformers (DiT) on multi-style data suffers from \emph{style bleed}: a single low-rank residual cannot represent several distinct artist fingerprints, and the optimizer converges to their average. Mixture-of-experts LoRA in the HydraLoRA style replaces the up-projection with

E

heads under a router, but when every expert is zero-initialized the router receives identical gradient from each head and remains at the uniform prior. The experts then evolve permutation-symmetrically, and the network trains as a single rank-

r

LoRA at

E{\times}

the cost. We present \textbf{Ortho-Hydra}, a re-parameterisation that combines an OFT-style Cayley-orthogonal shared basis with per-expert \emph{disjoint output subspaces} carved from the top-

(Er)

left singular vectors of the pretrained weight. Disjointness makes the router's per-expert score non-degenerate at step~

0

, so specialization receives gradient signal before any expert has trained. We test the predicted deadlock on a DiT pipeline by comparing two HydraLoRA baselines, a zero-initialized shared-basis variant and the original

σ{=}0.1

Gaussian-jitter mitigation, against Ortho-Hydra under a matched optimiser, dataset, and step budget. Neither baseline leaves the uniform prior within the first

1\text{k}

steps; Ortho-Hydra begins de-uniformising within the first few hundred. End-task generation quality on multi-style data is out of scope; we report the construction, the cold-start mechanism, and the routing dynamics it changes. Code: https://github.com/sorryhyun/anima_lora.

Seunghyun Ji

Diffusion Transformers Multi-Lora Training-Side Stage-Aware Low-Rank Adaptation Orthogonality Experts

May 3, 2026cs.CV

Exploring Data-Free LoRA Transferability for Video Diffusion Models

Video diffusion models leveraging step distillation or causal distillation have achieved remarkable performance. However, adapting existing LoRAs to these variants remains a critical challenge due to weight space mismatches. We observe that direct application leads to style degradation and structural collapse, yet the underlying mechanisms remain poorly understood. To fill this gap, we delve into the weight space and identify that the incompatibility stems from spectral interference within shared functional clusters defined over singular subspaces. Specifically, our analysis reveals that while both paradigms respect spectral rigidity, they establish conflicting routing pathways that clash through constructive overload or destructive cancellation. To address this issue, we propose Cluster-Aware Spectral Arbitration (CASA), a data-free framework that dynamically arbitrates between safeguarding the target's manifold and restoring LoRA alignment based on spectral density. Extensive experiments demonstrate that CASA effectively mitigates artifacts and revives LoRA functionality. Our code is available at https://github.com/Noahwangyuchen/CASA

Yuchen Wang, Wenliang Zhong, Lichen Bai +6

Thrust of Data Science and Analytics, The Hong Kong University of Science and Technology (Guangzhou) · Thrust of Microelectronics, The Hong Kong University of Science and Technology (Guangzhou) · School of Intelligence Science and Technology, Nanjing University +1

Training-Side Stage-Aware Low-Rank Adaptation

May 2, 2026cs.AI

SCALE-LoRA: Auditing Post-Retrieval LoRA Composition with Residual Merging and View Reliability

Libraries of Low-Rank Adaptation (LoRA) adapters are becoming a practical by-product of parameter-efficient adaptation. Once such adapters accumulate, a natural question is no longer how to train one adapter for one task, but how to reuse an open pool of adapters for a new task given only a small support set. Prior work has shown that LoRA modules can be composed at the task level and dynamically selected at the instance level. However, open-pool LoRA reuse is not automatic: retrieving relevant adapters does not guarantee that their parameter updates are compatible, and composing adapters does not guarantee reliable outputs. We introduce the Sparse-Composition Agreement Layer (SCALE), a post-retrieval audit and composition framework for open-pool LoRA reuse. SCALE contains a deployable 1.0* merge path, Layer-Adaptive Sparse Residual Composition (LASRC), and a higher-cost reliability-analysis layer for multi-view disagreement. LASRC addresses merge interference by preserving a linear anchor while residualizing block-wise adapter update directions. The reliability layer treats disagreement among sparse composition views as an observable uncertainty signal and compares agreement, support-loss proxy selection, and oracle headroom under explicit path cost. In matched FLAN-T5-Large, BIG-Bench Hard (BBH), and 97-LoRA experiments, LASRC gives a directional single-view gain under fixed retrieval, while SCALE-support is reported as a query-label-free 3.0* reliability-analysis variant rather than as a calibrated or throughput-equivalent selector. Protocol-distinct BBH-8 validation shows the same qualitative trend on three decoder-only backbones. Detailed scores, paired audits, and path-cost records are reported in the experimental section.

Shuaipeng Zhou, Yu Zhang

School of Computer Science and Technology, Shenyang University of Chemical Technology · 2Liaoning Key Laboratory of Intelligent Technology for Chemical Process Industry

Multi-Lora Low-Rank Adaptation Framework Training-Side Stage-Aware Low-Rank Adaptation Model Auditing

Apr 29, 2026cs.CL

Decoupling Knowledge and Task Subspaces for Composable Parametric Retrieval Augmented Generation

Parametric Retrieval-Augmented Generation (PRAG) encodes external documents into lightweight parameter modules that can be retrieved and merged at inference time, offering a promising alternative to in-context retrieval augmentation. Despite its potential, many PRAG implementations train document adapters with task-supervised objectives, which may cause each adapter to encode both document-specific facts and reusable task-solving behavior. This entanglement may make adapter composition less reliable: when multiple adapters are merged at inference time, their overlapping task behaviors can accumulate together with document-specific updates, potentially making the merged adapter less stable and less focused on the intended document knowledge. To examine this issue, we explore Orthogonal Subspace Decomposition (OSD), an adapter-training setup that separates reusable task behavior from document-specific knowledge adapters. Concretely, we first train a Task LoRA to capture reusable task behavior, and then train document LoRAs to encode document-specific knowledge in a orthogonal subspace. This setup provides a controlled way to examine how orthogonalizing task and document LoRA updates affects adapter composition in multi-document PRAG. Experiments across multiple knowledge-intensive tasks and model scales suggest that this orthogonalization strategy can improve compositional robustness in parametric RAG, especially when multiple document adapters are merged.

Weihang Su, Hanwen Zhang, Qingyao Ai +1

Department of Computer Science and Technology, Tsinghua University

Agentic Retrieval-Augmented Generation Systems Multimodal Retrieval Augmented Generation Training-Side Stage-Aware Low-Rank Adaptation Search-Augmented Reasoning Subspace Orthogonality

Apr 24, 2026cs.CL

Where Should LoRA Go? Component-Type Placement in Hybrid Language Models

Hybrid language models that interleave attention with recurrent components are increasingly competitive with pure Transformers, yet standard LoRA practice applies adapters uniformly without considering the distinct functional roles of each component type. We systematically study component-type LoRA placement across two hybrid architectures -- Qwen3.5-0.8B (sequential, GatedDeltaNet + softmax attention) and Falcon-H1-0.5B (parallel, Mamba-2 SSM + attention) -- fine-tuned on three domains and evaluated on five benchmarks. We find that the attention pathway -- despite being the minority component -- consistently outperforms full-model adaptation with 5-10x fewer trainable parameters. Crucially, adapting the recurrent backbone is destructive in sequential hybrids (-14.8 pp on GSM8K) but constructive in parallel ones (+8.6 pp). We further document a transfer asymmetry: parallel hybrids exhibit positive cross-task transfer while sequential hybrids suffer catastrophic forgetting. These results establish that hybrid topology fundamentally determines adaptation response, and that component-aware LoRA placement is a necessary design dimension for hybrid architectures.

Hector Borobia, Elies Seguí-Mas, Guillermina Tormo-Carbó

VRAIN – Valencian Research Institute for Artificial Intelligence, Universitat Politècnica de València, Valencia, Spain · Department of Economics and Social Sciences, Universitat Politècnica de València, Valencia, Spain · Department of Business Organisation, Universitat Politècnica de València, Valencia, Spain

Large Language Model Adaptation Training-Side Stage-Aware Low-Rank Adaptation Recurrence

Apr 21, 2026cs.CL

SAMoRA: Semantic-Aware Mixture of LoRA Experts for Task-Adaptive Learning

The combination of Mixture-of-Experts (MoE) and Low-Rank Adaptation (LoRA) has shown significant potential for enhancing the multi-task learning capabilities of Large Language Models. However, existing methods face two primary challenges: (1)Imprecise Routing in the current MoE-LoRA method fails to explicitly match input semantics with expert capabilities, leading to weak expert specialization. (2)Uniform weight fusion strategies struggle to provide adaptive update strengths, overlooking the varying complexity of different tasks. To address these limitations, we propose SAMoRA (Semantic-Aware Mixture of LoRA Experts), a novel parameter-efficient fine-tuning framework tailored for task-adaptive learning. Specifically, A Semantic-Aware Router is proposed to explicitly align textual semantics with the most suitable experts for precise routing. A Task-Adaptive Scaling mechanism is designed to regulate expert contributions based on specific task requirements dynamically. In addition, a novel regularization objective is proposed to jointly promote expert specialization and effective scaling. Extensive experiments on multiple multi-task benchmarks demonstrate that SAMoRA significantly outperforms the state-of-the-art methods and holds excellent task generalization capabilities. Code is available at https://github.com/boyan-code/SAMoRA

Boyan Shi, Wei Chen, Shuyuan Zhao +4

School of Computer Science and Technology, Beijing Jiaotong University, China · 2Guangxi Key Lab of Trusted Software, Guilin University of Electronic Technology, China · 3Beijing Key Lab of Traffic Data Mining and Embodied Intelligence, China +2

Training-Side Stage-Aware Low-Rank Adaptation Multi--Task Learning

Jan 8, 2026cs.AI

DR-LoRA: Dynamic Rank LoRA for Fine-Tuning Mixture-of-Experts Models

Mixture-of-Experts (MoE) has become a prominent paradigm for scaling Large Language Models (LLMs). Parameter-efficient fine-tuning methods, such as LoRA, are widely adopted to adapt pretrained MoE LLMs to downstream tasks. However, existing approaches typically assign identical LoRA ranks to all expert modules, ignoring the heterogeneous specialization of pretrained experts. This uniform allocation leads to a resource mismatch: task-relevant experts are under-provisioned, while less relevant ones receive redundant parameters. To address this, we propose DR-LoRA, a Dynamic Rank LoRA framework for fine-tuning pretrained MoE models. Specifically, DR-LoRA initializes all expert LoRA modules with a small active rank and uses an expert saliency score, which combines routing frequency and gradient-based rank importance, to identify which experts would benefit most from additional capacity. It then periodically expands the active ranks of the task-critical expert LoRA, progressively constructing a heterogeneous rank distribution tailored to the target task. Experiments on three MoE models across six tasks show that DR-LoRA consistently outperforms LoRA and other strong baselines, demonstrating that task-adaptive heterogeneous rank allocation is an effective strategy to improve active capacity utilization in MoE fine-tuning.

Guanzhi Deng, Bo Li, Ronghao Chen +5

City University of Hong Kong · Tsinghua University · Peking University +2

Training-Side Stage-Aware Low-Rank Adaptation Hip-Lora Model Fine-Tuning

Sep 22, 2025cs.LG

On-the-Fly Adaptation to Quantization: Configuration-Aware LoRA for Efficient Fine-Tuning of Quantized LLMs

As increasingly large pre-trained models are released, deploying them on edge devices for privacy-preserving applications requires effective compression. Recent works combine quantization with the fine-tuning of high-precision LoRA adapters, which can substantially reduce model size while mitigating the accuracy loss from quantization. However, edge devices have inherently heterogeneous capabilities, while performing configuration-wise fine-tuning for every quantization setting is computationally prohibitive. In this paper, we propose CoA-LoRA, a method that dynamically adjusts the LoRA adapter to arbitrary quantization configurations (i.e., the per-layer bit-width choices of a pre-trained model) without requiring repeated fine-tuning. This is accomplished via a configuration-aware model that maps each configuration to its low-rank adjustments. The effectiveness of this model critically depends on the training configuration set, a collection of configurations chosen to cover different total bit-width budgets. However, constructing a high-quality configuration set is non-trivial. We therefore design a Pareto-based configuration search that iteratively optimizes the training configuration set, yielding more precise low-rank adjustments. Our experiments demonstrate that, unlike the state-of-the-art methods that require fine-tuning a separate LoRA adapter for each configuration, CoA-LoRA incurs no additional time cost while achieving comparable or even superior performance to those methods.

Rongguang Ye, Ming Tang, Edith C. H. Ngai

Southern University of Science and Technology · The University of Hong Kong

Large Language Model Quantization Training-Side Stage-Aware Low-Rank Adaptation Low-Rank Adaptation Adapters Model Fine-Tuning Pretraining Edge Devices

Aug 4, 2025cs.LG

PLoRA: Efficient Concurrent LoRA Training for Large Language Models

Low-Rank Adaptation (LoRA) has gained popularity as a fine-tuning approach for Large Language Models (LLMs) due to its low resource requirements and good performance. While numerous studies have investigated ways to improve LoRA serving efficiency by serving multiple LoRAs concurrently, existing methods assume that a wide range of LoRA adapters are available for serving. In our work, we conduct extensive empirical studies to show that current LoRA training paradigms do not efficiently utilize hardware resources and incur high overhead to obtain a performant LoRA adapter. Leveraging these insights, we propose PLoRA, which automatically orchestrates concurrent LoRA fine-tuning jobs under given hardware and model constraints and develops performant kernels to improve training efficiency. Across a range of LLMs and LoRA configurations, PLoRA improves training throughput by up to 12.8x and reduces the overall fine-tuning makespan by up to 7.52x compared to existing approaches.

Minghao Yan, Zhuang Wang, Zhen Jia +2

University of Wisconsin-Madison · Amazon Web Services

Training-Side Stage-Aware Low-Rank Adaptation Low-Rank Adaptation Framework Low-Rank Adaptation Adapters

Jun 12, 2025cs.LG

TreeLoRA: Efficient Continual Learning via Layer-Wise LoRAs Guided by a Hierarchical Gradient-Similarity Tree

Many real-world applications collect data in a streaming environment, where learning tasks are encountered sequentially. This necessitates continual learning (CL) to update models online, enabling adaptation to new tasks while preserving past knowledge to prevent catastrophic forgetting. Nowadays, with the flourish of large pre-trained models (LPMs), efficiency has become increasingly critical for CL, due to their substantial computational demands and growing parameter sizes. In this paper, we introduce TreeLoRA (K-D Tree of Low-Rank Adapters), a novel approach that constructs layer-wise adapters by leveraging hierarchical gradient similarity to enable efficient CL, particularly for LPMs. To reduce the computational burden of task similarity estimation, we employ bandit techniques to develop an algorithm based on lower confidence bounds to efficiently explore the task structure. Furthermore, we use sparse gradient updates to facilitate parameter optimization, making the approach better suited for LPMs. Theoretical analysis is provided to justify the rationale behind our approach, and experiments on both vision transformers (ViTs) and large language models (LLMs) demonstrate the effectiveness and efficiency of our approach across various domains, including vision and natural language processing tasks.

Yu-Yang Qian, Yuan-Ze Xu, Zhen-Yu Zhang +2

National Key Laboratory for Novel Software Technology, Nanjing University, China · School of Artificial Intelligence, Nanjing University, China · RIKEN AIP, Tokyo, Japan

Replay-Based Continual Learning Training-Side Stage-Aware Low-Rank Adaptation Continual Learning Vision Transformer Similarity and Metrics

Aug 8, 2024cs.CL

BA-LoRA: Bias-Alleviating Low-Rank Adaptation to Mitigate Catastrophic Inheritance in Large Language Models

Parameter-efficient fine-tuning (PEFT) has become a de facto standard for adapting large language models (LLMs). However, we identify a critical vulnerability within popular low-rank adaptation methods such as LoRA: they can exacerbate "Catastrophic Inheritance" - the unchecked propagation of biases, noise, and data imbalances from pre-training. This phenomenon can degrade model robustness and fairness, undermining the benefits of efficient adaptation. To address this, we introduce Bias-Alleviating Low-Rank Adaptation (BA-LoRA). Our approach is founded on a principled decomposition of Catastrophic Inheritance into three core challenges: Knowledge Drift, Representation Collapse, and Overfitting to Noise. BA-LoRA systematically mitigates these issues by incorporating a trio of targeted regularizers: consistency, diversity, and an SVD-based term, designed to preserve core knowledge, promote representational richness, and encourage robust, low-rank output representations, respectively. We conduct comprehensive evaluations on a suite of Natural Language Generation (NLG) and Natural Language Understanding (NLU) tasks using diverse, prominent open-source language models (e.g., LLaMA-2-7B and DeBERTa-v3-base). Our results show that BA-LoRA not only outperforms state-of-the-art LoRA variants in terms of performance and stability, but also demonstrates superior robustness and bias mitigation on targeted evaluations. These results provide evidence that BA-LoRA can counteract the adverse effects of Catastrophic Inheritance.

Yupeng Chang, Yi Chang, Yuan Wu

School of Artificial Intelligence, Jilin University · Key Laboratory of Symbolic Computation and Knowledge Engineering, Jilin University · International Center of Future Science, Jilin University

Training-Side Stage-Aware Low-Rank Adaptation Large Language Model Adaptation Large Language Model Bias

Date pendingcs.CV

SPECTRA: Band-Routed Embedding and Stage-Wise LoRA for Cross-Sensor Fine-Tuning of Geospatial Foundation Models

Geospatial foundation models (GeoFMs), pretrained on large-scale geospatial data such as Earth observation (EO), climate, and weather data, have shown promising performance when fine-tuned on diverse downstream tasks. However, there are two challenges of adapting EO-pretrained GeoFMs to practical downstream datasets. The first challenge is how to handle spectral mismatch: pretrained patch embeddings expect a fixed set of input bands, whereas downstream sensors may provide different channels. The second challenge is how to reduce fine-tuning cost and make it efficient. While existing work has made efforts on these challenges individually, jointly improving fine-tuning performance under spectral mismatch while reducing adaptation cost remains underexplored. We propose SPECTRA, a parameter-efficient fine-tuning framework that addresses both spectral mismatch and adaptation cost. To handle spectral mismatch, SPECTRA introduces Band-Routed Embedding (BRE), which maps all available downstream bands into the band space expected by the pretrained GeoFM. By using BRE, all available bands in the downstream dataset are utilized to improve the selected-band input without changing the pretrained patch embedding interface. To reduce adaptation cost, SPECTRA further introduces a Stage-wise Transferability-aware LoRA (ST-LoRA) fine-tuning. ST-LoRA estimates stage-wise transferability before fine-tuning and assigns stage-specific LoRA ranks, concentrating trainable parameters on the stages with high transferability for the target task. Across three EO-pretrained GeoFMs and four downstream segmentation datasets, experiments show that BRE improves performance by utilizing all spectral bands, while ST-LoRA reduces trainable parameters compared with full fine-tuning and standard LoRA. Code is available at https://github.com/big-data-lab-umbc/SPECTRA.

Xingyan Li, Jordan A. Caraballo-Vega, Jie Gong +2

University of Maryland, Baltimore County Baltimore, Maryland, USA · NASA Goddard Space Flight Center Greenbelt, Maryland, USA

Geospatial Foundation Models Parameter-Efficient Fine-Tuning Methods Training-Side Stage-Aware Low-Rank Adaptation Spectral Rapid Adaptation