CtM merges T LoRAs into one rank-r LoRA by computing shared r-dimensional subspaces from the LoRA weights, projecting adapters into r x r coordinates, and merging in that reduced space, outperforming merge-then-compress baselines in experiments.
Decouple and Orthogonalize: A Data-Free Framework for LoRA Merging
6 Pith papers cite this work. Polarity classification is still indexing.
6
Pith papers citing it
abstract
With more open-source models available for diverse tasks, model merging has gained attention by combining models into one, reducing training, storage, and inference costs. Current research mainly focuses on model merging for full fine-tuning, overlooking the popular LoRA. However, our empirical analysis reveals that: a) existing merging methods designed for full fine-tuning perform poorly on LoRA; b) LoRA modules show much larger parameter magnitude variance than full fine-tuned weights; c) greater parameter magnitude variance correlates with worse merging performance. Considering that large magnitude variances cause deviations in the distribution of the merged parameters, resulting in information loss and performance degradation, we propose a Decoupled and Orthogonal merging approach(DO-Merging). By separating parameters into magnitude and direction components and merging them independently, we reduce the impact of magnitude differences on the directional alignment of the merged models, thereby preserving task information. Furthermore, we introduce a data-free, layer-wise gradient descent method with orthogonal constraints to mitigate interference during the merging of direction components. We provide theoretical guarantees for both the decoupling and orthogonal components. And we validate through extensive experiments across vision, language, and multi-modal domains that our proposed DO-Merging can achieve significantly higher performance than existing merging methods at a minimal cost. Notably, each component can be flexibly integrated with existing methods, offering near free-lunch improvements across tasks.
years
2026 6representative citing papers
Online SFT data selection acts as an implicit preference model, shifting refusal rates, verbosity, and sycophancy in directions predictable from the selected data's attribute mixture.
Prompt-aware weighting strategies W-Switch and W-Composite improve multi-concept LoRA composition in diffusion models without training.
M2A uses null-space model merging to combine mathematical and agentic reasoning in LLMs, raising SWE-Bench Verified performance from 44.0% to 51.2% on Qwen3-8B without retraining.
FREE-Switch dynamically switches LoRA adapters using frequency importance per diffusion step and adds semantic alignment to reduce content drift when merging specialized image generators.
SOLAR introduces a self-optimizing agent using meta-learning on model weights and RL-driven strategy discovery for lifelong adaptation in LLMs, claiming superior performance on reasoning tasks across domains.
citing papers explorer
-
Compress then Merge: From Multiple LoRAs into One Low-Rank Adapter
CtM merges T LoRAs into one rank-r LoRA by computing shared r-dimensional subspaces from the LoRA weights, projecting adapters into r x r coordinates, and merging in that reduced space, outperforming merge-then-compress baselines in experiments.
-
Online Data Selection Is Implicit Alignment
Online SFT data selection acts as an implicit preference model, shifting refusal rates, verbosity, and sycophancy in directions predictable from the selected data's attribute mixture.
-
Training-Free Multi-Concept LoRA Composition with Prompt-Aware Weighting
Prompt-aware weighting strategies W-Switch and W-Composite improve multi-concept LoRA composition in diffusion models without training.
-
M2A: Synergizing Mathematical and Agentic Reasoning in Large Language Models
M2A uses null-space model merging to combine mathematical and agentic reasoning in LLMs, raising SWE-Bench Verified performance from 44.0% to 51.2% on Qwen3-8B without retraining.
-
FREE-Switch: Frequency-based Dynamic LoRA Switch for Style Transfer
FREE-Switch dynamically switches LoRA adapters using frequency importance per diffusion step and adds semantic alignment to reduce content drift when merging specialized image generators.
-
SOLAR: A Self-Optimizing Open-Ended Autonomous Agent for Lifelong Learning and Continual Adaptation
SOLAR introduces a self-optimizing agent using meta-learning on model weights and RL-driven strategy discovery for lifelong adaptation in LLMs, claiming superior performance on reasoning tasks across domains.