arXiv:2509. 02555v2 Announce Type: replace-cross Abstract: Model merging techniques aim to integrate the abilities of multiple models into a single model.
arXiv:2608. 14264v1 Announce Type: cross Abstract: Model merging combines trained models directly in weight space, offering a compute-efficient alternative to additional fine-tuning.
arXiv:2606. 28373v1 Announce Type: cross Abstract: Model merging integrates the capabilities of multiple expert models to create strong models for multiple tasks without additional training, thereby reducing computational resource requirements.
The paper introduces a data‑free method for model merging that estimates per‑layer covariance matrices directly from difference matrices, eliminating the need for auxiliary data. This approach reduces computational costs while maintaining a principled interference‑minimization framework. Experiments on vision and language benchmarks with models from 86 M to 7 B parameters show that the method outperforms existing data‑free merging techniques.
arXiv:2608. 12842v1 Announce Type: new Abstract: Model merging has recently attracted significant attention as a promising paradigm for constructing unified multi-task models without requiring additional retraining.
arXiv:2610.07819v1 Announce Type: new Abstract: Model merging offers a promising solution for combining multiple fine-tuned checkpoints into a single model through parameter arithmetic. However, find...
arXiv:2510. 17426v3 Announce Type: replace-cross Abstract: The "alignment tax" of post-training is typically framed as a drop in task accuracy.
Mixture-Trained Merging (MTM) is a method for creating unified language models that combine multiple objectives—such as mathematics, code, instruction following, and controllable thinking—into a single parameter set. Instead of sequentially post‑training on each objective, MTM trains each branch on a mixture of objectives, ensuring that the branches remain compatible in weight space and can be merged without degrading performance. The approach iteratively refines merge coefficients using low‑cost evaluations and multi‑objective Bayesian optimization, outperforming naive merging and preserving distinct behaviors across domains.
arXiv:2606. 03073v1 Announce Type: cross Abstract: Reinforcement learning (RL) for large language models (LLMs) is highly sensitive to hyperparameter configurations, making hyperparameter optimization (HPO) essential yet computationally expensive.
arXiv:2607. 11997v1 Announce Type: cross Abstract: Multi-task model merging combines separately trained expert models into a single model that handles all tasks without co-training.
The paper introduces Power‑Law Entropy Search (PLES), a computational‑cost‑aware acquisition function that uses multi‑fidelity Bayesian optimization to efficiently estimate optimal hyperparameter scaling laws for large language model training. PLES focuses on reducing the overall uncertainty of scaling law estimates rather than optimizing a single objective, selecting configurations that maximize uncertainty reduction per unit computational cost. Experiments on synthetic benchmarks, surrogate models, and real LLM pre‑training runs show that PLES converges to accurate scaling laws using less than one‑tenth of the computational budget required by conventional grid search and other baselines.
arXiv:2211. 14411v5 Announce Type: replace-cross Abstract: Hyperparameter optimization (HPO) is crucial for strong performance of deep learning algorithms and real-world applications often impose some constraints, such as on memory usage or latency, on top of the performance requirement.
CoMerge is a conflict‑driven preference optimization framework for merging multiple expert language models into a single multi‑task model without full retraining. It treats model merging as a preference optimization problem, using self‑supervised, conflict‑driven hard negative samples derived from naive merging defects to refine lightweight, tensor‑wise merging coefficients. Experiments show CoMerge achieves near‑perfect performance on MergeBench and improves instruction‑following and safety on Llama‑3.1‑8B‑Instruct while optimizing only 1,445 scalar coefficients.