LGAIJul 2

Model Merging as Probabilistic Inference in Fine-Tuning Parameter Space

arXiv:2607.0168911.1Has Code
Predicted impact top 22% in LG · last 90 daysOriginality Highly original
AI Analysis

This work addresses the problem of effectively combining multiple single-task models into a multi-task model without retraining, which is important for practical deployment of large models.

The paper proposes a probabilistic inference framework for model merging, treating each single-task solution as an energy-based expert model. By replacing Gaussian assumptions with heavy-tailed Cauchy experts, they achieve significant improvements over state-of-the-art baselines across multiple tasks and architectures.

Model merging aims to combine existing single-task solutions into a multi-task solution without additional data-driven fine-tuning.~Most existing approaches achieve this using geometric properties of local solution spaces. However, such geometric views provide limited guidance for scoring how statistically useful each task-specific update direction is across tasks during merging. To address this, we formulate model merging from a new perspective of probabilistic inference under a product-of-experts (PoE) scenario where each single-task solution defines an energy-based expert model (EBM) over the merged parameters. We show that several existing model merging methods arise as special cases of our framework under energy designs that impose implicit Gaussian assumptions on directional residuals between merged and task-specific models. Empirically, we find that these residuals are often heavy-tailed which exposes a mismatch with the imposed light-tailed Gaussian structures. We address this with a heavy-tailed PoE design based on Cauchy experts, which better captures the observed residual behavior while admitting a provably convergent inference procedure. Experiments across multiple tasks and architectures show significant improvements over state-of-the-arts baselines. Our code is available at https://github.com/MinhLong210/PoE-EBM-Merging.git.

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes