Harnessing Optimization Dynamics for Curvature-Informed Model Merging

• URL: http://arxiv.org/abs/2509.11167v1
• Date: Sun, 14 Sep 2025 08:59:53 GMT
• Title: Harnessing Optimization Dynamics for Curvature-Informed Model Merging
• Authors: Pouria Mahdavinia, Hamed Mahdavi, Niloofar Mireshghallah, Mehrdad Mahdavi,
• Abstract summary: In supervised fine-tuning, multiple capability-based SFT checkpoints must be consolidated into a single model.<n>We introduce Optimization Trajectory Aware (OTA) Merging and Fast Fisher Grafting (FFG)<n>OTA+FFG improves merged-model quality over strong weight-space baselines, reduces negative transfer, and remains robust across sparsity levels.
• Score: 17.42364575754576
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Model merging is an effective post-training strategy for composing capabilities in large language models without joint retraining. We study this in the supervised fine-tuning (SFT) stage, where multiple capability-based SFT checkpoints -- spanning math, code, precise instruction following, general instruction following, and knowledge recall -- must be consolidated into a single model. We introduce Optimization Trajectory Aware (OTA) Merging, a curvature-aware aggregation that leverages optimizer second-moment statistics as a diagonal curvature proxy to reweight parameter edits and mitigate interference. Complementing OTA, we propose Fast Fisher Grafting (FFG), a curvature-driven task-localization step that sparsifies conflicting or low-importance edits. FFG induces extremely low-rank masks concentrated in early attention query/key projections and token embeddings, exploiting shared curvature across capabilities. We further develop a memory-light compression of the second moments that preserves OTA's effect. Across diverse capability-based SFT checkpoints, OTA+FFG improves merged-model quality over strong weight-space baselines, reduces negative transfer, and remains robust across sparsity levels. Analyses reveal substantial curvature overlap between checkpoints, offering a novel lens on why simple linear merging can be effective in practice. Ablations confirm that FFG is critical for reducing task interference and that the compressed second moments retain the gains of the full formulation. To facilitate reproducibility, we open-source all code, training and evaluation scripts, visualization artifacts, and capability-specific SFT checkpoints at https://github.com/pmahdavi/ota-merge.

Related papers

• Merging Beyond: Streaming LLM Updates via Activation-Guided Rotations [55.047454145941366]
  Streaming Merging is an innovative model updating paradigm that conceptualizes merging as an iterative optimization process.<n> ARM is a strategy designed to approximate gradient descent dynamics.<n> ARM requires only early SFT checkpoints and, through iterative merging, surpasses the fully converged SFT model.
  arXiv  Detail & Related papers  (2026-02-03T08:15:57Z)
• PT$^2$-LLM: Post-Training Ternarization for Large Language Models [52.4629647715623]
  Large Language Models (LLMs) have shown impressive capabilities across diverse tasks, but their large memory and compute demands hinder deployment.<n>We propose PT$2$-LLM, a post-training ternarization framework tailored for LLMs.<n>At its core is an Asymmetric Ternary Quantizer equipped with a two-stage refinement pipeline.
  arXiv  Detail & Related papers  (2025-09-27T03:01:48Z)
• DEFT: Decompositional Efficient Fine-Tuning for Text-to-Image Models [103.18486625853099]
  DEFT, Decompositional Efficient Fine-Tuning, adapts a pre-trained weight matrix by decomposing its update into two components.<n>We conduct experiments on the Dreambooth and Dreambench Plus datasets for personalization, the InsDet dataset for object and scene adaptation, and the VisualCloze dataset for a universal image generation framework.
  arXiv  Detail & Related papers  (2025-09-26T18:01:15Z)
• Optimal Sparsity of Mixture-of-Experts Language Models for Reasoning Tasks [17.067788440109137]
  Mixture-of-Experts (MoE) models are now standard in state-of-the-art systems.<n>We investigate how MoE sparsity influences two distinct capability regimes: memorization and reasoning.
  arXiv  Detail & Related papers  (2025-08-26T04:31:28Z)
• Communication-Efficient Wireless Federated Fine-Tuning for Large-Scale AI Models [13.742950928229078]
  Low-Rank Adaptation (LoRA) addresses these issues by training compact, low-rank matrices instead of fully fine-tuning large models.<n>This paper introduces a wireless federated LoRA fine-tuning framework that optimize both learning performance and communication efficiency.
  arXiv  Detail & Related papers  (2025-05-01T06:15:38Z)
• Decentralized Nonconvex Composite Federated Learning with Gradient Tracking and Momentum [78.27945336558987]
  Decentralized server (DFL) eliminates reliance on client-client architecture.<n>Non-smooth regularization is often incorporated into machine learning tasks.<n>We propose a novel novel DNCFL algorithm to solve these problems.
  arXiv  Detail & Related papers  (2025-04-17T08:32:25Z)
• SuperFlow++: Enhanced Spatiotemporal Consistency for Cross-Modal Data Pretraining [62.433137130087445]
  SuperFlow++ is a novel framework that integrates pretraining and downstream tasks using consecutive camera pairs.<n>We show that SuperFlow++ outperforms state-of-the-art methods across diverse tasks and driving conditions.<n>With strong generalizability and computational efficiency, SuperFlow++ establishes a new benchmark for data-efficient LiDAR-based perception in autonomous driving.
  arXiv  Detail & Related papers  (2025-03-25T17:59:57Z)
• Unlock the Correlation between Supervised Fine-Tuning and Reinforcement Learning in Training Code Large Language Models [12.656574142412484]
  We make an attempt to understand the correlation between supervised fine-tuning and reinforcement learning.<n>We find that both atomic and synthetic functions are indispensable for SFT's generalization.
  arXiv  Detail & Related papers  (2024-06-14T03:39:01Z)
• Communication Efficient and Provable Federated Unlearning [43.178460522012934]
  We study federated unlearning, a novel problem to eliminate the impact of specific clients or data points on the global model learned via federated learning (FL) This problem is driven by the right to be forgotten and the privacy challenges in FL. We introduce a new framework for exact federated unlearning that meets two essential criteria: textitcommunication efficiency and textitexact unlearning provability.
  arXiv  Detail & Related papers  (2024-01-19T20:35:02Z)
• Over-the-Air Federated Learning and Optimization [52.5188988624998]
  We focus on Federated learning (FL) via edge-the-air computation (AirComp) We describe the convergence of AirComp-based FedAvg (AirFedAvg) algorithms under both convex and non- convex settings. For different types of local updates that can be transmitted by edge devices (i.e., model, gradient, model difference), we reveal that transmitting in AirFedAvg may cause an aggregation error. In addition, we consider more practical signal processing schemes to improve the communication efficiency and extend the convergence analysis to different forms of model aggregation error caused by these signal processing schemes.
  arXiv  Detail & Related papers  (2023-10-16T05:49:28Z)
• DSEE: Dually Sparsity-embedded Efficient Tuning of Pre-trained Language Models [152.29364079385635]
  As pre-trained models grow bigger, the fine-tuning process can be time-consuming and computationally expensive. We propose a framework for resource- and parameter-efficient fine-tuning by leveraging the sparsity prior in both weight updates and the final model weights. Our proposed framework, dubbed Dually Sparsity-Embedded Efficient Tuning (DSEE), aims to achieve two key objectives: (i) parameter efficient fine-tuning and (ii) resource-efficient inference.
  arXiv  Detail & Related papers  (2021-10-30T03:29:47Z)
• Over-the-Air Federated Learning from Heterogeneous Data [107.05618009955094]
  Federated learning (FL) is a framework for distributed learning of centralized models. We develop a Convergent OTA FL (COTAF) algorithm which enhances the common local gradient descent (SGD) FL algorithm. We numerically show that the precoding induced by COTAF notably improves the convergence rate and the accuracy of models trained via OTA FL.
  arXiv  Detail & Related papers  (2020-09-27T08:28:25Z)

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.