---
title: "Model Merging Does Not Require Uniform LoRA Ranks"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2611.06486v1"
permalink: "https://feed.astralor.com/items/itm_84981ea7a4dac0a58205ce84"
published: "2026-11-04T13:26:44.000Z"
score: 59
channel: "ai"
tags: ["RAG","Model Merging Does","Not Require Uniform","LoRA Ranks Model","Standard","LoRA","Model"]
---

# Model Merging Does Not Require Uniform LoRA Ranks

## 速览

Model merging combines separately fine-tuned experts into a single model without retraining. Standard practice trains every expert at the same LoRA rank, which couples the rank choice across experts even though each dom…

## 关键点

- Model merging combines separately fine-tuned experts into a single model without retraining.
- Standard practice trains every expert at the same LoRA rank, which couples the rank choice across experts even though each domain has its own best rank.
- We show this coupling is a property of the merge method, not of merging.

## 全文

Model merging combines separately fine-tuned experts into a single model without retraining. Standard practice trains every expert at the same LoRA rank, which couples the rank choice across experts even though each domain has its own best rank. We show this coupling is a property of the merge method, not of merging. We train 36 experts on three domains (mathematics, instruction following, multilingual) at four ranks and three model scales, then merge with six methods at uniform ranks and at each domain's own best rank. Giving experts different ranks costs averaging and utility-based selection nothing, and gains them up to 1.8 pp of multi-domain accuracy, while the conflict-resolution methods are inconsistent and can lose up to 4 pp. The split has a simple cause. Merging keeps only a fixed share of the pooled spectral components, and methods that choose them by size discard many that would have reduced loss. They also tend to discard the same expert: the instruction adapter has the smallest singular values, so it is dropped almost entirely and IFEval falls by 16 to 25 pp. Choosing components by their effect on loss instead keeps all three experts, and leaves accuracy nearly flat across an 8$\times$ rank range (0.7 pp, against 6.7 pp for TIES at 4B scale). Rank can therefore be chosen per expert whenever the merge operator does not select by magnitude.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_84981ea7a4dac0a58205ce84