---
title: "Pretraining Data Scale Reverses the Muon-AdamW Fine-Tuning Gap"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2611.03633v1"
permalink: "https://feed.astralor.com/items/itm_7f7f3ba1729dbdbcfbd99eb6"
published: "2026-11-02T18:56:43.000Z"
score: 56
channel: "ai"
tags: ["LLM","Pretraining Data Scale","Reverses","Muon-AdamW Fine-Tuning Gap","Muon","AdamW","However"]
---

# Pretraining Data Scale Reverses the Muon-AdamW Fine-Tuning Gap

## At a glance

Muon has recently emerged as a promising optimizer for pretraining large language models, offering a speedup over AdamW in minimizing pretraining loss. However, it remains unclear whether Muon's advantage persists after…

## Key points

- Muon has recently emerged as a promising optimizer for pretraining large language models, offering a speedup over AdamW in minimizing pretraining loss.
- However, it remains unclear whether Muon's advantage persists after fine-tuning.
- In this paper, we study the downstream behavior of Muon and AdamW and uncover a surprising trend: beyond a sufficiently large pretraining data budget, AdamW outperforms Muon after fine-tuning.

## Full text

Muon has recently emerged as a promising optimizer for pretraining large language models, offering a speedup over AdamW in minimizing pretraining loss. However, it remains unclear whether Muon's advantage persists after fine-tuning. In this paper, we study the downstream behavior of Muon and AdamW and uncover a surprising trend: beyond a sufficiently large pretraining data budget, AdamW outperforms Muon after fine-tuning. Moreover, as model size increases, AdamW overtakes Muon at progressively smaller pretraining token-parameter ratios. To explain this reversal, we view fine-tuning as a perturbation of the outputs and introduce the margin-to-spread ratio, which measures how far the correct logit lies above the wrong ones relative to how widely the wrong logits are spread. In language models, Muon has the larger margin-to-spread ratio at small data budgets and AdamW at large ones, mirroring the fine-tuning trend. In a linear associative memory, we prove that after one update the margin-to-spread ratio of gradient descent gains on that of Muon as the data budget grows, and simulations show that this trend persists after extended training.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_7f7f3ba1729dbdbcfbd99eb6