大模型
arxiv.org ·
用于稳定 LLM 预训练的输出嵌入中心化
该论文针对 LLM 预训练末期常见的 output logit divergence 不稳定现象，从输出嵌入几何角度分析，认为各向异性嵌入是其根源，并提出输出嵌入中心化（OEC）作为缓解策略，包含确定性的 μ-centering 和正则化的 μ-loss 两种实现。实验显示两种变体在训练稳定性上均优于 z-loss，与 logit soft-capping 相当，且在有无权重共享时均成立；次要结论是 μ-loss 对正则超参调优的敏感度明显低于 z-loss。
AI推荐理由 arXiv 论文（v3，2026-09-10 修订），作者提出输出嵌入中心化（OEC）方法，从输出嵌入几何角度分析 LLM 预训练末期的 output logit divergence，指出各向异性嵌入是根源，并给出 μ-centering 与 μ-loss 两种实现；实验称两者在训练稳定性上优于 z-loss、与 logit soft-capping 相当，且 μ-loss 对正则超参更不敏感。属于有明确方法、对比基线和结论的预训练稳定性研究，对训练工程有参考价值；但摘要未给出具体模型规模、数据集与量化指标，效果仍需正文验证。
78
正文
Computer Science > Machine Learning
arXiv:2601.02031 (cs)
[Submitted on 5 Jan 2026 (v1), last revised 10 Sep 2026 (this version, v3)]
Title:Output Embedding Centering for Stable LLM Pretraining
Authors:Felix Stollenwerk, Anna Lokrantz, Niclas Hertzberg
View a PDF of the paper titled Output Embedding Centering for Stable LLM Pretraining, by Felix Stollenwerk and 2 other authors
View PDF
Abstract:Pretraining of large language models is not only expensive but also prone to certain training instabilities. A specific instability that often occurs at the end of training is output logit divergence. The most widely used mitigation strategies, z-loss and logit soft-capping, merely address the symptoms rather than the underlying cause of the problem. In this paper, we analyze the instability from the perspective of the output embeddings' geometry and identify anisotropic embeddings as its source. Based on this, we propose output embedding centering (OEC) as a new mitigation strategy, and demonstrate that it suppresses output logit divergence. OEC can be implemented in two different ways: as a deterministic operation called $\mu$-centering, or a regularization method called $\mu$-loss. Our experiments show that both variants outperform z-loss in terms of training stability, while being on par with logit soft-capping. This holds true both in the presence and the absence of weight tying. As a secondary result, we find that $\mu$-loss is significantly less sensitive to regularization hyperparameter tuning than z-loss.
Comments:
Subjects:
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
Cite as:
arXiv:2601.02031 [cs.LG]
(or
arXiv:2601.02031v3 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2601.02031
Focus to learn more
arXiv-issued DOI via DataCite
Submission history
From: Felix Stollenwerk [view email]
[v1]
Mon, 5 Jan 2026 11:44:05 UTC (2,036 KB)
[v2]
Thu, 2 Apr 2026 10:34:01 UTC (2,822 KB)
[v3]
Thu, 10 Sep 2026 09:26:21 UTC (4,153 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled Output Embedding Centering for Stable LLM Pretraining, by Felix Stollenwerk and 2 other authors
View PDF
TeX Source
view license
Current browse context:
cs.LG
new
|
recent
| 2026-01
Change to browse by:
cs
cs.AI
cs.CL
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
export BibTeX citation
Loading...
BibTeX formatted citation
×
loading...
Data provided by:
Bookmark
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
IArxiv recommender toggle
IArxiv Recommender
(What is IArxiv?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? |
Disable MathJax (What is MathJax?)