大模型
arxiv.org ·
Diffract：大语言模型领域自适应的谱视角
该论文研究持续预训练（CPT）用于将通用大模型适配到数学、指令、代码和自然文本等专业领域。通过对权重矩阵做奇异值分解，作者发现 CPT 基本不改变奇异值谱，适配主要由奇异向量变化驱动；注意力头投影矩阵呈现强领域相关异质性，据此提出头重要性准则，可移除最多 60% 的头更新而无明显质量损失，将低重要性头回退至预训练状态可使基准准确率较全量训练基线提升最多 4%。作者还发现领域连通性——CPT 检查点间线性插值可实现平滑的领域质量过渡，并开源了面向十亿参数级模型的可扩展谱分析工具包 Diffract。
AI推荐理由 arXiv 论文（v2，2026-09-11 修订），来源可信、摘要信息密度高。核心新信息：用权重矩阵 SVD 发现持续预训练（CPT）基本不改变奇异值谱，领域适配主要由奇异向量变化驱动；注意力头投影矩阵存在强领域相关异质性，据此定义头重要性准则，可移除最多 60% 的头更新而无明显质量损失，将低重要性头回退到预训练状态可使基准准确率较全量训练基线提升最多 4%；并观察到领域连通性（CPT 检查点线性插值可平滑过渡）并开源 Diffract 工具包。局限：摘要未给出具体基准、模型规模与实验细节，4% 提升为作者自述，需查看全文验证。
78
正文
Computer Science > Machine Learning
arXiv:2608.10850 (cs)
[Submitted on 11 Aug 2026 (v1), last revised 11 Sep 2026 (this version, v2)]
Title:Diffract: Spectral View of LLM Domain Adaptation
Authors:Nikita Borodin, Maria Krylova, Artem Zabolotnyi, Dmitry Aspisov, Egor Shikov, Nikita Tyuplyaev, Oleg Travkin, Roman Alferov, Dmitry Vinichenko
View a PDF of the paper titled Diffract: Spectral View of LLM Domain Adaptation, by Nikita Borodin and Maria Krylova and Artem Zabolotnyi and Dmitry Aspisov and Egor Shikov and Nikita Tyuplyaev and Oleg Travkin and Roman Alferov and Dmitry Vinichenko
View PDF
HTML (experimental)
Abstract:We study continual pre-training (CPT) as a mechanism for adapting general-purpose large language models to specialized domains: mathematics, instruction, code, and natural text. Using singular value decomposition of weight matrices, we find that CPT leaves singular value spectra largely invariant, with adaptation driven mainly by changes in singular vectors. An analysis of attention-head projection matrices reveals strong, domain-dependent head heterogeneity, which we exploit to define a head importance criterion: up to 60% of head updates can be removed without measurable quality loss. Selectively rewinding low-importance heads to their pre-trained state improves benchmark accuracy by up to 4% versus the fully trained baseline. Finally, we identify domain connectivity - linear interpolation between CPT checkpoints yields smooth domain-quality interpolation without notable degradation on either domain - and release Diffract, an open-source toolkit for scalable spectral analysis of billion-parameter models.
Comments:
Subjects:
Machine Learning (cs.LG)
Cite as:
arXiv:2608.10850 [cs.LG]
(or
arXiv:2608.10850v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.10850
Focus to learn more
arXiv-issued DOI via DataCite
Submission history
From: Nikita Borodin S. [view email]
[v1]
Tue, 11 Aug 2026 12:23:28 UTC (5,295 KB)
[v2]
Fri, 11 Sep 2026 17:24:34 UTC (5,295 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled Diffract: Spectral View of LLM Domain Adaptation, by Nikita Borodin and Maria Krylova and Artem Zabolotnyi and Dmitry Aspisov and Egor Shikov and Nikita Tyuplyaev and Oleg Travkin and Roman Alferov and Dmitry Vinichenko
View PDF
HTML (experimental)
TeX Source
view license
Current browse context:
cs.LG
new
|
recent
| 2026-08
Change to browse by:
cs
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?)