AI Navigate

UltraEdit: Training-, Subject-, and Memory-Free Lifelong Editing in Language Models

arXiv cs.CL / 3/11/2026

Tools & Practical UsageModels & Research

Key Points

  • UltraEdit is a novel lifelong model editing approach for large language models that is training-, subject-, and memory-free, enabling ultra-scalable and practical lifelong learning.
  • It uniquely computes parameter shifts in one step using only a hidden state and its gradient, ensuring efficiency and simplicity.
  • UltraEdit implements a lifelong normalization strategy to handle distributional shifts and maintain consistency over time for scalable editing.
  • It achieves editing speeds over 7 times faster and uses 4 times less VRAM than previous state-of-the-art methods, allowing editing of 7B parameter models on consumer-grade GPUs.
  • The authors introduce UltraEditBench, the largest dataset with over 2 million editing pairs, demonstrating that UltraEdit can support up to 2 million edits while maintaining high accuracy across diverse scenarios and models.

Computer Science > Computation and Language

arXiv:2505.14679 (cs)
[Submitted on 20 May 2025 (v1), last revised 10 Mar 2026 (this version, v3)]

Title:UltraEdit: Training-, Subject-, and Memory-Free Lifelong Editing in Language Models

View a PDF of the paper titled UltraEdit: Training-, Subject-, and Memory-Free Lifelong Editing in Language Models, by Xiaojie Gu and 3 other authors
View PDF
Abstract:Lifelong learning enables large language models (LLMs) to adapt to evolving information by continually updating their internal knowledge. An ideal system should support efficient, wide-ranging updates while preserving existing capabilities and ensuring reliable deployment. Model editing stands out as a promising solution for this goal, offering a focused and efficient way to revise a model's internal knowledge. Although recent paradigms have made notable progress, they often struggle to meet the demands of practical lifelong adaptation at scale. To bridge this gap, we propose UltraEdit, a training-, subject-, and memory-free approach that is well-suited for ultra-scalable, real-world lifelong model editing. UltraEdit fundamentally differs from traditional paradigms by computing parameter shifts in one step using only a hidden state and its gradient, making the approach simple yet efficient. To improve scalability in lifelong settings, UltraEdit employs a lifelong normalization strategy that continuously updates feature statistics across turns, allowing it to adapt to distributional shifts and maintain consistency over time. UltraEdit achieves editing speeds more than $7\times$ faster than the previous state-of-the-art method, while requiring $4\times$ less VRAM. This makes it the only method currently capable of editing a 7B LLM on a 24GB consumer-grade GPU. Furthermore, we construct UltraEditBench, the largest dataset in the field to date with over 2M editing pairs, and demonstrate that our method supports up to 2M edits while maintaining high accuracy. Comprehensive experiments on five datasets and six models show that UltraEdit consistently achieves superior performance across diverse model editing scenarios, taking a further step towards safe and scalable lifelong learning. Our code is available at this https URL.
Comments:
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2505.14679 [cs.CL]
  (or arXiv:2505.14679v3 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2505.14679
Focus to learn more
arXiv-issued DOI via DataCite

Submission history

From: Xiaojie Gu [view email]
[v1] Tue, 20 May 2025 17:59:04 UTC (630 KB)
[v2] Fri, 26 Sep 2025 04:37:01 UTC (632 KB)
[v3] Tue, 10 Mar 2026 07:07:15 UTC (617 KB)
Full-text links:

Access Paper:

    View a PDF of the paper titled UltraEdit: Training-, Subject-, and Memory-Free Lifelong Editing in Language Models, by Xiaojie Gu and 3 other authors
  • View PDF
  • TeX Source
Current browse context:
cs.CL
< prev   |   next >
Change to browse by:

References & Citations

export BibTeX citation Loading...

BibTeX formatted citation

×
Data provided by:

Bookmark

BibSonomy logo Reddit logo
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?)
Links to Code Toggle
Papers with Code (What is Papers with Code?)
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?)
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.