How LoRA Remembers? A Parametric Memory Law for LLM Finetuning

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Xu, Ziwen, Hong, Haiwen, Yu, Linsong, Cui, Benglei, Huang, Longtao, Xue, Hui, Zhang, Ningyu
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918529968635904
author Xu, Ziwen
Hong, Haiwen
Yu, Linsong
Cui, Benglei
Huang, Longtao
Xue, Hui
Zhang, Ningyu
author_facet Xu, Ziwen
Hong, Haiwen
Yu, Linsong
Cui, Benglei
Huang, Longtao
Xue, Hui
Zhang, Ningyu
contents Large Language Models (LLMs) must continuously learn and update knowledge to remain effective in dynamic real-world environments. While Low-Rank Adaptation (LoRA) is widely used for such memory updates, existing studies mainly rely on qualitative downstream evaluations, leaving the quantitative capacity limits and underlying dynamics of exact parametric memory largely unexplored. To bridge this gap, we employ LoRA as a controlled memory capacity probe within the latent space to systematically quantify exact parametric memory. We introduce the Parametric Memory Law, a robust power law linking loss reduction Delta L to effective parameters and sequence length. At the token level, fine-grained analysis reveals a deterministic phase transition, demonstrating that a prediction probability of p > 0.5 constitutes a sufficient condition for verbatim recall under greedy decoding. Driven by these insights, we introduce MemFT, a threshold-guided optimization strategy that dynamically redistributes the training budget toward sub-threshold tokens. Empirical evaluations demonstrate that MemFT can enhance memory fidelity and efficiency. Code will be released at https://github.com/zjunlp/ParametricMemoryLaw.
format Preprint
id arxiv_https___arxiv_org_abs_2605_30260
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle How LoRA Remembers? A Parametric Memory Law for LLM Finetuning
Xu, Ziwen
Hong, Haiwen
Yu, Linsong
Cui, Benglei
Huang, Longtao
Xue, Hui
Zhang, Ningyu
Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
Machine Learning
Large Language Models (LLMs) must continuously learn and update knowledge to remain effective in dynamic real-world environments. While Low-Rank Adaptation (LoRA) is widely used for such memory updates, existing studies mainly rely on qualitative downstream evaluations, leaving the quantitative capacity limits and underlying dynamics of exact parametric memory largely unexplored. To bridge this gap, we employ LoRA as a controlled memory capacity probe within the latent space to systematically quantify exact parametric memory. We introduce the Parametric Memory Law, a robust power law linking loss reduction Delta L to effective parameters and sequence length. At the token level, fine-grained analysis reveals a deterministic phase transition, demonstrating that a prediction probability of p > 0.5 constitutes a sufficient condition for verbatim recall under greedy decoding. Driven by these insights, we introduce MemFT, a threshold-guided optimization strategy that dynamically redistributes the training budget toward sub-threshold tokens. Empirical evaluations demonstrate that MemFT can enhance memory fidelity and efficiency. Code will be released at https://github.com/zjunlp/ParametricMemoryLaw.
title How LoRA Remembers? A Parametric Memory Law for LLM Finetuning
topic Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2605.30260