Saved in:
Bibliographic Details
Main Authors: Morris, John X., Sitawarin, Chawin, Guo, Chuan, Kokhlikyan, Narine, Suh, G. Edward, Rush, Alexander M., Chaudhuri, Kamalika, Mahloujifar, Saeed
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.24832
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913899741184000
author Morris, John X.
Sitawarin, Chawin
Guo, Chuan
Kokhlikyan, Narine
Suh, G. Edward
Rush, Alexander M.
Chaudhuri, Kamalika
Mahloujifar, Saeed
author_facet Morris, John X.
Sitawarin, Chawin
Guo, Chuan
Kokhlikyan, Narine
Suh, G. Edward
Rush, Alexander M.
Chaudhuri, Kamalika
Mahloujifar, Saeed
contents We propose a new method for estimating how much a model knows about a datapoint and use it to measure the capacity of modern language models. Prior studies of language model memorization have struggled to disentangle memorization from generalization. We formally separate memorization into two components: unintended memorization, the information a model contains about a specific dataset, and generalization, the information a model contains about the true data-generation process. When we completely eliminate generalization, we can compute the total memorization, which provides an estimate of model capacity: our measurements estimate that GPT-style models have a capacity of approximately 3.6 bits per parameter. We train language models on datasets of increasing size and observe that models memorize until their capacity fills, at which point "grokking" begins, and unintended memorization decreases as models begin to generalize. We train hundreds of transformer language models ranging from $500K$ to $1.5B$ parameters and produce a series of scaling laws relating model capacity and data size to membership inference.
format Preprint
id arxiv_https___arxiv_org_abs_2505_24832
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle How much do language models memorize?
Morris, John X.
Sitawarin, Chawin
Guo, Chuan
Kokhlikyan, Narine
Suh, G. Edward
Rush, Alexander M.
Chaudhuri, Kamalika
Mahloujifar, Saeed
Computation and Language
We propose a new method for estimating how much a model knows about a datapoint and use it to measure the capacity of modern language models. Prior studies of language model memorization have struggled to disentangle memorization from generalization. We formally separate memorization into two components: unintended memorization, the information a model contains about a specific dataset, and generalization, the information a model contains about the true data-generation process. When we completely eliminate generalization, we can compute the total memorization, which provides an estimate of model capacity: our measurements estimate that GPT-style models have a capacity of approximately 3.6 bits per parameter. We train language models on datasets of increasing size and observe that models memorize until their capacity fills, at which point "grokking" begins, and unintended memorization decreases as models begin to generalize. We train hundreds of transformer language models ranging from $500K$ to $1.5B$ parameters and produce a series of scaling laws relating model capacity and data size to membership inference.
title How much do language models memorize?
topic Computation and Language
url https://arxiv.org/abs/2505.24832