Memory-based Language Models : An Efficient, Explainable, and Eco-friendly Approach to Large Language Modeling
(2025)- Abstract
- We present memory-based language modeling as an efficient, eco-friendly alternative to deep neural network-based language modeling. It offers log-linearly scalable next-token prediction performance and strong memorization capabilities. Implementing fast approximations of k-nearest neighbor classification, memory-based language modeling leaves a relatively small ecological footprint both in training and in inference mode, as it relies fully on CPUs and attains low token latencies. Its internal workings are simple and fully transparent. We compare our implementation of memory-based language modeling, OLIFANT, with GPT-2 and GPT-Neo on next-token prediction accuracy, estimated emissions and speeds, and offer some deeper analyses of the model.
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/a6723008-96f5-414f-8ad3-7f4674c2e745
- author
- van den Bosch, Antal
; Risco Patón, Ainhoa
; Buijse, Teun
; Berck, Peter
LU
and van Gompel, Maarten
- organization
- publishing date
- 2025-10-25
- type
- Working paper/Preprint
- publication status
- published
- subject
- keywords
- AI, machine learning, language modelling
- pages
- 15 pages
- publisher
- arXiv.org
- language
- English
- LU publication?
- yes
- id
- a6723008-96f5-414f-8ad3-7f4674c2e745
- alternative location
- https://arxiv.org/abs/2510.22317
- date added to LUP
- 2025-11-17 12:45:16
- date last changed
- 2025-12-01 16:23:17
@misc{a6723008-96f5-414f-8ad3-7f4674c2e745,
abstract = {{We present memory-based language modeling as an efficient, eco-friendly alternative to deep neural network-based language modeling. It offers log-linearly scalable next-token prediction performance and strong memorization capabilities. Implementing fast approximations of k-nearest neighbor classification, memory-based language modeling leaves a relatively small ecological footprint both in training and in inference mode, as it relies fully on CPUs and attains low token latencies. Its internal workings are simple and fully transparent. We compare our implementation of memory-based language modeling, OLIFANT, with GPT-2 and GPT-Neo on next-token prediction accuracy, estimated emissions and speeds, and offer some deeper analyses of the model.}},
author = {{van den Bosch, Antal and Risco Patón, Ainhoa and Buijse, Teun and Berck, Peter and van Gompel, Maarten}},
keywords = {{AI; machine learning; language modelling}},
language = {{eng}},
month = {{10}},
note = {{Preprint}},
publisher = {{arXiv.org}},
title = {{Memory-based Language Models : An Efficient, Explainable, and Eco-friendly Approach to Large Language Modeling}},
url = {{https://arxiv.org/abs/2510.22317}},
year = {{2025}},
}