Saved in:
Bibliographic Details
Main Authors: Rafiuddin, S M, Khan, Muntaha Nujat
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.08798
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914085735497728
author Rafiuddin, S M
Khan, Muntaha Nujat
author_facet Rafiuddin, S M
Khan, Muntaha Nujat
contents Transformer attention scales quadratically with sequence length O(n^2), limiting long-context use. We propose Adaptive Retention, a probabilistic, layer-wise token selection mechanism that learns which representations to keep under a strict global budget M. Retention is modeled with Bernoulli gates trained via a Hard-Concrete/variational relaxation and enforced with a simple top-M rule at inference, making the method differentiable and drop-in for standard encoders. Across classification, extractive QA, and long-document summarization, keeping only 30-50% of tokens preserves >= 95% of full-model performance while cutting peak memory by ~35-45% and improving throughput by up to ~1.8x. This architecture-agnostic approach delivers practical long-context efficiency without modifying base attention or task heads.
format Preprint
id arxiv_https___arxiv_org_abs_2510_08798
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Learning What to Remember: Adaptive Probabilistic Memory Retention for Memory-Efficient Language Models
Rafiuddin, S M
Khan, Muntaha Nujat
Computation and Language
Transformer attention scales quadratically with sequence length O(n^2), limiting long-context use. We propose Adaptive Retention, a probabilistic, layer-wise token selection mechanism that learns which representations to keep under a strict global budget M. Retention is modeled with Bernoulli gates trained via a Hard-Concrete/variational relaxation and enforced with a simple top-M rule at inference, making the method differentiable and drop-in for standard encoders. Across classification, extractive QA, and long-document summarization, keeping only 30-50% of tokens preserves >= 95% of full-model performance while cutting peak memory by ~35-45% and improving throughput by up to ~1.8x. This architecture-agnostic approach delivers practical long-context efficiency without modifying base attention or task heads.
title Learning What to Remember: Adaptive Probabilistic Memory Retention for Memory-Efficient Language Models
topic Computation and Language
url https://arxiv.org/abs/2510.08798