Adaptive Memory Scaling: Reproducibility Artifact for Resource-Constrained Transformer Experiments

Read full text
Authors: Anonymous
Cited by: 0

Abstract

This record provides the reproducibility artifact accompanying the manuscript “Adaptive Memory Scaling as an Emerging Mechanism for Stable Transformer-Based Computing under Resource Constraints.” The artifact contains experimental notebooks, configurations, and result files for the controlled multi-seed English→Hindi and English→Marathi translation experiments, including the Baseline, Fixed-Sqrt, Learnable-Plain, Memory RMSNorm, and AMS configurations. It also contains the controlled Baseline–AMS experiments for XSum abstractive summarization. Per-run summaries and epoch-level training histories are provided to support verification of task-level performance, validation-loss variability, learned scaling behavior, resource measurements, and seed-matched statistical analyses. Third-party datasets are not redistributed; dataset acquisition and preparation information is provided in the accompanying reproducibility documentation.

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
29 August 2026