Published June 2, 2025 | Version v1
Software Open

Artifact of the paper: Interval-Asynchrony: Delimited Intervals of Localised Asynchrony for Fast Parallel SGD

  • 1. EDMO icon Chalmers University of Technology
  • 2. ROR icon University of Gothenburg

Description

This artifact allows you to reproduce the main results of the paper “Interval-Asynchrony: Delimited Intervals of Localised Asynchrony for Fast Parallel SGD” as presented in Euro-Par 2025. Specifically, it will generate plots corresponding to figures 4 & 5 in the paper, which compare the accuracy and training time of our method to the main two baselines – fully asynchronous with and without adaptive parallelism. Please refer to the paper for more details about our algorithm and the baselines.

To use this artifact, please refer to artifact_documentation.pdf, which resides within the ZIP archive. Building and running the code is achieved by executing a single script.

The following software is all you need to ensure manually. The included script will install a few necessary Python dependencies automatically, contained within a virtual environment.
• Linux. We used Linux 6.13.1, but there’s no reason that any recent version wouldn’t work.
• C++ compiler toolchain with support for C++20. We used GNU g++ 14.2.1.
• CMake. The specific version shouldn’t matter as long as it’s recent, but we used CMake 3.31.7.
• Python 3 (for rendering plots). We used Python 3.13.2.

Files

artifact-interval-asynchrony.zip

Files (1.3 MB)

Name Size Download all
md5:41669f5359802eb2c87d458a6d296ac6
1.3 MB Preview Download

Additional details

Software

Repository URL
https://github.com/j4cobgarby/shared-memory-sgd
Programming language
C++, Python
Development Status
Active