Data

Tools

Indexes

Applications

Experiments

Open Source Science

gru_aie

A low latency implementation of the Gated Recurrent Unit on the Versal AI Engines

https://github.com/msapkas/gru_aie

Science Score: 44.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • ✓
    CITATION.cff file
    Found CITATION.cff file
  • ✓
    codemeta.json file
    Found codemeta.json file
  • ✓
    .zenodo.json file
    Found .zenodo.json file
  • ○
    DOI references
  • ○
    Academic publication links
  • ○
    Academic email domains
  • ○
    Institutional organization owner
  • ○
    JOSS paper metadata
  • ○
    Scientific vocabulary similarity
    Low similarity (3.4%) to scientific vocabulary
Last synced: 6 months ago · JSON representation ·

Repository

A low latency implementation of the Gated Recurrent Unit on the Versal AI Engines

Basic Info
  • Host: GitHub
  • Owner: Msapkas
  • License: mit
  • Language: C
  • Default Branch: master
  • Size: 653 KB
Statistics
  • Stars: 0
  • Watchers: 1
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Created about 1 year ago · Last pushed 10 months ago
Metadata Files
Readme License Citation

README.md

gru_aie repo

A low latency implementation of the Gated Recurrent Unit on the Versal AI Engines

Different implementations of the model can be found by switching branches. Here are the branches and their descriptions:

  • gruactcompute_linear : The GRU model, using the MAC rows method with kernels that compute the activation functions as splines. A linear layer is connected at the end. (most recent)
  • grureducersw_linear : The GRU model, using MAC rows method with kernels that implement Look Up Tables to calculate the activation functions. A linear layer is connected at the end. (mosts recent and numerically validated)
  • macs_implementation: The GRU model, using the MAC columns method and kernels that implement Look Up Tables for activation functions.
  • master_aggregator: An old GRU model, that is not numericaly tested. In this branch I am testing an idea to double the aggregation limitation from 32 to 64 merges.
  • reducers_implementation: An old GRU model that is using the MAC rows method.

Owner

  • Name: Michail Sapkas
  • Login: Msapkas
  • Kind: user
  • Location: Padova, Veneto, Italy

Physicist "Physics of Data" - Master Student Loving nature, physics, code, and beer.

Citation (CITATION.cff) 

cff-version: 1.2.0
message: "If you use this code, please cite it as below:"
title: "Gated Recurrent Unit Implementation on the AMD Versal AI Engine"
version: 1.0.0
date-released: 2025-05-20
authors:
  - family-names: Sapkas
    given-names: Michail
    affiliation: INFN Department of Physics and Astronomy Padova
    orcid: https://orcid.org/0009-0000-2290-9520
abstract:
  This repository contains the first known implementation of a Gated Recurrent Unit (RNN)
  running on the Versal AI Engine, demonstrating novel usage of Tiled array.
  The code is intended for researchers and developers exploring AI on emerging platforms.
repository-code: https://github.com/Msapkas/gru_aie
license: MIT
keywords:
  - rnn
  - machine learning
  - hardware acceleration
  - michail sapkas
  - neural networks
  - ai engine
  - versal
  - gated recurrent unit

GitHub Events

Total
  • Delete event: 1
  • Public event: 1
  • Push event: 20
  • Create event: 5
Last Year
  • Delete event: 1
  • Public event: 1
  • Push event: 20
  • Create event: 5