Skip to content

maestro-project/gamma

Repository files navigation

GAMMA

This is the implementation of the paper GAMMA, GAMMA is an autonomous framework for optimizing the HW mapping of DNN models on the DNN Accelerators. We use MAESTRO as our cost model.

GAMMA Framework

Extended Works

This code repository also incorporates the code-bases of the extended works:


Sister Repo: Gamma-Timeloop

  • We also have GAMMA supporting Timeloop as cost model. The sister repo can be found here -- Gamma-Timeloop.

    It enables using GAMMA algorithm to search through the design space of Timeloop, a DNN cost model from NVIDIA.


Installation

  • Create virtual env
conda create --name gammaEnv python=3.6
conda activate gammaEnv
  • Install requirement
pip install -r requirements.txt
  • Download cost model and build symbolic link
python build.py
  • Setup larger limitation for opened file if there is warning "Too many open files." (for threading)
ulimit -n 4096

Take a Trial Run

./run_gamma.sh

Different Map Space Exploration Scenarios

  • Map space exploration for fully flexible accelerator (in a full search space): Related reading - GAMMA.
  • Map space exploration for partially flexible accelerator (in a constrained search space): Related reading - Formalism of Accelerator Flexibility.
  • PE(HW)-Mapping Co-exploration: Related reading - DiGamma.

More details can be found here

Resources

  • Tutorial of GAMMA, in IEEE/ACM International Symposium on Microarchitecture (MICRO), 2020 [video]
  • GAMMA paper presentation, in IEEE/ACM International Conference On Computer Aided Design (ICCAD), 2020 [video]

Contributor

  • Sheng-Chun (Felix) Kao
  • Tushar Krishna

Pull Request

Citation

@inproceedings{gamma,
    author       = {Kao, Sheng-Chun and Krishna, Tushar},
    title        = {GAMMA: Automating the HW Mapping of DNN Models on Accelerators via Genetic Algorithm},
    booktitle     = {ICCAD},
  year          = {2020}
}

@inproceedings{digamma,
title={DiGamma: Domain-aware Genetic Algorithm for HW-Mapping Co-optimization for DNN Accelerators},
author={Kao, Sheng-Chun and Pellauer, Michael and Parashar, Angshuman and Krishna, Tushar},
booktitle     = {DATE},
year={2022}
}
@inproceedings{kao2022formalism,
  title={A Formalism of DNN Accelerator Flexibility},
  author={Kao, Sheng-Chun and Kwon, Hyoukjun and Pellauer, Michael and Parashar, Angshuman and Krishna, Tushar},
  booktitle={Proceedings of the 2022 ACM SIGMETRICS/IFIP PERFORMANCE Joint International Conference on Measurement and Modeling of Computer Systems},
  year={2022}
}

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Contributors 3

  •  
  •  
  •  

Languages