warpspeed

TensorFlow GSPMD implementation of Microsoft's DeepSpeed library.

https://github.com/qonfused/warpspeed

Science Score: 54.0%

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
    Links to: zenodo.org
  • Academic email domains
  • Institutional organization owner
  • JOSS paper metadata
  • Scientific vocabulary similarity
    Low similarity (6.6%) to scientific vocabulary
Last synced: 6 months ago · JSON representation ·

Repository

TensorFlow GSPMD implementation of Microsoft's DeepSpeed library.

Basic Info
Statistics
  • Stars: 1
  • Watchers: 2
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Created almost 3 years ago · Last pushed almost 3 years ago
Metadata Files
Readme License Citation

README.md

WarpSpeed logo WarpSpeed

TensorFlow GSPMD implementation of Microsoft's DeepSpeed Library.

![DOI](https://zenodo.org/badge/625935606.svg) ![License](https://img.shields.io/badge/⚖_License-Apache%202%2E0-lightblue?labelColor=3f4551) ![SemVer](https://img.shields.io/badge/SemVer-v0.0.0-important?logo=SemVer&labelColor=3f4551)
> **Warning**: This project is still under construction and is not production ready!

This project aims to (i) port optimizations from Microsoft's DeepSpeed library to TensorFlow through various XLA-compiler optimizations like GSPMD, (ii) explore better strategies for mixed-precision quantization for facilitating sparser model architectures, and (iii) provide an extensible API for scalable training and inference.

WarpSpeed is developed in C++ and Python using TensorFlow with API compatibility for JAX + Flax.

License

The code in this repository is licensed under the Apache License 2.0.

Owner

  • Name: Cory Bennett
  • Login: Qonfused
  • Kind: user
  • Location: Austin, TX, USA

AI/ML Research Scientist, Security Researcher, Data Science Enthusiast.

Citation (CITATION.cff)

cff-version: 1.2.0
message: "If you use WarpSpeed in your research, please cite it using this metadata."
title: "WarpSpeed, A Tensorflow GSPMD implementation of Microsoft's DeepSpeed library."
abstract: Warp-Speed is a TensorFlow GSPMD implementation of Microsoft's DeepSpeed library. The project aims to (i) port optimizations from Microsoft's DeepSpeed library to TensorFlow through various XLA-compiler optimizations like GSPMD, (ii) explore better strategies for mixed-precision quantization for facilitating sparser model architectures, and (iii) provide an extensible API for scalable training and inference. WarpSpeed is developed in C++ and Python using TensorFlow with support for JAX + Flax.
authors:
- family-names: "Bennett"
  given-names: "Cory"
identifiers:
  - type: doi
    value: 10.5281/zenodo.7818463
    description: The concept DOI for the collection containing all versions of the Citation File Format.
date-released: "2023-03-06"
license: "Apache-2.0"
doi: 10.5281/zenodo.7818463

GitHub Events

Total
  • Watch event: 1
Last Year
  • Watch event: 1