MEAD

MEAD is a library for reproducible deep learning research and fast model development for NLP. It provides easily extensible abstractions and implementations for data loading, model development, training, experiment tracking and export to production.

It also provides implementations of high-performance deep learning models for various NLP tasks, against which newly developed models can be compared. Deep learning experiments are hard to reproduce, MEAD provides functionalities to track them. The goal is to allow a researcher to focus on model development, delegating the repetitive tasks to the library.

Documentation

Tutorials using Colab

MEAD Hub

Installation

Pip

Baseline can be installed as a Python package.

pip install mead-baseline

If you are using tensorflow 2 as your deep learning backend you will need to have tensorflow_addons already installed or have it get installed directly with:

pip install mead-baseline[tf2]

Note for TF 2.1 users: If you are using TF 2.1, you cannot just pip install tensorflow_addons (or the command above) -- it will pull a version that is dependent on a more recent version with breaking changes. If you are running TF 2.1, use a pinned version of the addons: pip install tensorflow_addons==0.9.1

From the repository

If you have a clone of this repostory and want to install from it:

cd layers
pip install -e .
cd ../
pip install -e .

This first installs mead-layers AKA 8 mile, a tiny layers API containing PyTorch and TensorFlow primitives, locally and then mead-baseline

Dockerhub

We use Github CI/CD to automatically cut releases for TensorFlow (1.x and 2.x) and PyTorch via this project:

https://github.com/mead-ml/mead-gpu

Links to the latest dockerhub images can be found there

A Note About Versions

Deep Learning Frameworks are evolving quickly and changes are not always backwards compatible. We recommend recent versions of whichever framework is being used underneath. We currently run on TF versions between 1.13 and 2.3, and we recommend using at least TF 2.1. The PyTorch backend requires at least version 1.3.0, though we recommend using a more recent version.

Citing

If you use the library, please cite the following paper:

@InProceedings{W18-2506,
  author =    "Pressel, Daniel
               and Ray Choudhury, Sagnik
               and Lester, Brian
               and Zhao, Yanjie
               and Barta, Matt",
  title =     "Baseline: A Library for Rapid Modeling, Experimentation and
               Development of Deep Learning Algorithms targeting NLP",
  booktitle = "Proceedings of Workshop for NLP Open Source Software (NLP-OSS)",
  year =      "2018",
  publisher = "Association for Computational Linguistics",
  pages =     "34--40",
  location =  "Melbourne, Australia",
  url =       "http://aclweb.org/anthology/W18-2506"
}

MEAD was selected for a Spotlight Poster at the NeurIPS MLOSS workshop in 2018. OpenReview link