Skip to content

README.md

Latest commit

 

History

History
47 lines (29 loc) · 1.97 KB

README.md

File metadata and controls

47 lines (29 loc) · 1.97 KB
Raw

Home | Data | Package | Performance Scores | Predictions | Results

Package

The folder Core is a wrapper for the HuggingFace Transformers library. This wrapper implements many higher-level functions : formatting data, training models, prediction... This makes HuggingFace Transformers easier to use for people lacking a computer science background when dealing with non-standard textual data.

WARNING : recent GPU required. Our configuration (and subsequent batch size) makes use of a NVidia V100 (can be used for free using Google Colab for instance)

Installing locally the repository

Requirements

Software

  • Unix-based OS (OSX/Linux)
  • Recent version of pip
  • [Optionnal] Conda (we recommend installing miniconda). Conda enables to install all needed packages in a news, clean environment separated from the user's base environment.

Hardware

  • GPU : we use one Nvidia V100, which has 32Go of RAM. Batch sizes of our models were computed so that they make full use of the graphic cards.

Installing the required packages

cd Core
conda create --name AugmentedSocialScientist python pip jupyter
conda activate AugmentedSocialScientist
pip install .
cd ..

Uninstall:

pip uninstall AugmentedSocialScientist

Troubleshooting

  • conda not found -> Make sure that conda is properly configured with your shell (add export PATH="/home/username/miniconda3/bin:$PATH" in your .bash_profile or try conda init --help)
  • No matching distribution found for ... -> Update your version of pip, or use pip3
  • In case this doesn't work, you can install by yourself the packages listed in Core/setup.py