Developed by Maxime MĂ©loux and Van Duy Ngo for the Terminology class of the NLP Master in IDMC, Nancy.
- Install the required packages by using
pip install -r requirements.txt
andpython -m spacy download en_core_web_sm
. - Download the required additional files and move them into the folders
embeds/
(forenglish_fasttext_2017_10.vectors.npy
andenglish_fasttext_2017_10
) andoutput/
(forfinal-model.pt
). - The full training and inference process can be triggered by running
main.py
. - Alternatively,
sandbox.py
can be run as a standalone program to play with TARS' capabilities.
Warning: TARS requires a lot of VRAM for training. We recommend using Grid5k or any other similar resource. However, inference can be performed using common resources such as a personal computer or laptop.
The results of running the fine-tuned TARS tagger on the test set can be found in output/tars_results.txt
.