This repository is made to reproduce results from the paper Disentangling semantics in language through VAEs and a certain architectural choice.
A model can be trained by running disentangle_train.py
with default arguments.
The checkpoint for trained parameters used for our analysis is provided under the experiment name
nlilm/StructuredAutoreg5
.
The analysis of the model is performed using the jupyter notebook Model_Analysis.ipynb
under the folder disentanglement_transformer
.
Documentation and usage guidelines are still a work in progress. Don't hesitate to file an issue if you have questions.