Code for The Grammar-Learning Trajectories of Neural Language Models
Note that most of the project code is in borgr-code dir in transformers. Ther rest of the code is redistributed code of KenLM and Transformers (with their own licenses) that is redistributed for ease of making the experiments anew. Transformers versions were already a bit problematic with the different models used. The code is very messy, but honestly, I doubt it will be massively reused, so if you need anything send a line I might just help directly instead of cleaning it for no reason..