This repository contains the supplementary code and features for our submission. 
Specifically, train.py and generator.py are excerpts from our actual Tokenizer implementation, 
while metrics.py provides the evaluation code for computing alignment metrics such as CKNNA. 
We also include precomputed features from our Tokenizer and from related VFM model. 
After installing the dependencies in requirements.txt, 
running metrics.py with the provided feature files will reproduce our results, 
giving a CKNNA score of about 0.188.