Off by one error in model naming

[cbrew]

I found that if I run the train and test commands as in the README.md, which implies a default value of
20 epochs, training creates models 0 through 19, but testing expects model 20, which does not exist.

This is with Python 3.9.1 and the following requirements.txt in a conda-created virtual environment

tqdm==4.56.0
scikit_learn==0.24.1
spacy==2.3.5
torch==1.7.1
tornado==6.1