Let’s use byte pair encoding tokenization along with Poisson regression to understand which tokens are more more often (or less often) in US place …