How lm_score and logit_Score is calculated in decorder.py?

I recently started studying language models. I was wondering how `lm_score` and `logit_score` are calculated mathematically in the context of `decoder.py` code.

I would really appreciate the help!