Intuition of calculating the ProGen2 likelihood of the reverse sequence

[georgedashen]

Hi all,

I am curious about the intuition of calculating the likelihood of the reverse order sequence in progen_score in scripts/model.py. Such a procedure of taking the mean of original and reverse order is not adopted for IgLM, another decoder-only model.