I am currently wanting to implement the paper “Text Generation by Learning From Demonstrations” found here: [2009.07839] Text Generation by Learning from Demonstrations, but not sure if there is a way already in OpenNMT to implement these methods. I would love to use this at Validation time, it looks like a GOLD score is already present during onmt_translate, but please let me know if there is a way to log the average GOLD score for validation runs, thank you!