Is there a model that takes a question/query as its sole input, constructs the entity and attribute graphs from the query, and outputs the state transitions? I see in 4_KIEST_cosntrain_reward/training/generation.py that there is a predictor that takes the question together with the graphs as input, as well as a second predictor in 6_test_our_model/generation.py that takes only the question. However, the second one isn't covered in the readme; we're only instructed to run evaluation on already-generated outputs in 6_test_our_model as far as I can see. I haven't been able to determine where its checkpoint can be found. Thanks!