Training/Testing datafiles for finetuning

[sfillwo]

Finetuning blender expects all_train_responses.json and all_test_responses.json but the only data that is provided is organized_Reflect_9k_responses.json. Based on the agents.py code, the format of organized_Reflect_9k_responses.json is different than that of the expected files for finetuning. Will training/testing data files themselves be released, or at least the preprocessing code to convert organized_Reflect_9k_responses.json?

[yana-xuyan]

I don't think the authors collected training data in this dataset. It's just a test set from what I understand...