We prepare the data manually. Split it into a train and an eval dataset. Using [prompt-collection-v0.1](https://huggingface.co/datasets/OpenRLHF/prompt-collection-v0 ...
from communicator_objects import unity_output_pb2 as communicator__objects_dot_unity__output__pb2 from communicator_objects import unity_input_pb2 as communicator__objects_dot_unity__input__pb2 from ...