query regarding the reference model

[SachinVashisth]

Hi, for the task of imdb_text_continuation, I looked into the config file gpt2_ppo_on_supervised.yml where policy contains model_name: rajkumarrrk/gpt2-fine-tuned-on-imdb-positive-reviews. This is the model that will be improved but which is the reference model?
I want to provide the checkpoints for both the reference model and the model that I want to improve i.e. policy model. In fact, they are the same models.