This is a testing model created for RLHF. The reward model used for training is martin-arguments and was trained on 1000 examples of the Anthropic/hh-rlhf dataset.