Bigger RLHF model with the reward model from annaovesnaatatt/reward-model. The reward model was trained on 3000 examples from the Anthropic/hh-rlhf dataset.