Mitigating Post-Training Effects on Generative Diversity in Language Models

artificial intelligence, creative writing, creative writing, diverse sampling, language modeling, model evaluation, post-training, reinforcement learning