Generation Probabilities: How to compute probabilities of output scores for GPT2

Thank you! But why are the scores the same in my example, given that they “correspond” to different texts? In addition, how can I “restore” the correct sequence having length_penalty=0?