Computing change sores

[shanefeng123]

Hi,

In your current implementation for change_scores.py, you are loading the dataset with a prompt column. However, in the originalhh_rlhf_harmless dataset, or after reformatting it with your provided code, there's no such column. How did you evaluate the change scores exactly?

[laujianmin]

exactly! Same Issue

[chenjianhuii]

Sorry for missing these issues! All the evaluation datasets are the form of {'prompt': xxx} which can be obtained via prepare_test_data.py