Introduction:

The “src” folder contains the code for all experiments in the paper, but running them might be a bit annoying. Sorry for not well organizing the code more systematically. However, as the code is based on this repo and the official DPO repo, and the experiments are easy to understand, it is easy for you to write your own code using your favorite framework.

The training and probing data:

• Prepare the data files
• Specify which data files are used for training

The training and evaluation pipeline:

• Training Pipeline
• Where the results are saved and how to analyze them

Other explanations of the code

• Structure of the code
• How different metrics are calculated