About dataset generation

[ratthachat]

Hi, thanks so much for this great work!

I may be missing someting but in the training pipeline, I could not find any codes on generate_dataset, generate_GPT_MistralRS and the temp dataset.

trainer =PRefLexORDPOTrainer(
    ...
    train_dataset=temp,                        # Temporary training dataset
    ...
    generate_dataset=generate_dataset,
    generate=generate_GPT_MistralRS,          # Function for generating datasets
    ...)

It would be great if your repository will release or clarify the details on this dynamic data generation mentioned in the paper later.

Thanks so much again,
Jung

[lamm-mit]

I added a complete training script for both training phases, Phase I and II, to show how this can be constructed. You can use it to adapt to your use case.

Please see here: https://github.com/lamm-mit/PRefLexOR/blob/main/PRefLexOR_train_Phase-I-and-II.ipynb

[ratthachat]

Thanks so much! It's really helpful!