Open XinyuGuan01 opened 2 years ago
Thanks for the question. You are right! different categories of WebNLG uses the same soft prompt.
I think the current branch has all the setups to reproduce the results, but feel free to checkout the old branch if you are interested. I think those get_prompt_p1... are preliminary experiments.
Thanks for the question. You are right! different categories of WebNLG uses the same soft prompt.
Thank you for your answer. Did this mean that the category information was not used in the proposed method?
Hi, thanks for sharing the codes.
I have tried the webnlg task and data2text task with the 'cleaned' branch. But I found that the "control_code" argument is not used in all the implementations of PrefixTuning.get_prompt(). Does this mean that different categories of webnlg dataset will use the same soft prompt? I found that there are get_prompt_p3, get_prompt_p1 and get_prompt_p4 in the master branch. Can I use them to reproduce the results of the paper?
Thanks.