Question about 1-p(unsure) experimental results

[namin-an]

Hello,

Thank you for always replying in detail and helping me for my research!

I also happened to have another question regarding the results! In the OpenReview, I saw that you experimented with 1-p(unsure) instead of the p(yes) ratio! Are the results in the OpenReview (general response) the average performance across the models? If it is, would you mind sharing the InternLM-8b debiasing results using 1-p(unsure)?

Thank you so much!

[LGirrbach]

Hi,

Thank you very much for the kind words!
I assume I still have these results, but I will check them in detail after the CVPR deadline (tomorrow). Thanks for your patience!

Best,
Leander

[LGirrbach]

Sorry for the delay. I found the results:

task	Prompt Tuning	Tuning (Full)	Tuning (LoRA)
occupations	0.35	0.15	0.35
personality traits	0.3	0.1	0.1
skills	0.1	0	0

These are all for InternVL2-8b.
I hope this helps!

[namin-an]

Thanks!🙂