Great Job! Would the "Clean" training set (i.e., after removing the identified harmful sample) you used to re-train the VLLMs be released?

ys-zong / VLGuard

[ICML 2024] Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Models.

https://ys-zong.github.io/VLGuard/

45 stars 2 forks source link

Great Job! Would the "Clean" training set (i.e., after removing the identified harmful sample) you used to re-train the VLLMs be released? #6

Open pzs19 opened 2 weeks ago

ys-zong commented 2 weeks ago

Yes, I'll aim to release it soon. At the same time, if you want to reproduce it yourself, you can take the LlamaGuard model to iterate over the whole dataset and filter the samples labelled as "unsafe".