meta-llama / PurpleLlama

Set of tools to assess and improve LLM security.
Other
2.73k stars 453 forks source link

Will the dataset be released? #21

Closed para-zhou closed 7 months ago

para-zhou commented 7 months ago

Appreciate your nice work. Is there any plan to release the dataset or just test set for comparison? Thanks!

SimonWan commented 7 months ago

Hi there, could you please provide more information, such as whether your question is about Llama Guard or CyberSecEval, and what exact dataset you are looking for? Thanks.

haidequanbu commented 7 months ago

Hey, I have the same question, will the dataset used for training LLaMAGuard be released?

para-zhou commented 7 months ago

hi thank you for asking. I mean for LlamaGuard, Thx!

Hi there, could you please provide more information, such as whether your question is about Llama Guard or CyberSecEval, and what exact dataset you are looking for? Thanks.

ujjwalkarn commented 7 months ago

Hi, we leverage the human preference data from Anthropic to collect the prompts. At this point, we are unable to share the dataset that was used, but more details about how the data was curated is mentioned in the "Data Collection" section of the Llama Guard paper and the model card.