ethz-spylab / satml-llm-ctf

Code used to run the platform for the LLM CTF colocated with SaTML 2024
https://ctf.spylab.ai
MIT License
25 stars 6 forks source link

Contents of `chat_history` #31

Closed s-zanella closed 10 months ago

s-zanella commented 10 months ago

The contents of the chat_history parameter passed to a Python filter have changed with respect to the original API. They now include not only the user prompts and filtered assistant responses, but also unfiltered assistant responses and intermediate results after applying the first filter. I think this change might have been introduced at the time the "Debug defense" functionality was added to the web interface. Is this change intended?

dedeswim commented 10 months ago

Hi, thanks a lot for pointing this out. This was indeed not intended, and made us realise about another bug, which we have now fixed. Please see our latest Google Group announcement.

Sorry for the inconvenience!

s-zanella commented 10 months ago

No worries. Thanks for the quick fix!