[USTC-Sec] - [matr1x] - [Why does the Llama-2-7b-chat-hf (ASR) score 0?]

AISG-Technology-Team / GCSS-Track-1A-Submission-Guide

Submission Guide + Discussion Board for AI Singapore Global Challenge for Safe and Secure LLMs (Track 1A).

16 stars 2 forks source link

[USTC-Sec] - [matr1x] - [Why does the Llama-2-7b-chat-hf (ASR) score 0?] #6

Closed hacker-jerry closed 2 months ago

hacker-jerry commented 3 months ago

Hello, we noticed that Llama-2-7b-chat-hf (ASR) has many teams with a score of 0. This is not intuitive as it was able to trigger the vulnerability consistently when tested locally, and wondered if it is possible to view the model output and if it is running out of steam?

aisg-py commented 3 months ago

Hi @hacker-jerry ,

We keep model outputs private to ensure fairness and maintain the integrity of the competition. We continuously monitor all submissions and system performance. Thank you for your understanding.