In the Model Drifting tab, you can see the attack success rate (ASR) trends over time for a selected AI target so you can manage your targets and prevent model degradation/decay. For each target, you can track its performance based on the data gathered from all the prompts and responses overtime.
The Attack Success Rate line graph shows how the ASR changes across the 10 most recent test rounds. Under Target, you can choose the AI target to analyze. By default, it:
-
Selects from the top 100 targets (by creation date) with the highest number of test rounds.
-
If multiple targets share the same maximum number of test rounds, it displays the most recently created target.
You can also see a tabular view of the recent test history with ASR progression for the chosen target.
