Results
Results reported in the BrowseSafe++ paper, measured on the private test split. Detection tables classify every test page, attacked or control, so a model is rewarded for catching injections and penalised for false alarms on otherwise identical pages.
Accuracy, latency, cost and size
Detection accuracy on the text views against what each model costs to run.
Submit results
Coming soon.
The leaderboard is computed on the 9,175-pair test split. A submission endpoint that scores a detector on it, view by view and with the same metrics as the tables above, is coming soon. Submission instructions will be posted here.
The training split is available now through the SDK (see Quick start).
shell
pip install "browsesafe-pp[eval] @ git+https://github.com/perplexityai/browsesafe-plus-plus.git"
browsesafe evaluate --predict mymod:predict --split train \
--view html --output results/mymod-html