LLM Redteam Framework — Live

Runs redteam-eval via dashboard_server, streams stdout · Source repo ↗
dashboard_server not running. Start it to run live evals: python dashboard_server.py
Metrics below are from the last committed benchmark run.
Run evaluation
~5–15s · offline, no API calls
Evaluation results
0.97
F1 — grouped held-out
deepset/prompt-injections · committed benchmark
→
−0.23 on transfer
0.7188
F1 — novel-phrasing OOD
Out-of-distribution · committed benchmark
0.94
Precision
0.92
Recall
0.97
F1
0.73
Jailbreak ASR
Precision / Recall / F1 by dataset
Confidence score distribution