Pooled Cross-Model Evaluation Suite
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Pooled Cross-Model Evaluation Suite (all split)
31.2Attack Success Rate (ASR)
5
Pooled Cross-Model Evaluation Suite whitebox
22.2ASR
5
Pooled Cross-Model Evaluation Suite (adaptive split)
27.6Attack Success Rate (ASR)
5
Pooled Cross-Model Evaluation Suite static split
43.8ASR (%)
5
Pooled Cross-Model Evaluation Suite
95.9Utility
4