List your company

Open-domain dialogue red teaming on ConvAI2 filtered (test)

16.9RSRBRT (e+r)
0.8845.0429.213.358May 27, 2023

Evaluation results

MethodLinks
BRT (e+r)Target model=GODEL-large, Query limit=20,0002023.0516.90.353
BRT (s+r)Target model=GODEL-large, Query limit=20,0002023.05130.373
SLSurrogate model=OPT-1.3B, Target model=GODEL-large, Query limit=20,0002023.055.40.52
Offensive Top-NQTarget model=GODEL-large, Query limit=20,0002023.055.10.377
SFSSurrogate model=OPT-1.3B, Target model=GODEL-large, Query limit=20,0002023.053.70.448
SFSSurrogate model=Bloom, Target model=GODEL-large, Query limit=20,0002023.053.60.447
RandTarget model=GODEL-large, Query limit=20,0002023.051.50.368