General
MergeBench
Released safety assessment records for two RegMeanPlusPlus merged language models, with explicit attribution to WildGuardTest, HarmBench, XSTest and Do Anything Now.
1,813items
2subjects
The captured MergeBench repository does not declare a license for its code or cached output archive; the old MIT label was unsupported. The linked safety-eval fork has Apache-2.0 code terms. Original component datasets retain their own terms (WildGuardMix ODC-BY; XSTest CC-BY-NC-4.0, as documented by their sources). The paper and website do not establish a separate output-archive redistribution grant. Captured data and formatted tables remain local pending clarification; this metadata is not a license grant.license
generaldomain
textmodality
item-level responses released
Response matrix
Loading response matrix…
Correct (1)Incorrect (0)Unobserved
Scale: 1 = correct · 0 = incorrect