List your company

Bias Evaluation on BBQ (Averaged across Gender, Nationality, Religion)

87.73Accuracy (Ambiguous)Self-Debiasing
51.205260.687670.1779.6524Feb 4, 2026

Evaluation results

MethodLinks
Self-DebiasingModel=Llama-3.12026.0287.7373.0280.38
Prefix PromptingModel=Llama-3.22026.0282.8169.0375.92
Prefix PromptingModel=Llama-3.12026.0275.1386.0880.61
BBAModel=Llama-3.12026.0274.4688.9581.7
FBAModel=Llama-3.22026.0268.7881.5975.19
BBAModel=Llama-3.22026.0268.0879.9574.02
FBAModel=Llama-3.12026.0266.4588.8777.66
Self-DebiasingModel=Llama-3.22026.0266.0674.1770.11
DDPModel=Llama-3.22026.0260.3883.471.89
IG^2Model=Llama-3.12026.025989.4674.23
BaseModel=Llama-3.22026.0258.9584.6171.78
Auto-DebiasModel=Llama-3.22026.0258.5384.6271.58
DDPModel=Llama-3.12026.0256.8789.9373.4
Auto-DebiasModel=Llama-3.12026.0254.6191.0672.84
BaseModel=Llama-3.12026.0254.591.1872.84
IG^2Model=Llama-3.22026.0252.6175.8464.22