Large Language Model Debiasing on BBQ and CrowS-Pairs In-Distribution (test)
0.94Mean BiasUGID (OURS)
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| UGID (OURS)Eval Protocol=Main Eval protocol, Backbone=LLaMA-3-8B2026.03 | 0.94 | 0.94 | 0.007 | 0.058 | 100 | 121.11 | — | — | — | — | — | |
| KLAAD-LoRAEval Protocol=Main Eval protocol, Backbone=LLaMA-3-8B2026.03 | 1.03 | 1.13 | 0.148 | 3.576 | 50 | 10.66 | — | — | — | — | — | |
| CDAEval Protocol=Main Eval protocol, Backbone=LLaMA-3-8B2026.03 | 1.16 | 1.29 | 0.11 | 3.813 | 100 | 3.76 | — | — | — | — | — | |
| SELF-DEBIAS (Ex.)Eval Protocol=Self-Debias Eval protocol, Backbone=LLaMA-3-8B2026.03 | 3.47 | 7.88 | — | — | — | — | — | 5.29 | 18.81 | 1.08 | 1 | |
| ORIGINAL (SD)Eval Protocol=Self-Debias Eval protocol, Backbone=LLaMA-3-8B2026.03 | 4.42 | 11.8 | — | — | — | — | — | 4.32 | 16.91 | 1.19 | 1 | |
| SELF-DEBIAS (Ex.)Eval Protocol=Main Eval protocol, Backbone=LLaMA-3-8B2026.03 | 6.34 | 19.42 | 0.211 | 5.198 | 100 | 118.07 | — | — | — | — | — | |
| ORIGINALEval Protocol=Main Eval protocol, Backbone=LLaMA-3-8B2026.03 | 7.14 | 21.99 | 0.211 | 5.198 | 100 | 118.07 | — | — | — | — | — | |
| SELF-DEBIAS (Re.)Eval Protocol=Self-Debias Eval protocol, Backbone=LLaMA-3-8B2026.03 | 13.04 | 27.28 | — | — | — | — | — | 14.38 | 184.97 | 2.32 | 1 | |
| SELF-DEBIAS (Re.)Eval Protocol=Main Eval protocol, Backbone=LLaMA-3-8B2026.03 | 25.97 | 58.25 | 0.211 | 5.198 | 75 | 118.07 | — | — | — | — | — |