Question Answering on TruthfulQA (True*Info (TQA) and Rank Metrics)
81.6True*Info Score (TQA)Comm Steer
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Comm SteerBackbone=LLaMA-3.1-8B-Instruct2026.05 | 81.6 | 4.63 | |
| CAABackbone=LLaMA-3.1-8B-Instruct2026.05 | 76.8 | 5.44 | |
| SADIBackbone=LLaMA-3.1-8B-Instruct2026.05 | 75.9 | 3.38 | |
| ITIBackbone=LLaMA-3.1-8B-Instruct2026.05 | 66.8 | 4.75 | |
| DISCO-QBackbone=LLaMA-3.1-8B-Instruct2026.05 | 66.1 | 4.75 | |
| SKOPBackbone=LLaMA-3.1-8B-Instruct2026.05 | 65.9 | 2.69 | |
| CASTBackbone=LLaMA-3.1-8B-Instruct2026.05 | 58.2 | 5.63 | |
| Angular SteerBackbone=LLaMA-3.1-8B-Instruct2026.05 | 56.9 | 4.75 | |
| LoRABackbone=LLaMA-3.1-8B-Instruct2026.05 | 55.4 | — | |
| BaselineBackbone=LLaMA-3.1-8B-Instruct2026.05 | 46.1 | — |