Multiple-Choice Reading on LogiQA MCR 2.0 (test)
54.4AccuracyLSS (Phi-3-Mini)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LSS (Phi-3-Mini)Base Model=Phi-3-Mini, Steering=LSS2026.04 | 54.4 | 2.6 | |
| LSS (Llama-3.1-8B)Base Model=Llama-3.1-8B, Steering=LSS2026.04 | 51.8 | 3 | |
| Phi-3-MiniSteering=Greedy2026.04 | 51.8 | — | |
| Llama-3.1-8BSteering=Greedy2026.04 | 48.8 | — |