Natural Language Inference on LogiQA NLI 2.0 (test)
56.2AccuracyLSS (Llama-3.1-8B)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LSS (Llama-3.1-8B)Base Model=Llama-3.1-8B, Steering=LSS2026.04 | 56.2 | 4.4 | |
| LSS (Phi-3-Mini)Base Model=Phi-3-Mini, Steering=LSS2026.04 | 56.2 | 3.6 | |
| Phi-3-MiniSteering=Greedy2026.04 | 52.6 | — | |
| Llama-3.1-8BSteering=Greedy2026.04 | 51.8 | — |