Open-Ended Generation on TruthfulQA (test)
0.0307Steering EffectiPASa
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| iPASaModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Strength={15, 19}, Intervention Layer={1, 4}, Decoding Strategy=greedy decoding2025.09 | 0.0307 | — | |
| iPASwoModel=Nous-Hermes-2-Mistral-7B-DPO, Steering Strength={15, 19}, Intervention Layer={1, 4}, Decoding Strategy=greedy decoding2025.09 | 0.0307 | — | |
| iPASwoModel=DeepSeek-R1-Distill-Llama-8B, Steering Strength={15, 19}, Intervention Layer={1, 4}, Decoding Strategy=greedy decoding2025.09 | 0.0307 | — | |
| iPASaModel=DeepSeek-R1-Distill-Llama-8B, Steering Strength={15, 19}, Intervention Layer={1, 4}, Decoding Strategy=greedy decoding2025.09 | 0.0123 | — | |
| DeepSeek-R1-Distill-Llama-8BSteering Method=Raw, Decoding Strategy=greedy decoding2025.09 | — | 26.99 | |
| Nous-Hermes-2-Mistral-7B-DPOSteering Method=Raw, Decoding Strategy=greedy decoding2025.09 | — | 28.2 |