Knowledge Recovery on EntityQuestions (test)
36Steered AccuracyPre-trained (MD0) Baseline
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Pre-trained (MD0) BaselineModel=Gemma 2 9B2026.04 | 36 | — | — | |
| Pre-trained (MD0) BaselineModel=Llama 3.1 8B2026.04 | 35.5 | — | — | |
| Pre-trained (MD0) BaselineModel=Mistral v03 7B2026.04 | 31.9 | — | — | |
| MoRFIModel=Llama 3.1 8B, Direction=Unknown, Feature Trend=Decreasing, Intervention=Negative (-), Top Latent=216822026.04 | 23.8 | 70.1 | 495 | |
| MoRFIModel=Gemma 2 9B, Direction=Unknown, Feature Trend=Decreasing, Intervention=Positive (+), Top Latent=51052026.04 | 22.7 | 80.7 | 17 | |
| MoRFIModel=Gemma 2 9B, Direction=Unknown, Feature Trend=Increasing, Intervention=Negative (-), Top Latent=51522026.04 | 22.6 | 85.1 | 495 | |
| MoRFIModel=Mistral v03 7B, Direction=Unknown, Feature Trend=Decreasing, Intervention=Negative (-), Top Latent=78592026.04 | 21.9 | 77.1 | 495 | |
| MoRFIModel=Mistral v03 7B, Direction=Unknown, Feature Trend=Increasing, Intervention=Negative (-), Top Latent=613252026.04 | 21.8 | 83.8 | 495 | |
| MoRFIModel=Llama 3.1 8B, Direction=Unknown, Feature Trend=Decreasing, Intervention=Positive (+), Top Latent=100072026.04 | 21.7 | 72.5 | 495 | |
| MoRFIModel=Llama 3.1 8B, Direction=Unknown, Feature Trend=Increasing, Intervention=Negative (-), Top Latent=206492026.04 | 21.5 | 76.9 | 495 | |
| MoRFIModel=Gemma 2 9B, Direction=Unknown, Feature Trend=Decreasing, Intervention=Negative (-), Top Latent=82772026.04 | 21.3 | 72.3 | 17 | |
| MoRFIModel=Gemma 2 9B, Direction=Unknown, Feature Trend=Increasing, Intervention=Positive (+), Top Latent=183602026.04 | 20.8 | 76.1 | 17 | |
| MoRFIModel=Mistral v03 7B, Direction=Unknown, Feature Trend=Decreasing, Intervention=Positive (+), Top Latent=188482026.04 | 20.6 | 82.2 | 495 | |
| MoRFIModel=Mistral v03 7B, Direction=Unknown, Feature Trend=Increasing, Intervention=Positive (+), Top Latent=418442026.04 | 19.6 | 83 | 495 | |
| MoRFIModel=Llama 3.1 8B, Direction=Unknown, Feature Trend=Increasing, Intervention=Positive (+), Top Latent=271912026.04 | 19.1 | 69 | 17 | |
| Fine-tuned (MD100) BaselineModel=Gemma 2 9B2026.04 | 19.1 | — | — | |
| Fine-tuned (MD100) BaselineModel=Llama 3.1 8B2026.04 | 17.8 | — | — | |
| Fine-tuned (MD100) BaselineModel=Mistral v03 7B2026.04 | 15.7 | — | — |