Short-form generation on Short-form generation datasets LOO - near OOD
0.64PRRMSP
Evaluation Results
| Method | Links | |
|---|---|---|
| MSPModel Backbone=Gemma-2-9B2026.04 | 0.64 | |
| HUQ-SATRMDModel Backbone=Gemma-2-9B2026.04 | 0.62 | |
| HUQ-SATMDModel Backbone=Llama 3.1-8B2026.04 | 0.59 | |
| MSPModel Backbone=Llama 3.1-8B2026.04 | 0.57 | |
| HUQ-SATRMDModel Backbone=Llama 3.1-8B2026.04 | 0.55 | |
| HUQ-SATMDModel Backbone=Gemma-2-9B2026.04 | 0.52 | |
| SAPLMA (middle)Model Backbone=Llama 3.1-8B2026.04 | 0.46 | |
| SATMDModel Backbone=Llama 3.1-8B2026.04 | 0.41 | |
| SATRMD-MSPModel Backbone=Gemma-2-9B2026.04 | 0.4 | |
| SAPLMA (middle)Model Backbone=Gemma-2-9B2026.04 | 0.35 | |
| SAPLMA (top)Model Backbone=Gemma-2-9B2026.04 | 0.34 | |
| Uhead (1 layer)Model Backbone=Gemma-2-9B2026.04 | 0.34 | |
| Uhead (2 layers)Model Backbone=Gemma-2-9B2026.04 | 0.33 | |
| SATRMDModel Backbone=Gemma-2-9B2026.04 | 0.33 | |
| SATRMD-MSPModel Backbone=Llama 3.1-8B2026.04 | 0.3 | |
| SATMD-MSPModel Backbone=Llama 3.1-8B2026.04 | 0.28 | |
| SAPLMA (top)Model Backbone=Llama 3.1-8B2026.04 | 0.26 | |
| Uhead (1 layer)Model Backbone=Llama 3.1-8B2026.04 | 0.17 | |
| SATRMDModel Backbone=Llama 3.1-8B2026.04 | 0.17 | |
| SATMDModel Backbone=Gemma-2-9B2026.04 | 0.16 | |
| Uhead (2 layers)Model Backbone=Llama 3.1-8B2026.04 | 0.15 | |
| SATMD-MSPModel Backbone=Gemma-2-9B2026.04 | 0.13 | |
| Lookback-lensModel Backbone=Llama 3.1-8B2026.04 | 0 | |
| Lookback-lensModel Backbone=Gemma-2-9B2026.04 | -0.05 |