Preference Classification on Anthropic HH Helpful (test)
57.6AccuracyUMM-RM
Evaluation results
| Method | Links | |
|---|---|---|
| UMM-RMBackbone=TinyLlama-1.1B, Number of experts=62025.11 | 57.6 | |
| UMM-RMBackbone=TinyLlama-1.1B, Number of experts=42025.11 | 55.2 | |
| Mean OptimizationBackbone=TinyLlama-1.1B, Ensemble Strategy=ensemble RM2025.11 | 55 | |
| Worst-Case OptimizationBackbone=TinyLlama-1.1B, Ensemble Strategy=ensemble RM2025.11 | 54.8 | |
| Uncertainty-Weighted OptimizationBackbone=TinyLlama-1.1B, Ensemble Strategy=ensemble RM2025.11 | 54.6 | |
| UMM-RMBackbone=TinyLlama-1.1B, Number of experts=22025.11 | 54.2 | |
| Dense RMBackbone=TinyLlama-1.1B2025.11 | 44.6 |