Mathematical Reasoning on ASDiv Out of Distribution (Top-1 Accuracy)
89.1Top-1 Accuracy (maj@1)CausalRM
Evaluation Results
| Method | Links | |
|---|---|---|
| CausalRMEvaluation Type=Downstream RLHF2026.01 | 89.1 | |
| GoalRMEvaluation Type=Downstream RLHF2026.01 | 88.4 | |
| Standard RMEvaluation Type=Downstream RLHF2026.01 | 88 | |
| SFTEvaluation Type=Downstream RLHF2026.01 | 82.4 | |
| SIKeD (All)Model=Gemma 7B2024.10 | 70.76 | |
| SIKeD (Adaptive)Model=Gemma 7B2024.10 | 70.59 | |
| CombinedModel=Gemma 7B2024.10 | 69.11 | |
| SIKeD (Adaptive)Model=Qwen 1.5B2024.10 | 68.98 | |
| CoTModel=Gemma 7B2024.10 | 68.76 | |
| CoTModel=Qwen 1.5B2024.10 | 68.76 | |
| SIKeD (All)Model=Qwen 1.5B2024.10 | 68.55 | |
| PoTModel=Gemma 7B2024.10 | 67.85 | |
| CombinedModel=Qwen 1.5B2024.10 | 67.64 | |
| PoTModel=Qwen 1.5B2024.10 | 66.94 | |
| L2MModel=Gemma 7B2024.10 | 64.69 | |
| L2MModel=Qwen 1.5B2024.10 | 63.69 | |
| SIKeD (Adaptive)Model=Gemma 2B2024.10 | 59.05 | |
| SIKeD (All)Model=Qwen 0.5B2024.10 | 58.61 | |
| SIKeD (Adaptive)Model=Qwen 0.5B2024.10 | 58.44 | |
| PoTModel=Gemma 2B2024.10 | 58.13 | |
| SIKeD (All)Model=Gemma 2B2024.10 | 58 | |
| CombinedModel=Gemma 2B2024.10 | 57.96 | |
| CombinedModel=Qwen 0.5B2024.10 | 57.79 | |
| PoTModel=Qwen 0.5B2024.10 | 56.83 | |
| CoTModel=Qwen 0.5B2024.10 | 54.66 | |
| CoTModel=Gemma 2B2024.10 | 54.01 | |
| L2MModel=Qwen 0.5B2024.10 | 49.76 | |
| SIKeD (All)Model=SmolLM 1.7B2024.10 | 48.16 | |
| SIKeD (Adaptive)Model=SmolLM 1.7B2024.10 | 47.2 | |
| CombinedModel=SmolLM 1.7B2024.10 | 46.77 | |
| InfoRMEvaluation Type=Downstream RLHF2026.01 | 46.3 | |
| PoTModel=SmolLM 1.7B2024.10 | 43.77 | |
| L2MModel=Gemma 2B2024.10 | 43.47 | |
| CoTModel=SmolLM 1.7B2024.10 | 30.37 | |
| L2MModel=SmolLM 1.7B2024.10 | 22.13 |