Mathematical Reasoning on GSM8K (Acc, NFE)
67.3AccuracySWD
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SWDBackbone=LLaDA-8B, Scoring Metric=Confidence, Decoding Strategy=Stability-Weighted Decoding, lambda=52026.04 | 67.3 | 163.2 | |
| SWDBackbone=Dream-7B, Scoring Metric=Entropy, Decoding Strategy=Stability-Weighted Decoding, lambda=52026.04 | 65 | 132.1 | |
| SWDBackbone=LLaDA-8B, Scoring Metric=Margin, Decoding Strategy=Stability-Weighted Decoding, lambda=52026.04 | 64.4 | 166.9 | |
| SWDBackbone=Dream-7B, Scoring Metric=Confidence, Decoding Strategy=Stability-Weighted Decoding, lambda=52026.04 | 61.6 | 119.3 | |
| SWDBackbone=Dream-7B, Scoring Metric=Margin, Decoding Strategy=Stability-Weighted Decoding, lambda=52026.04 | 61 | 111.5 | |
| SWDBackbone=LLaDA-8B, Scoring Metric=Entropy, Decoding Strategy=Stability-Weighted Decoding, lambda=52026.04 | 48.7 | 191.8 | |
| BaseBackbone=LLaDA-8B, Scoring Metric=Margin, Decoding Strategy=Vanilla EB Sampler2026.04 | 38.7 | 215.8 | |
| BaseBackbone=LLaDA-8B, Scoring Metric=Confidence, Decoding Strategy=Vanilla EB Sampler2026.04 | 31.5 | 224.1 | |
| BaseBackbone=Dream-7B, Scoring Metric=Margin, Decoding Strategy=Vanilla EB Sampler2026.04 | 29.5 | 275.6 | |
| BaseBackbone=Dream-7B, Scoring Metric=Confidence, Decoding Strategy=Vanilla EB Sampler2026.04 | 29.3 | 286.6 | |
| BaseBackbone=Dream-7B, Scoring Metric=Entropy, Decoding Strategy=Vanilla EB Sampler2026.04 | 28.4 | 292.3 | |
| BaseBackbone=LLaDA-8B, Scoring Metric=Entropy, Decoding Strategy=Vanilla EB Sampler2026.04 | 26.3 | 226.6 |