Mathematical Reasoning on GSM8K (Acc, Steps)
79.55AccuracyTop-1
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Top-1Parallel=No, Backbone=Dream 7B Instruct2025.11 | 79.55 | 256 | |
| KLASSParallel=Yes, Backbone=Dream 7B Instruct2025.11 | 79.43 | 155.67 | |
| KL divergenceParallel=Yes, Backbone=Dream 7B Instruct2025.11 | 76.7 | 150.02 | |
| KLASSParallel=Yes, Backbone=LLaDA 8B Instruct2025.11 | 76.5 | 98.57 | |
| ConfidenceParallel=Yes, Backbone=LLaDA 8B Instruct2025.11 | 75.21 | 74.35 | |
| Top-1Parallel=No, Backbone=LLaDA 8B Instruct2025.11 | 75.13 | 256 | |
| KL divergenceParallel=Yes, Backbone=LLaDA 8B Instruct2025.11 | 74.52 | 155.88 | |
| ConfidenceParallel=Yes, Backbone=Dream 7B Instruct2025.11 | 73.67 | 74.81 | |
| Top-2Parallel=Yes, Backbone=LLaDA 8B Instruct2025.11 | 72.4 | 128 | |
| Top-2Parallel=Yes, Backbone=Dream 7B Instruct2025.11 | 71.69 | 128 | |
| RandomParallel=No, Backbone=LLaDA 8B Instruct2025.11 | 67.1 | 256 | |
| RandomParallel=No, Backbone=Dream 7B Instruct2025.11 | 37.35 | 256 |