Code Generation on MBPP 256
57.6AccuracyFull
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FullBackbone=Dream-7B-Base, Sampling Strategy=Full-step2026.04 | 57.6 | 256 | 29.7 | |
| NI SamplingBackbone=Dream-7B-Base2026.04 | 56 | 68.87 | 106.7 | |
| ThresholdBackbone=Dream-7B-Base, Sampling Strategy=Confidence Threshold2026.04 | 53.4 | 92.41 | 82.6 | |
| NI SamplingModel=LLaDA 1.52026.04 | 38.6 | 28.92 | 201.8 | |
| FullModel=LLaDA 1.52026.04 | 38.4 | 256 | 23.3 | |
| ThresholdModel=LLaDA 1.52026.04 | 38.4 | 41.89 | 143 | |
| ThresholdModel=LLaDA-8B-Instruct2026.04 | 38 | 43.01 | 137 | |
| NI SamplingModel=LLaDA-8B-Instruct2026.04 | 37.4 | 30.08 | 192.4 | |
| FullModel=LLaDA-8B-Instruct2026.04 | 36.6 | 256 | 22.8 |