Code Generation on HumanEval 512
40.85AccuracyFull
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FullModel=LLaDA 1.52026.04 | 40.85 | 512 | 32.1 | |
| NI SamplingModel=LLaDA 1.52026.04 | 39.63 | 105.3 | 144.2 | |
| ThresholdModel=LLaDA 1.52026.04 | 39.02 | 158.4 | 101.1 | |
| NI SamplingModel=LLaDA-8B-Instruct2026.04 | 36.59 | 88.45 | 172.4 | |
| FullModel=LLaDA-8B-Instruct2026.04 | 35.37 | 512 | 31.5 | |
| ThresholdModel=LLaDA-8B-Instruct2026.04 | 35.37 | 158.6 | 100.8 |