Code Generation on HumanEval (Accuracy, Accuracy Gain, Speedup)
55.5AccuracyProphet
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ProphetModel=Dream-7B-Instruct, Decoding Strategy=Prophet, Evaluation Setting=Zero-shot2025.08 | 55.5 | 60 | 1.44 | |
| Full-step decodingModel=Dream-7B-Instruct, Decoding Strategy=Full-step, Evaluation Setting=Zero-shot2025.08 | 54.9 | — | — | |
| Full-step decodingModel=LLaDA-8B-Instruct, Decoding Strategy=Full-step, Evaluation Setting=Zero-shot2025.08 | 30.5 | — | — | |
| ProphetModel=LLaDA-8B-Instruct, Decoding Strategy=Prophet, Evaluation Setting=Zero-shot2025.08 | 30.5 | 0 | 1.2 |