Code Generation on HumanEval (Accuracy, Throughput)
83.54AccuracyQwen2.5 7B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen2.5 7BGeneration mode=Autoregressive2026.06 | 83.54 | 45.83 | |
| PoE-BridgeImportance sampling=false2026.06 | 80.69 | 84.82 | |
| PoE-BridgeImportance sampling=true2026.06 | 79.47 | 76.13 | |
| Dream 7BSampling strategy=Entropy, Tokens per step=22026.06 | 47.56 | 17.44 | |
| Dream 7BSampling strategy=Random, Tokens per step=22026.06 | 35.13 | 12.34 | |
| Dream 7BSampling strategy=Entropy, Tokens per step=42026.06 | 32.32 | 30.6 |