Code Generation on MBPP+ (Pass@1, Pass@k)
69.44Pass@1ASFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ASFTModel=Qwen3-14B2026.07 | 69.44 | 77.81 | |
| EAFTModel=Qwen3-14B2026.07 | 68.88 | 81.03 | |
| CEModel=Qwen3-14B2026.07 | 68.43 | 79.92 | |
| LP-SFTModel=Qwen3-14B2026.07 | 68.2 | 83.73 | |
| PriFT-probBackbone=Qwen3-4B-Instruct2026.06 | 67.99 | — | |
| EntropyModel=WeDLM-8B, Default decoding=true2026.04 | 67.7 | 73.9 | |
| GEMModel=Qwen3-14B2026.07 | 67.59 | 82.46 | |
| DFTModel=Qwen3-14B2026.07 | 66.42 | 69.67 | |
| IMHModel=WeDLM-8B2026.04 | 66.4 | 83.6 | |
| OriginalBackbone=Qwen3-4B-Instruct2026.06 | 65.61 | — | |
| TALRBackbone=Qwen3-4B-Instruct2026.06 | 65.34 | — | |
| PriFT-massBackbone=Qwen3-4B-Instruct2026.06 | 64.81 | — | |
| ASFTBackbone=Qwen3-4B-Instruct2026.06 | 64.29 | — | |
| EAFTBackbone=Qwen3-4B-Instruct2026.06 | 63.76 | — | |
| SFTBackbone=Qwen3-4B-Instruct2026.06 | 63.23 | — | |
| IDFTBackbone=Qwen3-4B-Instruct2026.06 | 61.9 | — | |
| ASFTModel=Qwen3-4B2026.07 | 61.6 | 71.12 | |
| EAFTModel=Qwen3-4B2026.07 | 61.2 | 73.55 | |
| DFTBackbone=Qwen3-4B-Instruct2026.06 | 60.85 | — | |
| CEModel=Qwen3-4B2026.07 | 60.42 | 72.4 | |
| GEMModel=Qwen3-4B2026.07 | 60.19 | 76.21 | |
| LP-SFTModel=Qwen3-4B2026.07 | 60.19 | 78.93 | |
| ConfidenceModel=WeDLM-8B2026.04 | 59.8 | 71.7 | |
| DFTModel=Qwen3-4B2026.07 | 58.99 | 62.68 | |
| MarginModel=WeDLM-8B2026.04 | 49 | 64.3 | |
| CEModel=Llama3.1-8B2026.07 | 46.92 | 62.31 | |
| LP-SFTModel=Llama3.1-8B2026.07 | 46.52 | 66.91 | |
| GEMModel=Llama3.1-8B2026.07 | 46.49 | 65.39 | |
| ASFTModel=Llama3.1-8B2026.07 | 43.36 | 57.21 | |
| EAFTModel=Llama3.1-8B2026.07 | 42.01 | 60.73 | |
| ConfidenceModel=LLaDA-8B, Default decoding=true2026.04 | 40.4 | 52.1 | |
| IMHModel=LLaDA-8B2026.04 | 39.9 | 67.2 | |
| MarginModel=LLaDA-8B2026.04 | 38.8 | 58.9 | |
| EntropyModel=LLaDA-8B2026.04 | 38.7 | 65.6 | |
| DFTModel=Llama3.1-8B2026.07 | 30.7 | 35.98 | |
| RandomModel=LLaDA-8B2026.04 | 25.1 | 60.5 |