Code Generation on HumanEval (greedy pass@1)
54.88HumanEval Accuracy (greedy pass@1)CoDaPO
Evaluation Results
| Method | Links | |
|---|---|---|
| CoDaPO2026.06 | 54.88 | |
| GRPO2026.06 | 50.61 | |
| Dream-7B-Base + DyStructBackbone=Dream-7B, Decoding method=DyStruct, Few-shot examples=0, Unmasking iterations=256, Generation limit=2562026.05 | 47 | |
| FedRot-LoRA2026.02 | 40.88 | |
| Dream-7B-BaseBackbone=Dream-7B, Decoding method=Base, Few-shot examples=0, Unmasking iterations=256, Generation limit=2562026.05 | 40.2 | |
| FFA-LoRA2026.02 | 38.51 | |
| LLaDA-8B-Base + DyStructBackbone=LLaDA-8B, Decoding method=DyStruct, Few-shot examples=0, Unmasking iterations=256, Generation limit=2562026.05 | 34.8 | |
| Dream-7B-Base + DAEDALBackbone=Dream-7B, Decoding method=DAEDAL, Few-shot examples=0, Unmasking iterations=256, Generation limit=2562026.05 | 34.7 | |
| LLaDA-8B-Base + DAEDALBackbone=LLaDA-8B, Decoding method=DAEDAL, Few-shot examples=0, Unmasking iterations=256, Generation limit=2562026.05 | 33.5 | |
| LLaDA-8B-BaseBackbone=LLaDA-8B, Decoding method=Base, Few-shot examples=0, Unmasking iterations=256, Generation limit=2562026.05 | 32.3 | |
| RoLoRA2026.02 | 29.51 | |
| FlexLoRA2026.02 | 29.3 | |
| Base2026.06 | 29.27 | |
| FedIT2026.02 | 28.77 |