Code Generation on HumanEval+ (Acc, TPF)
80.49AccuracyMBD-LLaDA2-Mini
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MBD-LLaDA2-MiniBackbone=LLaDA2-Mini, bufsz=2, blksz=322026.06 | 80.49 | 7.85 | |
| SingleBD (Native)Backbone=LLaDA2-Mini, bufsz=2, blksz=322026.06 | 78.05 | 5.53 | |
| SingleBD (Native)Backbone=LLaDA2-Mini-DMax, bufsz=2, blksz=322026.06 | 77.44 | 7.44 | |
| MBD-LLaDA2-Mini-DMaxBackbone=LLaDA2-Mini-DMax, bufsz=2, blksz=322026.06 | 77.44 | 10.78 | |
| MultiBD (training-free)Backbone=LLaDA2-Mini-DMax, bufsz=2, blksz=322026.06 | 76.83 | 10.96 | |
| MultiBD (training-free)Backbone=LLaDA2-Mini, bufsz=2, blksz=322026.06 | 75.61 | 7.37 | |
| MultiBD (training-free)Backbone=SDAR-8B-Chat-b4, bufsz=4, blksz=42026.06 | 74.39 | 1.47 | |
| SingleBD (Native)Backbone=SDAR-8B-Chat-b4, bufsz=4, blksz=42026.06 | 73.7 | 1.07 | |
| MBD-SDAR-8B-Chat-b4Backbone=SDAR-8B-Chat-b4, bufsz=4, blksz=42026.06 | 72.56 | 2.24 | |
| SingleBD (Native)Backbone=SDAR-8B-Chat-b32, bufsz=4, blksz=322026.06 | 67.68 | 2 | |
| MultiBD (training-free)Backbone=SDAR-8B-Chat-b32, bufsz=4, blksz=322026.06 | 65.85 | 2.24 | |
| MBD-SDAR-8B-Chat-b32Backbone=SDAR-8B-Chat-b32, bufsz=4, blksz=322026.06 | 62.8 | 4.82 |