Multi-task Evaluation on Average (GSM8K-CoT, MATH, MBPP, HumanEval)
51.6AccuracyTAD-Q
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TAD-QBackbone=LLaDA, Configuration=Quality2026.05 | 51.6 | 5.08 | 225.2 | |
| TAD-SBackbone=LLaDA, Configuration=Speed2026.05 | 49.9 | 5.76 | 257.1 | |
| LLaDABackbone=LLaDA-Instruct2026.05 | 46.2 | 1 | 46.2 | |
| d3LLMBackbone=LLaDA2026.05 | 45.9 | 6.25 | 206.1 | |
| Fast-dLLMBackbone=LLaDA2026.05 | 45.5 | 2.36 | 84.7 | |
| dParallelBackbone=LLaDA2026.05 | 45.5 | 3.9 | 128.6 | |
| D2FBackbone=LLaDA2026.05 | 44.1 | 2.47 | 87.9 |