Language understanding on SQuAD 2.0
91.71ScoreCreditDecoding
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CreditDecodingModel=LLaDA-8B-Instruct, Gen Length=256, Block Size=64, Early Stop=true2025.10 | 91.71 | 16.84 | |
| BaselineModel=LLaDA-8B-Instruct, Gen Length=256, Block Size=64, Early Stop=true2025.10 | 91.43 | 1 | |
| Fast-dLLMModel=LLaDA-8B-Instruct, Gen Length=256, Block Size=64, Early Stop=true2025.10 | 91.43 | 13.55 | |
| CreditDecodingModel=LLaDA-MoE-Instruct, Gen Length=256, Block Size=64, Early Stop=true2025.10 | 87.27 | 9.64 | |
| BaselineModel=LLaDA-MoE-Instruct, Gen Length=256, Block Size=64, Early Stop=true2025.10 | 86.88 | 1 | |
| Fast-dLLMModel=LLaDA-MoE-Instruct, Gen Length=256, Block Size=64, Early Stop=true2025.10 | 86.88 | 7.09 |