Science Reasoning on GPQA (Accuracy, TPS, Efficiency)
32.25AccuracydKV-Cache
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| dKV-CacheAcceleration Method=dKV-Cache, Parallel Decoding=false2026.05 | 32.25 | 14.44 | 1.9 | |
| Fast-dLLMAcceleration Method=Fast-dLLM, Parallel Decoding=true2026.05 | 32.21 | 27.06 | 3.56 | |
| LLaDA-8B-InstructParallel Decoding=false2026.05 | 32.17 | 7.6 | 1 | |
| Fast-dLLMAcceleration Method=Fast-dLLM, Parallel Decoding=false2026.05 | 32.01 | 22.65 | 2.98 | |
| Dynamic-dLLMAcceleration Method=Dynamic-dLLM, Parallel Decoding=false2026.05 | 31.98 | 25.46 | 3.35 | |
| Dynamic-dLLMAcceleration Method=Dynamic-dLLM, Parallel Decoding=true2026.05 | 31.89 | 31.84 | 4.19 | |
| dLLM-CacheAcceleration Method=dLLM-Cache, Parallel Decoding=false2026.05 | 31.84 | 22.04 | 2.9 |