Mathematical Reasoning on GSM8k (Accuracy, Throughput, Efficiency)
79.17AccuracydKV-Cache
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| dKV-CacheAcceleration Method=dKV-Cache, Parallel Decoding=false2026.05 | 79.17 | 15.14 | 1.82 | |
| dLLM-CacheAcceleration Method=dLLM-Cache, Parallel Decoding=false2026.05 | 78.78 | 25.28 | 3 | |
| Fast-dLLMAcceleration Method=Fast-dLLM, Parallel Decoding=false2026.05 | 78.44 | 22.74 | 2.73 | |
| Dynamic-dLLMAcceleration Method=Dynamic-dLLM, Parallel Decoding=false2026.05 | 78.24 | 27.21 | 3.27 | |
| LLaDA-8B-InstructParallel Decoding=false2026.05 | 78.21 | 8.32 | 1 | |
| Dynamic-dLLMAcceleration Method=Dynamic-dLLM, Parallel Decoding=true2026.05 | 78.01 | 37.29 | 4.48 | |
| Fast-dLLMAcceleration Method=Fast-dLLM, Parallel Decoding=true2026.05 | 76.62 | 31.36 | 3.77 |