Language Understanding on MMLU-Pro (Accuracy, Throughput, and Speedup)
49.3AccuracydLLM-Cache
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| dLLM-CacheModel=Dream 7B, nu=12026.03 | 49.3 | 27.19 | 2.16 | |
| OriginalModel=Dream 7B, nu=12026.03 | 47.94 | 12.6 | 1 | |
| Fast-PrefixModel=Dream 7B, nu=12026.03 | 47.74 | 78.23 | 6.21 | |
| DyLLMModel=Dream 7B, Threshold (τ)=0.9975, nu=12026.03 | 47.45 | 83.1 | 6.6 | |
| DyLLMModel=Dream 7B, Threshold (τ)=0.995, nu=12026.03 | 47.21 | 87.98 | 6.98 | |
| Fast-DualModel=Dream 7B, nu=12026.03 | 46.73 | 128.52 | 10.2 | |
| dLLM-CacheModel=LLaDA 8B, nu=12026.03 | 37.9 | 31.56 | 2.75 | |
| OriginalModel=LLaDA 8B, nu=12026.03 | 37.31 | 11.46 | 1 | |
| DyLLMModel=LLaDA 8B, Threshold (τ)=0.995, nu=12026.03 | 37.06 | 60.16 | 5.25 | |
| Fast-PrefixModel=LLaDA 8B, nu=12026.03 | 36.7 | 46.45 | 4.05 | |
| DyLLMModel=LLaDA 8B, Threshold (τ)=0.99, nu=12026.03 | 36.42 | 66.44 | 5.8 | |
| Fast-DualModel=LLaDA 8B, nu=12026.03 | 36.18 | 64.62 | 5.64 |