Code Generation on HumanEval (Performance Metrics)
51.96Tokens Per Second (TPS)LLaDA Base + dLLM-Cache
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LLaDA Base + dLLM-CacheModel Size=8B, Caching=dLLM-Cache2025.05 | 51.96 | 2.6 | 2.04 | 2.96 | 32.31 | |
| LLaDA Instruct + dLLM-CacheModel Size=8B, Caching=dLLM-Cache2025.05 | 44.77 | 4.24 | 2.05 | 5.41 | 39.02 | |
| LLaDA BaseModel Size=8B, Caching=None2025.05 | 19.98 | 1 | 6.03 | 1 | 32.92 | |
| LLaDA InstructModel Size=8B, Caching=None2025.05 | 10.57 | 1 | 11.1 | 1 | 38.71 |