Mathematical Reasoning on GSM8K (Accuracy, TPS, Memory Usage)
53.55TPSLLaDA Base
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LLaDA BaseSteps=322025.05 | 53.55 | 22.25 | 16.94 | |
| LLaDA Base + dLLM-Cache + SlowFastSteps=-2025.05 | 49.86 | 67.17 | 17.93 | |
| Llama 3 8BSteps=-2025.05 | 47.73 | 49.05 | 16.06 | |
| LLaDA Base + dLLM-CacheSteps=2562025.05 | 20.64 | 70.66 | 17.93 | |
| LLaDA BaseSteps=2562025.05 | 7.37 | 69.06 | 16.94 |