Multi-task Language Understanding on MMLU (Throughput and Efficiency)
38.01Tokens Per Second (TPS)Dream Instruct + dLLM-Cache
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Dream Instruct + dLLM-CacheConfiguration=+ dLLM-Cache2025.05 | 38.01 | 4.5 | 2.42 | 6.1 | 73.42 | |
| Dream Base + dLLM-CacheConfiguration=+ dLLM-Cache2025.05 | 31.07 | 3.41 | 3.27 | 4.2 | 73.2 | |
| Dream BaseConfiguration=Dream Base2025.05 | 9.1 | 1 | 13.73 | 1 | 73.49 | |
| Dream InstructConfiguration=Dream Instruct2025.05 | 8.45 | 1 | 14.75 | 1 | 73.4 |