LLM Serving on MATH-plus (Average latency)
107.1Average Latency (ms/token)Oracle SJF
Evaluation Results
| Method | Links | |
|---|---|---|
| Oracle SJFModel=DeepSeek-R12025.09 | 107.1 | |
| PARSModel=DeepSeek-R12025.09 | 115.65 | |
| Listwise SJFModel=DeepSeek-R12025.09 | 123.69 | |
| Pointwise SJFModel=DeepSeek-R12025.09 | 132.91 | |
| Cross-Model PARSModel=DeepSeek-R12025.09 | 134.11 | |
| Oracle SJFModel=Llama2025.09 | 140.78 | |
| PARSModel=Llama2025.09 | 154.25 | |
| Cross-Model PARSModel=Llama2025.09 | 164.8 | |
| Listwise SJFModel=Llama2025.09 | 165.87 | |
| Pointwise SJFModel=Llama2025.09 | 167.76 | |
| FCFSModel=DeepSeek-R12025.09 | 289.41 | |
| FCFSModel=Llama2025.09 | 298.35 |