Mathematical Reasoning on Math500 (Throughput and Speedup)
20,417Throughput (tok/s)DFlash
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DFlashModel=Qwen3-4B, Concurrency=32, Inference Engine=SGLang, Backend=FA42026.02 | 20,417 | 2.9 | 8.01 | |
| DFlashModel=Qwen3-8B, Concurrency=32, Inference Engine=SGLang, Backend=FA42026.02 | 16,076 | 2.8 | 8.01 | |
| DFlashModel=Qwen3-4B, Concurrency=16, Inference Engine=SGLang, Backend=FA42026.02 | 14,477 | 3.5 | 8.01 | |
| DFlashModel=Qwen3-8B, Concurrency=16, Inference Engine=SGLang, Backend=FA42026.02 | 12,268 | 3.9 | 8.01 | |
| DFlashModel=Qwen3-4B, Concurrency=8, Inference Engine=SGLang, Backend=FA42026.02 | 9,066 | 4.1 | 8.01 | |
| DFlashModel=Qwen3-8B, Concurrency=8, Inference Engine=SGLang, Backend=FA42026.02 | 7,485 | 4.5 | 8.01 | |
| BaselineModel=Qwen3-4B, Concurrency=32, Inference Engine=SGLang, Backend=FA42026.02 | 7,136 | — | — | |
| BaselineModel=Qwen3-8B, Concurrency=32, Inference Engine=SGLang, Backend=FA42026.02 | 5,694 | — | — | |
| DFlashModel=Qwen3-4B, Concurrency=4, Inference Engine=SGLang, Backend=FA42026.02 | 4,943 | 4.3 | 8.01 | |
| BaselineModel=Qwen3-4B, Concurrency=16, Inference Engine=SGLang, Backend=FA42026.02 | 4,100 | — | — | |
| DFlashModel=Qwen3-8B, Concurrency=4, Inference Engine=SGLang, Backend=FA42026.02 | 3,884 | 4.5 | 8.01 | |
| BaselineModel=Qwen3-8B, Concurrency=16, Inference Engine=SGLang, Backend=FA42026.02 | 3,133 | — | — | |
| BaselineModel=Qwen3-4B, Concurrency=8, Inference Engine=SGLang, Backend=FA42026.02 | 2,201 | — | — | |
| BaselineModel=Qwen3-8B, Concurrency=8, Inference Engine=SGLang, Backend=FA42026.02 | 1,666 | — | — | |
| DFlashModel=Qwen3-4B, Concurrency=1, Inference Engine=SGLang, Backend=FA42026.02 | 1,531 | 4.8 | 8.01 | |
| DFlashModel=Qwen3-8B, Concurrency=1, Inference Engine=SGLang, Backend=FA42026.02 | 1,175 | 5.1 | 8.01 | |
| BaselineModel=Qwen3-4B, Concurrency=4, Inference Engine=SGLang, Backend=FA42026.02 | 1,145 | — | — | |
| BaselineModel=Qwen3-8B, Concurrency=4, Inference Engine=SGLang, Backend=FA42026.02 | 861 | — | — | |
| BaselineModel=Qwen3-4B, Concurrency=1, Inference Engine=SGLang, Backend=FA42026.02 | 316 | — | — | |
| BaselineModel=Qwen3-8B, Concurrency=1, Inference Engine=SGLang, Backend=FA42026.02 | 230 | — | — |