Text Generation on Math-500 (Throughput/Speed)
1,094.6Throughput (TPS)JetFlow
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| JetFlowBatch Size=16, Tree Budget=32, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 1,094.6 | 3.81 | |
| JetFlowBatch Size=16, Tree Budget=64, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 995.8 | 3.47 | |
| JetFlowBatch Size=16, Tree Budget=16, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 891.8 | 3.1 | |
| JetFlowBatch Size=8, Tree Budget=64, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 859.3 | 3.49 | |
| JetFlowBatch Size=8, Tree Budget=32, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 839.3 | 3.41 | |
| JetFlowBatch Size=8, Tree Budget=128, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 803.5 | 3.26 | |
| JetFlowBatch Size=16, Tree Budget=128, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 803.1 | 2.8 | |
| JetFlowBatch Size=4, Tree Budget=128, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 742.9 | 3.64 | |
| JetFlowBatch Size=8, Tree Budget=16, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 679.3 | 2.76 | |
| JetFlowBatch Size=4, Tree Budget=64, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 664.2 | 3.26 | |
| JetFlowBatch Size=2, Tree Budget=128, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 621.6 | 3.81 | |
| JetFlowBatch Size=1, Tree Budget=128, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 553.3 | 4.33 | |
| JetFlowBatch Size=4, Tree Budget=32, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 534.2 | 2.62 | |
| JetFlowBatch Size=2, Tree Budget=64, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 516 | 3.16 | |
| JetFlowBatch Size=1, Tree Budget=64, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 447.3 | 3.5 | |
| JetFlowBatch Size=4, Tree Budget=16, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 433.6 | 2.13 | |
| JetFlowBatch Size=2, Tree Budget=32, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 360.3 | 2.21 | |
| JetFlowBatch Size=1, Tree Budget=32, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 312 | 2.44 | |
| ARBatch Size=16, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 287.3 | — | |
| JetFlowBatch Size=2, Tree Budget=16, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 266.2 | 1.63 | |
| ARBatch Size=8, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 246.2 | — | |
| JetFlowBatch Size=1, Tree Budget=16, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 224 | 1.75 | |
| ARBatch Size=4, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 203.8 | — | |
| ARBatch Size=2, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 163.3 | — | |
| ARBatch Size=1, Model=Qwen3-8B, Hardware=single H100 GPU2026.06 | 127.8 | — |