Generation Latency and Statistical Metrics on MT-Bench
8.94Latency (ms/tok)CADDTREE
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CADDTREEBackbone=Qwen3-4B, Temperature=1.02026.06 | 8.94 | 6.05 | — | 235 | — | |
| DDTree-oracleBackbone=Qwen3-4B, Temperature=1.02026.06 | 9.02 | 5.84 | 256 | — | — | |
| CADDTREEBackbone=Qwen3-8B, Temperature=1.02026.06 | 9.47 | 5.77 | — | 169 | — | |
| DDTree-oracleBackbone=Qwen3-8B, Temperature=1.02026.06 | 9.68 | 5.54 | 128 | — | — | |
| DFlashBackbone=Qwen3-4B, Temperature=1.02026.06 | 13.53 | 4.01 | — | — | — | |
| DFlashBackbone=Qwen3-8B, Temperature=1.02026.06 | 14.46 | 3.78 | — | — | — | |
| ARBackbone=Qwen3-8B, Temperature=1.02026.06 | 30.92 | — | — | — | — | |
| ARBackbone=Qwen3-4B, Temperature=1.02026.06 | 31.09 | — | — | — | — |