Code Generation on MBPP (Latency and Budget Metrics)
5.26Time per Token (ms/tok)DDTree-oracle
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DDTree-oracleBackbone=Qwen3-4B, Temperature=0.02026.06 | 5.26 | 8.87 | 256 | |
| CADDTREEBackbone=Qwen3-4B, Temperature=0.02026.06 | 5.31 | 8.91 | 257 | |
| DDTree-oracleBackbone=Qwen3-8B, Temperature=0.02026.06 | 5.53 | 8.42 | 128 | |
| CADDTREEBackbone=Qwen3-8B, Temperature=0.02026.06 | 5.53 | 8.61 | 187 | |
| DFlashBackbone=Qwen3-4B, Temperature=0.02026.06 | 7.41 | 6.14 | — | |
| DFlashBackbone=Qwen3-8B, Temperature=0.02026.06 | 7.75 | 5.93 | — | |
| ARBackbone=Qwen3-4B, Temperature=0.02026.06 | 30.68 | — | — | |
| ARBackbone=Qwen3-8B, Temperature=0.02026.06 | 31.01 | — | — |