Mathematical Reasoning on GSM8K (Accuracy & Efficiency)
84.6AccuracyDAEDAL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DAEDALInitial Length=642026.01 | 84.6 | 246 | 331 | 74.5 | 1,090 | |
| ρ-EOSInitial Length=64, Threshold Configuration=Asym2026.01 | 84.2 | 252.5 | 361 | 70 | 823 | |
| Fixed-Length DenoisingSequence Length=10242026.01 | 83.9 | 282.9 | 1,024 | 27.6 | 8,238 | |
| Fixed-Length DenoisingSequence Length=5122026.01 | 83.6 | 286.8 | 512 | 56 | 2,458 | |
| Fixed-Length DenoisingSequence Length=20482026.01 | 82.7 | 297.3 | 2,048 | 14.5 | 30,305 | |
| ρ-EOSInitial Length=64, Threshold Configuration=Sym2026.01 | 81.8 | 198.4 | 231.8 | 85.6 | 645 | |
| Fixed-Length DenoisingSequence Length=2562026.01 | 77.8 | 233.7 | 256 | 91.3 | 860 | |
| Fixed-Length DenoisingSequence Length=1282026.01 | 68 | 124.1 | 128 | 97 | 360 | |
| Fixed-Length DenoisingSequence Length=642026.01 | 47.9 | 62.1 | 64 | 97.1 | 174 |