Mathematical Reasoning on GSM8K (Accuracy, Tokens, Latency)
84.3Accuracy (%)DySCo
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DySCon=6, k=2, R=32026.06 | 84.3 | 4.5 | 9.8 | |
| Dense MADn=6, k=2, R=32026.06 | 82 | 14.8 | 18.6 | |
| Trust-Only Sparsen=6, k=2, R=32026.06 | 81.8 | 4.2 | 9.6 | |
| Random-kn=6, k=2, R=32026.06 | 80.5 | 4.2 | 9.5 | |
| Static Ringn=6, k=2, R=32026.06 | 80.1 | 4.2 | 9.4 | |
| Self-Consistencyn=6, k=2, R=32026.06 | 78.5 | 2.5 | 4.5 | |
| Single-Agent CoTn=6, k=2, R=32026.06 | 72.3 | 0.4 | 3.2 |