Latent Reasoning on GSM8K (test)
1.84SG ScoreGTS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GTSBackbone=COCONUT2026.02 | 1.84 | 0.82 | 0.11 | |
| GTSBackbone=CODI2026.02 | 1.15 | 0.7 | 0.1 | |
| Dropoutp=0.1, Backbone=COCONUT2026.02 | 1.09 | 0.61 | 0.05 | |
| Dropoutp=0.1, Backbone=CODI2026.02 | 0.53 | 0.31 | 0.01 | |
| Dropoutp=0.5, Backbone=CODI2026.02 | 0.47 | 0.39 | 0.06 | |
| StandardGBackbone=COCONUT2026.02 | 0.34 | 0.51 | 0.2 | |
| StandardGBackbone=CODI2026.02 | 0.11 | 0.38 | 0.15 | |
| Dropoutp=0.5, Backbone=COCONUT2026.02 | -0.87 | 0.4 | 0.28 |