Out-of-Distribution Generalization on AIME'24
29.5Saving ScoreTTT QK (dh=128)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TTT QK (dh=128)Labeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 29.5 | 10 | |
| TTT no-QKLabeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 29.3 | 15 | |
| TTT QK (dh=128)Labeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 18.5 | 3.3 | |
| Static ProbeLabeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 15.8 | 5 | |
| TTT no-QKLabeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 14.1 | 3.3 | |
| Static ProbeLabeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 11.8 | 3.3 |