Out-of-Distribution Generalization on AIME '25
26.5Saving ScoreTTT no-QK
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TTT no-QKLabeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 26.5 | 5.6 | |
| TTT QK (dh=128)Labeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 25.8 | 0 | |
| TTT no-QKLabeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 16.6 | 6.7 | |
| Static ProbeLabeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 13.9 | 0 | |
| TTT QK (dh=128)Labeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 13.9 | 0 | |
| Static ProbeLabeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 10.1 | 0 |