Out-of-Distribution Generalization on AIME '26
19.8Saving ScoreTTT no-QK
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TTT no-QKLabeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 19.8 | 5 | |
| TTT no-QKLabeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 15.4 | 6.7 | |
| Static ProbeLabeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 14.7 | 5 | |
| Static ProbeLabeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 14.7 | 10 | |
| TTT QK (dh=128)Labeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 13.4 | 5 | |
| TTT QK (dh=128)Labeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 9.2 | 0 |