imagining syntax

← msize_long

n-embd_512

done · peak unseen_mismatch 77.1

Continued from

runs/20260704_225157_msize/n-embd_512

Figures

accuracy_at_iter_01200.png
accuracy_at_iter_01200.png
accuracy_at_iter_01500.png
accuracy_at_iter_01500.png
accuracy_at_iter_02400.png
accuracy_at_iter_02400.png
accuracy_at_iter_03600.png
accuracy_at_iter_03600.png
accuracy_at_iter_04800.png
accuracy_at_iter_04800.png
waypoint_evolution.png
waypoint_evolution.png

Results

at final waypoint (t = 4800), mean ± sd over seeds

seen_* generated at each grid point's own training α · unseen_* = held-out pairings, uniform (α-independent) · 1000 pairs per condition

αseen_matchseen_mismatchunseen_matchunseen_mismatch
0.9 95.2 ± 14.358.0 ± 28.893.3 ± 14.752.3 ± 25.9
1.0 94.8 ± 15.649.5 ± 42.094.5 ± 14.643.9 ± 37.3
1.1 94.7 ± 15.643.2 ± 43.594.3 ± 15.139.7 ± 39.9
1.2 94.8 ± 15.266.7 ± 35.693.7 ± 15.158.4 ± 32.5
1.3 99.9 ± 0.267.8 ± 41.697.0 ± 2.758.4 ± 35.9
1.4 99.8 ± 0.264.4 ± 39.496.2 ± 3.252.9 ± 32.9
1.5 99.8 ± 0.261.4 ± 44.295.5 ± 4.649.5 ± 35.5
1.6 99.7 ± 0.383.5 ± 26.793.6 ± 4.670.2 ± 25.1
1.7 99.7 ± 0.378.7 ± 39.493.4 ± 5.466.8 ± 34.6
1.8 94.9 ± 14.481.3 ± 31.086.9 ± 12.965.0 ± 25.2
1.9 95.0 ± 14.092.5 ± 15.083.9 ± 12.477.1 ± 11.0
2.0 94.8 ± 14.980.4 ± 31.487.3 ± 13.168.3 ± 25.3
2.1 99.8 ± 0.298.1 ± 1.987.6 ± 3.975.0 ± 9.4
2.5 99.8 ± 0.289.1 ± 29.782.5 ± 6.966.9 ± 22.9
3.0 99.8 ± 0.189.7 ± 29.980.2 ± 7.766.7 ± 22.6

Downloads