imagining syntax

← anaphor_n60

oneshot

done · peak unseen_mismatch 62.0

Configuration

n_layer
2
n_head
4
n_embd
256
base_seed
42

Figures

accuracy_at_iter_00300.png
accuracy_at_iter_00300.png
accuracy_at_iter_00600.png
accuracy_at_iter_00600.png
accuracy_at_iter_00900.png
accuracy_at_iter_00900.png
accuracy_at_iter_01200.png
accuracy_at_iter_01200.png
waypoint_evolution.png
waypoint_evolution.png

Results

at final waypoint (t = 1200), 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
0.0 100.0 ± 0.058.5 ± 18.8100.0 ± 0.057.9 ± 19.6

Downloads