imagining syntax

← bimodality_grid_v1

alpha_grid

done · peak unseen_mismatch 89.5

Configuration

n_layer
2
n_head
4
n_embd
256

Results

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
0.5 100.0 ± 0.016.4 ± 22.6100.0 ± 0.016.5 ± 22.6
0.8 100.0 ± 0.038.2 ± 27.7100.0 ± 0.038.3 ± 27.6
1.0 100.0 ± 0.058.0 ± 35.8100.0 ± 0.256.9 ± 35.4
1.2 99.9 ± 0.284.0 ± 26.299.8 ± 0.581.8 ± 25.5
1.4 99.8 ± 0.296.7 ± 3.298.5 ± 1.689.5 ± 7.4
1.6 99.5 ± 0.495.4 ± 13.196.1 ± 2.586.1 ± 13.4
2.0 99.3 ± 0.398.4 ± 0.989.2 ± 3.681.1 ± 5.5
2.5 99.4 ± 0.399.3 ± 0.484.0 ± 3.675.8 ± 6.0

Downloads