imagining syntax experiment results

← depth

n-layer_4

done · peak unseen_mismatch 86.0

Configuration

n_layer
4
n_head
4
n_embd
256
base_seed
42

Figures

accuracy_vs_alpha.png
accuracy_vs_alpha.png

Results

paramseen_matchseen_mismatchunseen_matchunseen_mismatch
0.0 89.6 ± 19.615.5 ± 22.089.7 ± 19.415.7 ± 22.8
0.1 93.5 ± 14.511.6 ± 15.293.3 ± 14.811.8 ± 14.8
0.2 95.1 ± 14.711.6 ± 16.295.3 ± 14.211.8 ± 16.9
0.3 100.0 ± 0.08.6 ± 16.1100.0 ± 0.08.4 ± 16.3
0.4 89.9 ± 20.114.8 ± 20.690.5 ± 19.114.3 ± 20.1
0.5 100.0 ± 0.011.7 ± 14.5100.0 ± 0.011.2 ± 13.7
0.6 95.4 ± 13.829.5 ± 18.495.3 ± 14.230.8 ± 19.1
0.7 100.0 ± 0.018.1 ± 20.2100.0 ± 0.017.8 ± 20.6
0.8 90.3 ± 19.426.8 ± 30.089.5 ± 21.026.9 ± 29.9
0.9 100.0 ± 0.013.7 ± 16.4100.0 ± 0.013.9 ± 17.4
1.0 95.0 ± 15.017.5 ± 21.995.1 ± 14.718.5 ± 22.2
1.1 100.0 ± 0.028.4 ± 21.0100.0 ± 0.027.9 ± 20.8
1.2 100.0 ± 0.023.8 ± 27.6100.0 ± 0.023.8 ± 27.0
1.3 100.0 ± 0.037.6 ± 33.4100.0 ± 0.038.8 ± 33.7
1.4 94.9 ± 15.232.7 ± 29.194.9 ± 15.333.0 ± 29.5
1.5 100.0 ± 0.064.0 ± 37.0100.0 ± 0.063.0 ± 36.2
1.6 99.8 ± 0.356.6 ± 39.699.2 ± 1.354.3 ± 37.7
1.7 99.9 ± 0.177.5 ± 21.999.7 ± 0.875.5 ± 20.6
1.8 99.6 ± 0.389.8 ± 21.196.9 ± 2.884.7 ± 20.1
1.9 99.5 ± 0.491.1 ± 21.596.1 ± 3.383.9 ± 21.1
2.0 99.5 ± 0.494.3 ± 5.794.1 ± 4.885.6 ± 7.0
2.1 99.0 ± 1.194.9 ± 6.692.0 ± 6.186.0 ± 5.9
2.2 99.2 ± 0.493.2 ± 13.190.0 ± 4.279.5 ± 12.0
2.3 99.4 ± 0.394.7 ± 10.590.7 ± 4.480.4 ± 8.4
2.4 99.1 ± 0.297.1 ± 2.785.2 ± 3.275.8 ± 6.1
2.5 98.3 ± 3.397.3 ± 3.480.7 ± 8.176.0 ± 9.3
2.6 99.1 ± 0.498.1 ± 0.781.2 ± 3.774.8 ± 2.0
2.7 99.5 ± 0.298.0 ± 2.580.2 ± 4.772.3 ± 7.0
2.8 99.2 ± 0.397.7 ± 1.277.4 ± 3.766.6 ± 6.2
2.9 99.3 ± 0.397.6 ± 2.278.0 ± 6.369.9 ± 12.4
3.0 99.5 ± 0.296.5 ± 2.575.2 ± 2.760.0 ± 9.3

Downloads