imagining syntax experiment results

← depth

n-layer_1

done · peak unseen_mismatch 90.7

Configuration

n_layer
1
n_head
4
n_embd
256
base_seed
42

Figures

accuracy_vs_alpha.png
accuracy_vs_alpha.png

Results

paramseen_matchseen_mismatchunseen_matchunseen_mismatch
0.0 100.0 ± 0.017.3 ± 15.5100.0 ± 0.018.1 ± 15.1
0.1 100.0 ± 0.015.6 ± 20.0100.0 ± 0.014.8 ± 19.6
0.2 100.0 ± 0.03.0 ± 4.3100.0 ± 0.04.0 ± 4.7
0.3 100.0 ± 0.024.9 ± 29.4100.0 ± 0.024.1 ± 29.4
0.4 100.0 ± 0.017.5 ± 17.3100.0 ± 0.018.2 ± 16.9
0.5 100.0 ± 0.016.7 ± 21.1100.0 ± 0.016.4 ± 21.3
0.6 100.0 ± 0.012.3 ± 14.0100.0 ± 0.014.3 ± 15.1
0.7 100.0 ± 0.019.6 ± 17.7100.0 ± 0.019.1 ± 17.5
0.8 100.0 ± 0.065.6 ± 24.9100.0 ± 0.064.7 ± 24.8
0.9 100.0 ± 0.089.3 ± 16.3100.0 ± 0.088.0 ± 16.1
1.0 100.0 ± 0.088.3 ± 11.099.6 ± 0.883.4 ± 12.7
1.1 99.9 ± 0.195.3 ± 6.099.4 ± 0.689.2 ± 11.9
1.2 99.9 ± 0.293.8 ± 8.898.8 ± 1.086.0 ± 15.5
1.3 99.9 ± 0.198.2 ± 1.196.9 ± 1.287.9 ± 4.9
1.4 99.9 ± 0.298.0 ± 1.295.8 ± 2.687.3 ± 7.2
1.5 99.8 ± 0.198.9 ± 1.496.0 ± 1.388.2 ± 8.1
1.6 99.8 ± 0.299.2 ± 0.695.1 ± 1.889.6 ± 5.4
1.7 99.8 ± 0.299.2 ± 0.494.3 ± 1.987.8 ± 4.6
1.8 99.7 ± 0.199.5 ± 0.393.5 ± 2.290.7 ± 3.3
1.9 99.7 ± 0.299.1 ± 0.591.2 ± 3.285.1 ± 3.6
2.0 99.7 ± 0.299.2 ± 0.491.4 ± 2.584.1 ± 6.1
2.1 99.8 ± 0.199.4 ± 0.391.9 ± 2.487.7 ± 4.3
2.2 99.8 ± 0.299.3 ± 0.590.2 ± 2.984.0 ± 7.7
2.3 99.7 ± 0.299.5 ± 0.388.6 ± 2.985.0 ± 2.9
2.4 99.8 ± 0.199.6 ± 0.288.7 ± 2.185.5 ± 3.3
2.5 99.8 ± 0.199.6 ± 0.387.0 ± 3.483.3 ± 4.3
2.6 99.7 ± 0.199.6 ± 0.287.7 ± 3.784.4 ± 5.2
2.7 99.8 ± 0.199.7 ± 0.285.3 ± 2.681.9 ± 3.1
2.8 99.9 ± 0.299.8 ± 0.184.9 ± 2.582.7 ± 2.8
2.9 99.8 ± 0.199.7 ± 0.185.0 ± 3.282.5 ± 3.7
3.0 99.8 ± 0.299.8 ± 0.183.9 ± 3.081.7 ± 3.0

Downloads