imagining syntax experiment results

← msize

n-embd_016

done · peak unseen_mismatch 51.1

Configuration

n_layer
2
n_head
4
n_embd
16
base_seed
42

Figures

accuracy_vs_alpha.png
accuracy_vs_alpha.png

Results

paramseen_matchseen_mismatchunseen_matchunseen_mismatch
0.0 51.0 ± 1.849.6 ± 1.550.3 ± 0.649.6 ± 1.7
0.1 49.9 ± 1.549.9 ± 1.449.1 ± 1.350.0 ± 2.0
0.2 50.9 ± 1.749.6 ± 1.349.4 ± 0.949.3 ± 1.6
0.3 50.5 ± 1.549.1 ± 1.349.8 ± 2.350.5 ± 1.7
0.4 50.5 ± 1.649.5 ± 1.550.0 ± 1.750.3 ± 1.7
0.5 50.9 ± 1.649.4 ± 1.250.8 ± 1.550.4 ± 1.8
0.6 49.7 ± 1.150.3 ± 1.849.8 ± 1.950.3 ± 1.6
0.7 50.3 ± 1.349.7 ± 1.649.3 ± 1.449.2 ± 1.2
0.8 49.9 ± 1.849.4 ± 1.649.8 ± 2.250.2 ± 1.0
0.9 50.6 ± 0.948.8 ± 1.548.7 ± 1.649.7 ± 1.3
1.0 50.2 ± 1.449.5 ± 1.350.5 ± 1.949.7 ± 1.1
1.1 50.6 ± 1.349.3 ± 1.849.4 ± 1.850.0 ± 1.5
1.2 51.0 ± 1.449.6 ± 1.449.9 ± 1.150.1 ± 0.9
1.3 49.8 ± 1.950.4 ± 1.349.3 ± 1.249.4 ± 1.5
1.4 50.6 ± 1.849.8 ± 1.050.2 ± 1.449.4 ± 1.2
1.5 50.0 ± 1.949.6 ± 1.150.9 ± 1.549.4 ± 1.2
1.6 51.1 ± 1.649.5 ± 1.549.3 ± 1.250.6 ± 1.6
1.7 49.8 ± 1.549.0 ± 1.450.1 ± 1.449.9 ± 1.6
1.8 50.3 ± 1.549.4 ± 0.950.4 ± 2.050.2 ± 1.9
1.9 50.7 ± 1.050.0 ± 1.449.9 ± 1.449.6 ± 2.0
2.0 51.1 ± 2.149.4 ± 1.749.8 ± 1.749.1 ± 1.9
2.1 49.5 ± 1.149.7 ± 1.250.2 ± 1.450.2 ± 1.2
2.2 50.2 ± 1.850.3 ± 1.649.7 ± 2.049.6 ± 1.7
2.3 50.2 ± 0.949.9 ± 1.550.3 ± 1.750.1 ± 1.6
2.4 50.4 ± 1.649.8 ± 1.449.6 ± 1.149.6 ± 2.2
2.5 50.5 ± 1.349.7 ± 1.449.9 ± 2.251.1 ± 2.0
2.6 49.9 ± 1.350.0 ± 1.650.1 ± 1.250.6 ± 1.4
2.7 50.5 ± 0.849.8 ± 1.650.5 ± 1.349.6 ± 2.1
2.8 50.0 ± 2.149.6 ± 1.349.8 ± 1.749.3 ± 1.0
2.9 50.3 ± 1.249.7 ± 1.549.9 ± 1.250.8 ± 1.1
3.0 51.5 ± 2.049.7 ± 1.550.4 ± 2.050.2 ± 1.3

Downloads