imagining syntax experiment results

← msize

n-embd_064

done · peak unseen_mismatch 71.6

Configuration

n_layer
2
n_head
4
n_embd
64
base_seed
42

Figures

accuracy_vs_alpha.png
accuracy_vs_alpha.png

Results

paramseen_matchseen_mismatchunseen_matchunseen_mismatch
0.0 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
0.1 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
0.2 100.0 ± 0.00.2 ± 0.5100.0 ± 0.00.2 ± 0.5
0.3 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
0.4 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
0.5 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
0.6 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
0.7 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
0.8 100.0 ± 0.00.1 ± 0.5100.0 ± 0.00.2 ± 0.6
0.9 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
1.0 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
1.1 100.0 ± 0.00.2 ± 0.6100.0 ± 0.00.3 ± 0.9
1.2 100.0 ± 0.00.0 ± 0.1100.0 ± 0.00.1 ± 0.1
1.3 100.0 ± 0.03.0 ± 8.4100.0 ± 0.02.9 ± 8.1
1.4 100.0 ± 0.00.0 ± 0.0100.0 ± 0.00.0 ± 0.0
1.5 100.0 ± 0.00.5 ± 1.6100.0 ± 0.00.5 ± 1.6
1.6 100.0 ± 0.00.2 ± 0.3100.0 ± 0.00.4 ± 0.6
1.7 100.0 ± 0.05.9 ± 13.6100.0 ± 0.06.1 ± 13.0
1.8 100.0 ± 0.05.3 ± 15.5100.0 ± 0.05.0 ± 14.7
1.9 100.0 ± 0.017.6 ± 22.9100.0 ± 0.017.9 ± 21.9
2.0 100.0 ± 0.013.7 ± 19.5100.0 ± 0.013.8 ± 19.5
2.1 100.0 ± 0.011.2 ± 20.4100.0 ± 0.011.7 ± 20.5
2.2 100.0 ± 0.026.7 ± 28.9100.0 ± 0.025.8 ± 26.9
2.3 100.0 ± 0.030.6 ± 36.099.6 ± 0.926.1 ± 29.1
2.4 100.0 ± 0.164.3 ± 36.498.9 ± 2.557.6 ± 32.0
2.5 100.0 ± 0.177.5 ± 18.698.4 ± 3.570.2 ± 16.5
2.6 99.8 ± 0.279.0 ± 19.098.2 ± 2.471.6 ± 13.8
2.7 99.9 ± 0.158.3 ± 33.199.5 ± 1.355.9 ± 30.3
2.8 100.0 ± 0.168.4 ± 34.399.6 ± 0.659.5 ± 32.7
2.9 100.0 ± 0.153.8 ± 36.298.6 ± 2.647.9 ± 29.7
3.0 99.9 ± 0.169.6 ± 30.897.8 ± 3.159.8 ± 25.2

Downloads