imagining syntax experiment results

← msize

n-embd_128

done · peak unseen_mismatch 89.2

Configuration

n_layer
2
n_head
4
n_embd
128
base_seed
42

Figures

accuracy_vs_alpha.png
accuracy_vs_alpha.png

Results

paramseen_matchseen_mismatchunseen_matchunseen_mismatch
0.0 100.0 ± 0.011.8 ± 16.2100.0 ± 0.011.5 ± 15.9
0.1 100.0 ± 0.06.1 ± 10.2100.0 ± 0.06.0 ± 10.3
0.2 100.0 ± 0.06.0 ± 14.8100.0 ± 0.06.0 ± 14.8
0.3 100.0 ± 0.06.5 ± 14.2100.0 ± 0.06.7 ± 14.6
0.4 100.0 ± 0.00.1 ± 0.3100.0 ± 0.00.2 ± 0.4
0.5 100.0 ± 0.06.9 ± 15.1100.0 ± 0.07.0 ± 15.0
0.6 100.0 ± 0.012.4 ± 18.7100.0 ± 0.012.4 ± 18.2
0.7 100.0 ± 0.07.8 ± 11.7100.0 ± 0.07.7 ± 11.5
0.8 100.0 ± 0.011.8 ± 14.5100.0 ± 0.012.3 ± 14.8
0.9 100.0 ± 0.020.9 ± 31.8100.0 ± 0.021.2 ± 31.7
1.0 100.0 ± 0.029.3 ± 26.7100.0 ± 0.028.9 ± 26.4
1.1 100.0 ± 0.030.8 ± 29.1100.0 ± 0.031.8 ± 29.1
1.2 100.0 ± 0.052.7 ± 40.5100.0 ± 0.052.4 ± 39.7
1.3 100.0 ± 0.195.1 ± 4.7100.0 ± 0.189.2 ± 5.9
1.4 99.9 ± 0.288.2 ± 15.199.8 ± 0.484.7 ± 14.4
1.5 99.9 ± 0.294.2 ± 9.298.7 ± 1.588.4 ± 9.6
1.6 99.7 ± 0.488.1 ± 23.098.0 ± 2.979.0 ± 19.9
1.7 99.6 ± 0.396.8 ± 1.296.3 ± 3.085.2 ± 6.1
1.8 99.3 ± 0.594.1 ± 3.995.9 ± 3.383.1 ± 2.6
1.9 99.2 ± 0.596.5 ± 3.192.6 ± 3.683.4 ± 3.4
2.0 99.2 ± 0.397.9 ± 0.792.1 ± 3.083.2 ± 5.1
2.1 99.2 ± 0.397.7 ± 0.889.7 ± 2.781.9 ± 3.7
2.2 99.0 ± 0.697.4 ± 1.387.2 ± 2.176.6 ± 4.4
2.3 99.2 ± 0.398.0 ± 0.884.8 ± 2.673.5 ± 6.5
2.4 99.3 ± 0.498.1 ± 0.784.8 ± 5.275.0 ± 4.7
2.5 99.2 ± 0.398.1 ± 0.680.8 ± 4.771.5 ± 5.3
2.6 99.0 ± 0.398.2 ± 0.781.6 ± 4.172.0 ± 6.8
2.7 99.3 ± 0.398.9 ± 0.380.0 ± 4.273.3 ± 5.3
2.8 99.4 ± 0.298.9 ± 0.578.5 ± 2.571.7 ± 2.3
2.9 99.3 ± 0.298.9 ± 0.677.0 ± 5.069.7 ± 4.6
3.0 99.6 ± 0.299.0 ± 0.578.3 ± 2.769.3 ± 1.6

Downloads