imagining syntax experiment results

← msize

n-embd_512

done · peak unseen_mismatch 76.2

Configuration

n_layer
2
n_head
4
n_embd
512
base_seed
42

Figures

accuracy_vs_alpha.png
accuracy_vs_alpha.png

Results

paramseen_matchseen_mismatchunseen_matchunseen_mismatch
0.0 61.6 ± 19.258.3 ± 18.559.4 ± 20.458.7 ± 18.2
0.1 79.3 ± 24.047.4 ± 26.979.3 ± 24.146.5 ± 26.8
0.2 60.4 ± 19.947.8 ± 18.860.3 ± 19.948.5 ± 19.4
0.3 64.7 ± 23.140.9 ± 18.265.2 ± 22.941.1 ± 18.0
0.4 84.9 ± 23.044.9 ± 35.885.3 ± 22.545.3 ± 35.7
0.5 70.0 ± 24.445.4 ± 14.970.4 ± 24.045.5 ± 14.4
0.6 70.6 ± 24.049.2 ± 29.269.7 ± 24.848.4 ± 29.1
0.7 85.1 ± 22.838.4 ± 25.285.0 ± 23.038.9 ± 25.8
0.8 80.2 ± 24.331.0 ± 25.180.0 ± 24.530.5 ± 24.6
0.9 74.5 ± 24.054.5 ± 24.574.5 ± 24.354.5 ± 25.1
1.0 69.5 ± 25.039.6 ± 29.269.7 ± 24.838.6 ± 28.8
1.1 65.3 ± 22.750.1 ± 22.464.7 ± 23.151.1 ± 22.4
1.2 64.7 ± 23.240.2 ± 20.164.8 ± 23.040.0 ± 20.0
1.3 75.0 ± 25.059.5 ± 29.474.6 ± 25.358.6 ± 28.8
1.4 70.7 ± 24.049.0 ± 30.270.3 ± 24.349.7 ± 30.3
1.5 69.4 ± 25.148.6 ± 30.970.0 ± 24.449.3 ± 30.1
1.6 70.0 ± 24.358.3 ± 16.669.2 ± 23.657.5 ± 14.8
1.7 75.1 ± 24.662.4 ± 30.774.2 ± 22.758.1 ± 26.4
1.8 76.9 ± 21.869.6 ± 23.671.4 ± 22.962.2 ± 23.4
1.9 82.8 ± 22.361.9 ± 36.778.6 ± 21.956.4 ± 32.8
2.0 80.4 ± 23.564.8 ± 31.276.0 ± 21.657.5 ± 29.0
2.1 90.1 ± 18.760.6 ± 38.786.0 ± 18.956.4 ± 34.1
2.2 89.5 ± 18.788.4 ± 19.979.9 ± 14.776.2 ± 14.2
2.3 70.1 ± 24.269.0 ± 21.867.6 ± 22.159.8 ± 14.3
2.4 89.6 ± 19.284.6 ± 21.781.2 ± 16.775.1 ± 17.1
2.5 81.1 ± 23.465.5 ± 31.274.5 ± 20.857.1 ± 24.8
2.6 78.7 ± 24.078.8 ± 23.768.4 ± 17.266.8 ± 14.6
2.7 83.8 ± 22.683.6 ± 22.269.6 ± 15.766.2 ± 13.4
2.8 89.5 ± 17.368.8 ± 38.678.4 ± 19.655.1 ± 31.0
2.9 99.7 ± 0.396.7 ± 8.083.0 ± 10.774.7 ± 14.9
3.0 84.9 ± 20.083.0 ± 20.965.9 ± 11.462.7 ± 8.4

Downloads