imagining syntax experiment results

← msize

n-embd_032

done · peak unseen_mismatch 48.7

Configuration

n_layer
2
n_head
4
n_embd
32
base_seed
42

Figures

accuracy_vs_alpha.png
accuracy_vs_alpha.png

Results

paramseen_matchseen_mismatchunseen_matchunseen_mismatch
0.0 72.1 ± 22.829.3 ± 23.470.9 ± 23.830.3 ± 23.8
0.1 52.8 ± 5.047.9 ± 3.352.0 ± 4.947.9 ± 5.1
0.2 65.1 ± 20.237.5 ± 17.664.0 ± 21.138.4 ± 18.2
0.3 67.9 ± 21.833.7 ± 20.868.1 ± 21.534.6 ± 21.2
0.4 60.9 ± 19.738.8 ± 19.460.7 ± 19.739.3 ± 19.6
0.5 72.1 ± 22.731.2 ± 20.271.8 ± 22.932.0 ± 20.7
0.6 68.9 ± 18.935.0 ± 17.167.8 ± 19.435.5 ± 17.4
0.7 66.2 ± 19.236.5 ± 17.464.1 ± 20.237.3 ± 17.6
0.8 80.3 ± 24.219.8 ± 24.180.5 ± 23.820.2 ± 24.6
0.9 57.5 ± 12.144.6 ± 9.955.5 ± 13.744.4 ± 10.1
1.0 65.6 ± 22.634.4 ± 22.565.7 ± 22.534.1 ± 22.3
1.1 66.0 ± 20.737.5 ± 17.965.2 ± 21.338.1 ± 17.9
1.2 72.9 ± 22.728.5 ± 22.472.3 ± 23.428.5 ± 22.1
1.3 61.2 ± 17.540.1 ± 16.360.6 ± 17.339.7 ± 16.4
1.4 72.6 ± 22.728.3 ± 22.973.0 ± 22.427.6 ± 22.4
1.5 56.0 ± 15.044.5 ± 15.056.4 ± 14.643.8 ± 14.9
1.6 67.6 ± 21.733.3 ± 22.066.5 ± 22.434.2 ± 22.5
1.7 65.0 ± 19.435.7 ± 18.565.2 ± 19.236.8 ± 19.0
1.8 52.2 ± 2.548.2 ± 1.352.7 ± 2.448.7 ± 2.3
1.9 75.7 ± 22.430.7 ± 20.074.6 ± 22.830.6 ± 18.9
2.0 56.6 ± 14.745.0 ± 13.555.9 ± 14.844.3 ± 12.9
2.1 64.9 ± 18.437.7 ± 16.364.8 ± 18.439.1 ± 16.1
2.2 76.0 ± 23.326.8 ± 21.875.4 ± 23.927.2 ± 21.7
2.3 69.0 ± 20.632.2 ± 21.068.9 ± 20.733.1 ± 21.5
2.4 60.6 ± 14.542.3 ± 14.559.9 ± 14.741.8 ± 14.4
2.5 60.9 ± 19.639.3 ± 18.460.2 ± 19.940.8 ± 19.4
2.6 56.9 ± 14.643.4 ± 14.657.3 ± 14.444.1 ± 14.9
2.7 62.8 ± 18.738.4 ± 19.162.5 ± 18.838.9 ± 19.4
2.8 68.3 ± 22.334.3 ± 20.667.7 ± 22.434.3 ± 20.4
2.9 69.0 ± 20.633.3 ± 20.668.2 ± 21.333.4 ± 20.4
3.0 57.4 ± 13.945.0 ± 9.656.8 ± 14.045.8 ± 10.4

Downloads