imagining syntax experiment results

← depth

n-layer_2

done · peak unseen_mismatch 93.1

Configuration

n_layer
2
n_head
4
n_embd
256
base_seed
42

Figures

accuracy_vs_alpha.png
accuracy_vs_alpha.png

Results

paramseen_matchseen_mismatchunseen_matchunseen_mismatch
0.0 100.0 ± 0.022.0 ± 21.2100.0 ± 0.022.7 ± 21.8
0.1 100.0 ± 0.019.7 ± 23.5100.0 ± 0.019.2 ± 22.9
0.2 100.0 ± 0.022.4 ± 24.0100.0 ± 0.023.4 ± 24.5
0.3 100.0 ± 0.015.2 ± 15.6100.0 ± 0.015.8 ± 16.0
0.4 100.0 ± 0.028.3 ± 20.9100.0 ± 0.027.9 ± 20.5
0.5 100.0 ± 0.018.2 ± 27.4100.0 ± 0.018.6 ± 28.1
0.6 100.0 ± 0.010.3 ± 12.6100.0 ± 0.010.9 ± 12.1
0.7 100.0 ± 0.030.2 ± 30.8100.0 ± 0.030.3 ± 30.6
0.8 100.0 ± 0.041.2 ± 30.7100.0 ± 0.040.4 ± 31.6
0.9 100.0 ± 0.041.8 ± 38.2100.0 ± 0.041.2 ± 38.0
1.0 100.0 ± 0.065.0 ± 26.2100.0 ± 0.064.4 ± 25.9
1.1 100.0 ± 0.182.4 ± 21.6100.0 ± 0.080.3 ± 21.1
1.2 99.9 ± 0.395.6 ± 6.299.7 ± 0.893.1 ± 9.1
1.3 99.9 ± 0.184.4 ± 28.699.1 ± 1.378.7 ± 27.0
1.4 99.7 ± 0.396.7 ± 2.897.5 ± 2.089.7 ± 4.1
1.5 99.5 ± 0.498.6 ± 0.997.0 ± 1.891.6 ± 5.1
1.6 99.3 ± 0.395.8 ± 5.194.3 ± 2.885.0 ± 7.5
1.7 99.4 ± 0.298.0 ± 1.094.7 ± 2.385.3 ± 5.7
1.8 99.4 ± 0.398.4 ± 1.293.1 ± 2.585.9 ± 5.9
1.9 99.3 ± 0.398.5 ± 0.790.3 ± 3.884.0 ± 6.4
2.0 99.4 ± 0.298.7 ± 0.289.8 ± 1.880.2 ± 4.0
2.1 99.4 ± 0.399.2 ± 0.588.1 ± 2.482.6 ± 4.5
2.2 99.4 ± 0.398.9 ± 0.386.3 ± 2.679.5 ± 1.1
2.3 99.5 ± 0.298.8 ± 0.687.6 ± 2.579.7 ± 6.0
2.4 99.6 ± 0.299.2 ± 0.485.7 ± 3.275.9 ± 6.2
2.5 99.7 ± 0.199.2 ± 0.484.9 ± 2.672.1 ± 4.4
2.6 99.6 ± 0.299.2 ± 0.283.5 ± 2.777.8 ± 4.5
2.7 99.7 ± 0.199.3 ± 0.282.6 ± 2.876.4 ± 5.5
2.8 99.6 ± 0.299.5 ± 0.182.2 ± 2.474.1 ± 4.2
2.9 99.6 ± 0.299.5 ± 0.479.7 ± 2.872.3 ± 5.1
3.0 99.8 ± 0.199.6 ± 0.282.6 ± 2.974.0 ± 1.4

Downloads