Visualize:  Figure 1 & 4 .
Description:  Attention in Transformer; Attention head view of GPT-2;

Visualize:  Figure 2.
Description: model view of GPT-2;

Visualize:  Figure 3.
Description: neuron view ;

Visualize:  Figure 5 & 6 & 10-12.
Description: Each heatmap shows the proportion of total attention

Visualize:  Table 1-3;
Description: word-level attention;