Fig. 11
From: From theory to experiment: transformer-based generation enables rapid discovery of novel reactions
The general flow of reaction formation. We randomly selected a symbol C to start the generation, a tensor of each character of vocab then imports into the model and the probability distribution P as an outcome. According to the distribution P, the model randomly selected the next symbol. Above steps will be cyclically execute until the appearance of EOF. Finally, the SMILES string as the result to output