Mol-BERT: An Effective Molecular Representation with BERT for Molecular Property Prediction

Table 2

The fine-tuning hyperparameters.

Parameter	Value/range
Learning rate	1-5∼1-3
Batch size	8
Epoch	100
Optimizer	Adam
Embedding dimension	300
Size of dictionary	13,325
Number of attention head	6
Layers of fully connected neural network	6