sentiment

• All training is under the base BERT model which has 110M parameters.
• The hyperparameters of training is in the log.txt.
• The learning rate will not affect training too much.
• The best batch size is from 32 to 64.