combine RE2RNN with Merryup method and train together
1st get tokens and tags
2nd get all_tokens and all_tags
3rd get dict(w2i, i2w, t2i, i2t)
4th add padding
5th create dataset
6th create dataloader
use bilstm and glove emeds
hidden2tag layer
out with softmax
remember to use forward and predict