text-generation-transformer

text generation based on transformer

Tokenize

Tokenizer is based on pytorch-pretrained-BERT

Bert pretraining model will probably be integrated into the model soon if better result is achieved.

Specify path to input data

See Data Loading / fashion data module

For details please refer to torchtext.data.TabularDataset

Online prediction

See Inference: one example prediction module

This code is heavily borrowed from The Annotated Transformer

Thanks for the great work!