This repo is based on https://github.com/pytorch/pytorch/blob/master/torch/nn/modules/rnn.py.

wget https://oneflow-public.oss-cn-beijing.aliyuncs.com/datasets/models/Imdb_ofrecord.tar.gz
tar zxf Imdb_ofrecord.tar.gz

bash train.sh

Note that the parameters of the provided pre-trained model is :

BATCH_SIZE=32
EPOCH=30
LEARNING_RATE=3e-4
SEQUENCE_LEN=128
EMBEDDING_DIM=100
NFC=128
HIDDEN_SIZE=256

bash infer.sh

The example text is a simple sentence "It is awesome! It is nice. The director does a good job!" (or "The film is digusting!"). You can change this text in infer.sh.