Use run.sh to train.
The argument --t2t_usr_dir=.
registers the model into tensor2tensor.
Please refer to tensor2tensor documentation for other usages.
Installation:
# Assumes tensorflow or tensorflow-gpu installed
pip install tensor2tensor
# Installs with tensorflow-gpu requirement
pip install tensor2tensor[tensorflow_gpu]
# Installs with tensorflow (cpu) requirement
pip install tensor2tensor[tensorflow]
Implementation of transformer: tensor2tensor/models
Implementation of Grave's ACT for RNN: DeNeutoy/act-tensorflow