train.sh

#!/bin/bash

# Before you train, please make sure your directory structure is as follow:
#
# usr_dir \
#     __init__.py
#     merge_vocab.py
# data_dir \
#     train.txt.up.clean
#     train.txt.down.clean
#     merge.txt.vocab.clean
#


TRAIN_DIR=./output
LOG_DIR=${TRAIN_DIR}
DATA_DIR=./data_dir
USR_DIR=./usr_dir

PROBLEM=translate_up2down
MODEL=transformer
HPARAMS_SET=transformer_small

# generate data
t2t-datagen \
  --t2t_usr_dir=${USR_DIR} \
  --data_dir=${DATA_DIR} \
  --problem=${PROBLEM}

# train data
t2t-trainer \
--t2t_usr_dir=${USR_DIR} \
--data_dir=${DATA_DIR} \
--problem=${PROBLEM} \
--model=${MODEL} \
--hparams_set=${HPARAMS_SET} \
--output_dir=${TRAIN_DIR} \
--keep_checkpoint_max=1000 \
--worker_gpu=1 \
--train_steps=200000 \
--save_checkpoints_secs=1800 \
--schedule=train \
--worker_gpu_memory_fraction=0.95 \
--hparams="batch_size=1024" 2>&1 | tee -a ${LOG_DIR}/train_default.log