Seq2Seq Generative Chat-bot Baseline

A Seq2Seq generative chat-bot baseline implemented by Tensorflow.

• Python: 3.6
• Tensorflow: 1.12

Model Architecture

GRU Seq2Seq + Luong Attention + Beam Search

Usage:

First, clone our repository to your computer and install dependencies:

git clone https://github.com/applenob/tf_chat_seq2seq.git
cd tf_chat_seq2seq
pip install -r requirements.txt

Prepare the Dataset

Download Chinese dialogue data from Dialog_Corpus.

git clone https://github.com/candlewill/Dialog_Corpus.git input

Then unzip the data.

Make sure we can visit data by path like input/xiaohuangji50w_nofenci.conv.

Then, you can change the configure in file config.yml.

Generate vectorized data:

python data_process.py --conv_path input/xiaohuangji50w_nofenci.conv

Try Tensorflow version model

Train Tensorflow version seq2seq model:

python main_tf.py --mode train

Have fun with the model you have trained:

python main_tf.py --mode predict

For Producing

Freezing model from checkpoint to proto-buffer file:

python tools/freeze_graph.py

Now, you can run your model with pb format:

python main_pb.py

Result Samples