This repo contains the source code of python API and HTML demo for Text Auto Completion.

You can use the pretrained GPT-2 model for general text, or follow the instruction at GPT-2 repo to finetune the GPT-2 model on your own data set.

Model in Tensorflow format can be converted to Pytorch using HuggingFace's Transformers library

from transformers.convert_gpt2_original_tf_checkpoint_to_pytorch import convert_gpt2_checkpoint_to_pytorch

gpt2_checkpoint_path = "path to TF checkpoint"
gpt2_config_file = "path to model config file"
pytorch_dump_folder_path = "path to save pytorch checkpoint"

convert_gpt2_checkpoint_to_pytorch(gpt2_checkpoint_path, gpt2_config_file, pytorch_dump_folder_path)

Source code of API is in webapp/src folder. Tested with python==3.6 and transformers==2.4.1

• Install libraries by pip install -r requirements.txt
• Update MODEL_PATH in config.py with the Python checkpoint path
• Run python app.py

The A Flask app is started at port 5000 which allow GET request.

# Request Arguments
input = flask.request.args.get('input')
temperature = float(flask.request.args.get('temperature', default=0.8))
top_p = float(flask.request.args.get('top_p',  default=0.9))

Please refer to this article for more information about temperature and top_p values

Source code of HTML demo is in webapp/html folder. Update URL constant in index.js file by the Text Generation API