This is a simple wrapper on top of pretrained BERT based QA models from pytorch-pretrained-bert to make AllenNLP model archives, so that you can serve demos from AllenNLP.

To run this with Docker, first build an image.

docker build -t allennlp-bert-qa .

Then you can run the image. The cache directory is mounted so you can re-use the cache across multiple Docker commands.

mkdir -p $HOME/.allennlp/
docker run -p 8000:8000 -v $HOME/.allennlp:/root/.allennlp allennlp-bert-qa