Simple QA

Question answering using pretrained models based on Xapian and XLNet

• Usage:

  • For anyone want to try, I suggest to use docker.

    • For CPU version, run to pull image:

          docker pull zhupengjia/simple-qa:cpu

    • For GPU version, run to pull image:

          docker pull zhupengjia/simple-qa:gpu

    • Download pretrained model from following link to your directory:, then decompress:

      https://1drv.ms/u/s!AnzH-f0hZoPctxAyoLAyA-b0ab6A?e=F2ks1i

      Then decompress:

          tar xzvf squad2_xlnet.tar

    • Make sure you have data that in the format of .pdf, .txt, .gzip, .bzip2

    • Then try to run manually:

      Assume your directory contains model and data

          docker run -d -v YOURDIRECTORY:/opt/chatbot/data --name simple_qa zhupengjia/simple-qa:cpu tail -f /dev/null

      For GPU version, please make sure you have installed nvidia-container-runtime, and has the following item in your /etc/docker/daemon.json file:

          "runtimes": {
              "nvidia": {
                  "path": "/usr/bin/nvidia-container-runtime",
                      "runtimeArgs": []
              }
          }

      Then create container as:

          docker run -d --runtime nvidia -v YOURDIRECTORY:/opt/chatbot/data --name simple_qa zhupengjia/simple-qa:cpu tail -f /dev/null

      Then manually do:

          docker exec -it simple_qa bash
          python3 interact.py  -m data/checkpoint-7900 --returnrelate --backend shell data/sample.txt

      Have fun

    • If you want to run a restfulapi:

          cd docker

      modify docker-compose.yml for docker-compose-gpu.yml for image, environment, volumes

          docker-compose up -d

      or

          docker-compose -f docker-compose-gpu.yml up -d

  • If you want to train model by yourself:

    cd ANYDIRECTORY
    git clone https://github.com/huggingface/transformers.git
    cd transformers/examples
    python run_squad.py --do_lower_case --version_2_with_negative --model_type xlnet --model_name_or_path xlnet-large-cased --do_train --do_eval --train_file data/train-v2.0.json --predict_file data/dev-v2.0.json --learning_rate 3e-6 --num_train_epochs 12 --max_seq_length 384 --doc_stride 128 --output_dir ./finetuned_squad_xlnet --per_gpu_eval_batch_size 2 --per_gpu_train_batch_size 2 --save_steps 100 --fp16 --gradient_accumulation_steps 100 --overwrite_output_dir --do_lower_case

  • If you want to run in local machine::

    Please make sure you have installed python-xapian. If you want to parse pdf file, please make sure you have installed poppler-utils

    • run:

      python interact.py -m MODELPATH --returnrelated --scorelimit 0.2 textfile_path