Skip to content

bloomsburyai/cape-webservices

Repository files navigation

cape-webservices CircleCI

Entrypoint for all backend cape webservices.

Frontend demo is here (only works if you already launched a Backend).

Overview of Cape

Cape is a suite of open-source libraries to manage a question-answering model that answers questions by "reading" documents automatically. It is based on state-of-the-art machine reading models trained on massive datasets, and includes several mechanisms to make it easy to use and improve based on user feedback. It has been designed to be portable, i.e. works on a single laptop or on a cluster of parallel machines to speedup computation, and is Open Source friendly to be used at all expertise levels.

It enables users to

  • upload documents and answer questions extracted from them,
  • update models by adding a "saved reply", i.e. a pre-defined answer,
  • manage users, documents and saved replies.

There are several ways to use Cape :

  • As a python library :
 from cape_responder.responder_core import Responder
 Responder.get_answers_from_documents('my-token','How easy is Cape to use', text ="Cape is an open source large-scale question answering system and is super easy to use!")
  • As a python service : python3 -m cape_webservices.run
  • As a standalone Docker container : docker run -p 5050:5050 bloomsburyai/cape
  • As an app with UI
  • As a distributed cluster
  • As a slack bot (video here, more info here)
  • AI-in-the-middle email answering system (video here, more info here)
  • As a Facebook bot (more info here)
  • As a Hangouts bot (more info here)

Quick start

Minimum Requirements

We recommend at least 3GB of RAM and at least 2 modern CPU cores (4 if virtual). If you're using Docker, ensure you increase the memory resource limits in the Docker preferences.

Standalone webapp with Docker

You can run a standalone version of the webapp that includes a management dashboard. After installing docker, update and run the Cape image:

docker pull bloomsburyai/cape && docker run -ti -p 5050:5050 -p 5051:5051 bloomsburyai/cape

This will launch both the backend and the frontend webservices, by default it will also create tunnels for both, outputting the public urls:

  • To use the frontend just browse to the given url, it will be something similar to : **https://RANDOM_STRING_HERE.ngrok.io?configuration={"api":{"backendURL":"https://RANDOM_STRING_HERE.ngrok.io:5050","timeout":"15000"}}
  • To use the backend you can use our client (documentation here or make your own by integrating our HTTP API (documentation here))

Quick Start Guide with Docker

  1. Pull the latest version of the Docker image (it will take a few moments to download all dependencies and a machine reading model): docker pull bloomsburyai/cape

  2. Run the Docker container and launch an IPython console within it using the following command: docker run -ti -p 5050:5050 -p 5051:5051 bloomsburyai/cape ipython3

  3. Import Responder: from cape_responder.responder_core import Responder

  4. Ask a question and store the response (which is a list of answers) and display the first answer using: response = Responder.get_answers_from_documents('my-token','How easy is Cape to use?', text="Cape is an open source large-scale question answering system and is super easy to use!"); print(response[0]['answerText'])

  5. If you are interested in understanding a bit more about what the response looks like, display the full response using: print(response)

Installing without docker

To natively install Cape on a linux system, take a look at deployment/Dockerfile.

Structure

Dependencies Diagram

In summary this is how Cape is organized:

  • cape-webservices Backend server providing the full HTTP API, depends on :
    • cape-responder Unique high level interface for distributing and creating machine reading tasks,depending on :
      • cape-machine-reader Module to integrate machine reading models
      • cape-document-qa Integration of a state of the art machine reading model, with training and evaluation scripts
    • cape-document-manager Interface to manage document and annotations, using SQLite as an example storage backend, depends on :
      • cape-splitter Package to split documents into chunks without breaking sentences
    • cape-userdb, Package to manage and store users and configurations
    • cape-api-helpers, HTTP API utility functions
    • Optional plugins:
  • cape-frontend Frontend server (not in the diagram) it uses the backend server API to provide a management dashboard to the users