Setup Quickstart

Setup

1. Install python, docker, and docker-compose and start the docker service
2. Open a terminal, clone the repository and go to the repository root
3. Run pip install -r requirements.txt, which will install all neccessary packages for the ./configure.py script

First application

Create a file named my.sw-config.yaml in the project root and write the following content to the file:

extern_environment:
  THREADS: 4
  BATCH_SIZE: 32
  CACHE_SIZE: 1000

metrics:
  - ./metrics/rouge
  - ./metrics/bleu
  - source: ./metrics/sbert
    disabled: true

summarizers:
  - ./summarizers/featuresum
  - ./summarizers/textrank
  - source: ./summarizers/neuralsum
    disabled: true
    environment:
      model: T5

This will configure BLEU, ROUGE, and Sentence-BERT as evaluation metrics, and neuralsum with the T5 model, our custom featuresum summarizer, and TextRank as summarizers.

note

If the disabled option is set to true, the plugin will not be loaded but information about the plugin will still be displayed in the demo. The disabled plugins in the example load large neural models. You can set disabled to false if you have enough space on your system.

note

With the environment option you can specify environment variables for the container in which the plugin gets loaded. For the neuralsum plugin the model variable is required.

You can find the configure.py file in the project root.
Run the following command to generate a docker-compose file in the project root:

./configure.py --config my.sw-config.yaml gen-docker-compose

Run the following command to start the application:

docker-compose up -d

The frontend will be exposed on localhost:3000 and the backend on localhost:5000 Go to http://localhost:3000 to use the application.

note

The startup may take some time and the frontend usually becomes available earlier than the backend. Also with the first startup the containers will be build and models will be downloaded. Therefore future startups won't take as long as the first.

note

The sw-config.yaml file in the project root will be used as the default config by configure.py when --config is omitted. Inspect this file if you want to get an overview of what a full config looks like.

note

For a overview of available plugins checkout the metrics/ and summarizers/ folders in the repository

Extern Environment

The extern_environment option sets environment variables in all plugin containers. The following environment variables are used to configure the plugin server.

Name	Default	Description
THREADS	1	The number of parallel calls to the summarize function.
BATCH_SIZE	32	This is the maximal length that the batch argument will have (see writing-a-plugin#required-arguments)
CACHE_SIZE	0	The size of the LRU cache. A unique sha256 key will be generated based on the input and the arguments. 0 disables the cache.