CONTRIBUTING.md

Contributing a new model to the demo

The following describes the steps necessary to add a model implemented in allennlp to the online demo.

These assume that the model to be added to the demo is already available in master of allennlp. We will follow the implementation of the SRL demo as an example.

Instructions

1. Fork and clone allennlp-demo, then follow installation instructions.
2. Dockerfile:
   • Make sure the commit id for allennlp/commit points to a commit where your model is implemented. Alternatively, you can use a release that includes your model with, for example, allennlp/allennlp:v0.6.2.
3. server/models.py: Instantiate a DemoModel object in the MODELS dictionary.
   • Provide a url to the trained model (e.g., https://s3-us-west-2.amazonaws.com/allennlp/models/srl-model-2018.05.25.tar.gz).
4. demo/src/components/menu.js:
   • Add a buildLink clause:

   {buildLink("semantic-role-labeling", "Semantic Role Labeling")}

5. demo/src/components/:
   • Add a <your-model-name>Component.js file.
   • this is where most of the model-specific js logic is implemented. See demo/src/components/SrlComponent.js for example.
6. demo/src/App.js:
   • Import the model component implemented in the previous step:

   import SrlComponent from './components/SrlComponent';

   • In the instantiation of ModelComponent, add an else if clause returning an instance of your new component:

      else if (selectedModel === "semantic-role-labeling") {
          return (<SrlComponent requestData={requestData} responseData={responseData}/>)
      }

7. app.py:
   • Add a app.route to the html page, using the same name provided in step (4). This allows users to link directly to your model.

   @app.route('/semantic-role-labeling/<permalink>')

   • Consider also adding logging of your model's outputs. Search for log_blob in the predict function.
8. Add a test for your model to tests/server_test.py.
   • For instance, see test_semantic_role_labeling.