This is a python / HTML template that you can hook up your NLG model to for people to interact with.

• python3 (http.server and urllib.parse native libraries)

The easiest thing to do is to create a symbolic link from the repo to the project/model folder:

cd your_project_directory
ln -s SOMEWHERE/NLGWebsite .

From a file in your_project_directory, import the main server method:

sys.path.append("NLGWebsite")
from server import run

Your NLG model must be created, instantiated and loaded

myNLGmodel = myNLGClass(hyperparams) # myNLGClass could be a from_pretrained transformer model, for example

It must also have a method that takes in an input string (it doesn't have to do anything with it if you don't want that) and returns an ouput string (your NLG response). For instance:

myNLGmodel.getOutput("Hello this is an input string!")
# returns a generated output string

Start up the demo server by calling the run method, feeding your nlg model to it:

run(nlg = myNLGmodel)

Then go to your web browser and access the page through http://localhost:8000 (or wherever your server is running).

You can also run the server with just an output function that takes a str and outputs a str (e.g., model.getOutput is one of those functions)

runNoClass(myOutputFunction)

Maarten Sap (& maybe other peeps!)