The server simply needs to know which model to load. By default, the server is run in the port 8080. Use the option --port to change it, specially if you have several servers, serving different languages, at once.

python3 src/mate-serve.py --port 8082 --model <path-or-huggingface-URL>

The model can be downloaded manually to some folder, it is easier to just use a Huggingface path. For, instance, for Emfatic you can use the following command:

python3 src/mate-serve.py --model models-lab/modelmate-emfatic-small 

For the moment we do not have an update site, but you can install the sources and test it a new Eclipse instance. For the moment, the plug-in expect to have the server running on port 8080.

For instance, for Emfatic you need to import these two projects.

modelmate.integration
modelmate.integration.emfatic

If you have an Xtext based language, you can easily create a plug-in for it reusing the infrastructure provided by these two plugins:

modelmate.integration
modelmate.integration.xtext

Then, you just create a new plug-in which the following extension point:

<plugin>
   <extension
         point="modelmate.integration.language">
      <language
            extension="mylang"
            tokenizer="modelmate.mylanguage.MyLanguageTokenizer">
      </language>
   </extension>
</plugin>

The tokenizer can be implemented by adapting the code in modelmate.integration.xtext.XtextTokenizer, something like:

public class MyLanguageTokenizer implements Tokenizer {

	private Injector injector = /* Get your language injector */;
	private Generator generator = new Generator(injector);
	
	@Override
	public String tokenize(String fragment) {
		return generator.toTokens(fragment);
	}
	
}

The experiments folder contains the source code to pre-process the dataset, build the ML models and evaluate them.

Checkout the experiments/README.md for more information about how to build the models.