The files in this directory are how I imagine an actual deployment of the model would work. A single instance (ml_model/) can handle all the predictions, while the actual visualization website (static/) is more simplistic, basically only using Leaflet.heat.
The trained model should be located at ml_model/weights/firecast.pth, and all .tiff and .xml files generated by GDAL should be located under ml_model/maps.
gen_heat handles everything necessary for generating the predictions.
The visualization demo uses Leaflet and Leaflet.heat to display predictions. It imports all the predictions from a single file, in this case named latest.js.