The main entry point is wsd.py which has three modes: train, eval, and topk. See the documentation for the command line flags in wsd.py for options. To download a sample pre-trained model, run the download_model.sh script. For an example call to generate top K results, see the topk.sh script (which assumes download_model.sh has been called in the root directory).