what do people use for word embeddings (word2vec etc.?) for the moment i just want to use existing models, not train new ones. i tried dl4j, but it takes 5-6 minutes just to load the Google News model, vs. 45 seconds for python gensim.