It seems like making pre-trained ULMFiT models available in many popular languages would fit with the mission of fast.ai. This isn’t unheard of - FastText makes it easy to download pre-trained language vectors, and word2vec-like vectors are available in many languages at Polyglot.
I brought this up with AWS a few weeks ago, and suggested they host the models, and they wanted to talk more about it. Last week at re:Invent, I ran this by Andrew Shaw and @matt.mcclean and they suggested I bring it up here. I assume the plan is to do something like this with http://nlp.fast.ai/category/model_zoo.html , but that page has been coming soon for a while, so I was hoping to get the ball rolling!