Hey, we at deepset.ai have been working very hard on an open source framework for Transfer Learning in NLP. It’s called FARM and the core idea is that we separate out the language model component (takes tokens, returns vectors) from the prediction head part (uses these vectors to make meaningful predictions). We currently have support for Text Classification, NER, Language Model Finetuning and Question Answering. Feel free to direct questions and suggestions to me. If this seems like something that you’re interested in, leave us a star on github!

​

https://github.com/deepset-ai/FARM