I am trying to use the huggingface.co pre-trained model of Google T5 (https://huggingface.co/t5-base) for a variety of tasks. But I can`t find a list of many tasks it really supports and how to address them. I found summarize: + the text to summarize. I also try to find an overview in the paper (https://arxiv.org/abs/1910.10683) there are for instance examples of question-answering in the appendix but without instructions how to tell T5 to answer a specific question. The huggingface.co documentation did not provide any further information besides the summarize: declaration.

  • $\begingroup$ As you already mentioned in the question, starting from page 47 of the paper, It is giving 18 different datasets along with how you need to preprocess them for these tasks. Each of these datasets is for different purposes which you can figure out when looking at the Processed input of each section. $\endgroup$ – Fatemeh Rahimi Jan 9 at 5:18
  • $\begingroup$ So if I want to apply SQuAD-like data I just have to format my query this way: question: <question> context: <context> and I will get an answer? And if I want to paraphrase an input I can´t use t5-base because this type of query is not listed in the appendix? $\endgroup$ – tschomacker Jan 9 at 15:46
  • $\begingroup$ Do the research. If you want to use T5 for question answering, look for code samples just google it(they are talking about it in the repo readme page). Or if you want to paraphrase something. Check if they have already provided a notebook for that. If not, send the authors an email to figure this out. Answer to questions doesn't always come easy. $\endgroup$ – Fatemeh Rahimi Jan 10 at 19:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.