Hello everybody,
I’m currently taking the NLP specialization and also a course about NLP at my university. As part of this course we’ve received the task to build a model that can answer questions about swiss employment law. We don’t have much time for the project (only 6 weeks) to generate labelled data, so my initial idea was to take an already available model from huggingface or another platform, fine-tune the training with unlabelled data which are texts from law documents (in german) and then further fine-tune it with a few input/output exemplars of potential questions/answers.
With so many different models available, I wanted to ask which model and procedure do you think would be best suited for this project? Bert, roberta, gpt or palm come to mind, but certainly there are other options.
Thank you very much in advance for your feedback,
Sven