Hi there,
I want to finetune a model (pytorch/transformers) for question generation tasks. However, I want to train it on longer contexts. How can I do this? I read online that sliding windows could be used for use cases similar to this one, but I can't seem to find an example online for question generation.
Question answering can be found, but differs from question generation, since for question generation the model needs to 'understand' the whole text before creating the question while for question answering, it can look per 'chunk' if it can answer the question.