Are there any good reading comprehension prompt or QA (question answering) prompt for squadv2.0 dataset?
I am trying to work on some QA tasks with GPT-3 but often find the model outputs ‘___’ at the end.
One prompt works for one dataset but not for another. The prompt in the GPT-3 paper for QA tasks just doesn’t work in my. case.
Currently my prompt is:
“”"
I read context and answer the questions.
Context: {}
Question:{}
Answer:
“”"
I also tried the openai.Answer API which works good in the 1-shot setting. But I am not sure how the API design the prompt and I also need to try. 0-shot in my setting but the api required 1 example at least.
Any suggestions?