We want to create a question and answer model. How can one create a question-answering model like chatgpt, which determines the level of technology readiness (from 1 to 9) as a result of articles, patents or any product descriptions given to this model?
We thought of using the robeta model as a ready-made model, but how will we integrate the masking, fine tune and pre train stages with the article format (title, description, author, etc.) coming from the API according to this format and get information from this format and determine the technology readiness level of the article at a certain level?
I thought about giving a coefficient of 2 to each section, that is, to the words that determine the maturity level in the title section, giving a coefficient of 1.5 to the words that show the maturity level in the description section, and then giving a coefficient of 1 to the words that show the maturity level in the subtext, but how will the model learn this in that part? I'm stuck. It won't be consistent if it freezes a different result each time. Can you help me explain in detail how to proceed through these stages?