I’m using “gpt-3.5-turbo” in my RAG implementation.
Providing json data as external knowledge. Getting accurate results. But the only problem here is for few objects it is giving answers based on the question format like “What is the ‘key’ ?” and for few objects it is giving answers based on the question format like "Give the details of the ‘key’ ? "
For example:
{‘query’: ‘give the details of the Date Formed?’, ‘result’: ‘“I don't know”’}
{‘query’: ‘what is Date Formed?’, ‘result’: ‘{\n “Date Formed”: [\n {\n “value”: "06/08/2015 ",\n “key_confidence”: 67.83586883544922,\n “val_confidence”: 99.92040252685547,\n “key_coordinate”: [\n [\n 0.5047077536582947,\n 0.2812928855419159,\n 0.029619064182043076,\n 0.00990480650216341\n ],\n [\n 0.5376865267753601,\n 0.28164637088775635,\n 0.053304523229599,\n 0.00980121549218893\n ]\n ],\n “value_coordinate”: [\n [\n 0.726016104221344,\n 0.2817370593547821,\n 0.07695826143026352,\n 0.011224324814975262\n ]\n ],\n “page_no”: 0\n }\n ]\n}’}
**Note:**Here for the same key “Date Formed”, the model is giving correct answer for the question type what is Date Formed? but when we ask like give the details of the Date Formed? it is giving I don’t know.
What problem are you trying to solve?
And what do you mean by “Date Formed”?
I want to get accurate results from RAG, irrespective of question format.
we have JSON data which contain multiple objects related to some loan application. I’m giving that json data as external knowledge base(in RAG).When I ask question related to any object it should supposed to give all the details related to that object.
Here the “Date Formed” is one of the objects.
