Return JSON data is not proper while extracting NER and RE from the Text

Currently, I am exploring the OpenAI API for Named Entity Recognition (NER) and Relation Extraction (RE) in the Japanese language.
I have tried different models and found that the text-davinci-003 model yields good results, so I am currently using it. However, I have encountered an issue where each time I use the model, it provides a different response. Additionally, the key names in the key-value pairs of the JSON output are also inconsistent.

What is your current completion temperature setting @devang.hingu ?



As per documentation i set it to temperature: 0.2 so i get more accurate instead of random data.

I also checked with temperature:0 but still issue with JSON format. sometimes i still get broken response!

@devang.hingu were you able to solve this? Do you have any tips or tricks?