Classification model - GPT 3.5 API - Best way to pass the labels (CSV or JSON format?)

I plan to use GPT 3.5 API to classify some merchant names into categories (e.g. restaurant, hotel). For each request/user I will be sending a merchant name and different list of categories.

I am wondering if it is better to structure the labels(categories) in a CSV or in JSON?

best to just try it and see the results and compare (do this in the playground first). I think if the data has a complex structure you’d be better off with JSON, but if it is a simple CSV, it will work well. If you are concerned about token costs becasue you have to process a large volume of data then CSV will be quite alot cheaper, becasue you’re not lengthening the data with property names on each record.

We are working on a similar use case. How is it coming along? We hage had trouble getting it to return Hotle each time it shows up in the name. We even give examples as such and set temp at 0.

I was not able to make much significant progress, how about you?

I trained babbage-002 and on fine tuned one when i run completion, it doesnot return me any label but random text. any help

Hi have you managed to do that with the agents? I have tried with the pandas dataframe but it is working slow and it costs too much. If you have experienced those I would like to get your opinion. Thanks