How to get a single response from fine-tuned model?

Hi all,

Thanks to all the amazing support on here, I’ve managed to get up and running with a fine-tuned model. However, I think I’m doing something wrong!

My training data is an attempt to answer 2 questions. The first is a yes/no/don’t know response to “Did I get your name right?” and the second is an attempt to ascertain which option was chosen from 5 choices.

I created my data in Excel and then used the openai CLI tool to prepare the data - it created two JSONL file, one whose name ends in _train.jsonl and the other ends _valid.jsonl.

Here is a sample chunk of one of the jsonl files:

{"prompt":"question: Did I get your name right?\\nReply: Yes, I think so. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Yeah, most definitely. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Somehow yep, you did. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Yes you did. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Yep, affirmative. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Totes. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Yup, assuredly. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Actually, yeah. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Yup, I'm pretty sure you did. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Yes, you sure did. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Most assuredly. ->","completion":" yes"}
{"prompt":"question: Did I get your name right?\\nReply: Do me a favor ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: Don't be stupid. ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: Nope not even close. ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: No, sadly not. ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: Nah, absolutely not. ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: No, that's not even close. ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: Nope. You having a giraffe? ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: Nope. Miles off. ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: Nope, that's not even close. ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: Course not. ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: No, I'm afraid not. ->","completion":" no"}
{"prompt":"question: Did I get your name right?\\nReply: Dunno. ->","completion":" unsure"}
{"prompt":"question: Did I get your name right?\\nReply: I haven't the foggiest. ->","completion":" unsure"}
{"prompt":"question: Did I get your name right?\\nReply: Perhaps ->","completion":" unsure"}
{"prompt":"question: Which option would you like to try?\\nReply: Desirables ->","completion":" desirableExpenses"}
{"prompt":"question: Which option would you like to try?\\nReply: I'd like to get a paper round. ->","completion":" extraIncome"}
{"prompt":"question: Which option would you like to try?\\nReply: Add extra income ->","completion":" extraIncome"}
{"prompt":"question: Which option would you like to try?\\nReply: Increase extra income ->","completion":" extraIncome"}
{"prompt":"question: Which option would you like to try?\\nReply: Look at adding my goal ->","completion":" goal"}
{"prompt":"question: Which option would you like to try?\\nReply: I would like to look at adding goals ->","completion":" goal"}
{"prompt":"question: Which option would you like to try?\\nReply: I would like to try changing my goals ->","completion":" goal"}
{"prompt":"question: Which option would you like to try?\\nReply: Reduce my spending budget ->","completion":" essentialExpenses"}
{"prompt":"question: Which option would you like to try?\\nReply: I want to try reducing my cost of living ->","completion":" essentialExpenses"}
{"prompt":"question: Which option would you like to try?\\nReply: Reduce my cost of living ->","completion":" essentialExpenses"}
{"prompt":"question: Which option would you like to try?\\nReply: Inheritance ->","completion":" windfall"}

There is a lot more data than this in my Excel sheet - hundreds of entries, and the openai tool put them in a weird order, but that snippet gives the gist.

I successfully created a fine-tune job and got a model I can use, based on “Ada”. However, when I try example answers in the play ground, I get 30 or 40 results! e.g.

Of course, I’d just like “no” as my response in that case.

Also, sometimes, the wrong question is being answered:

does this mean I need a fine-tune model for every single question?

And lastly (and this is probably the dumbest question of all, since I’m asking for a completion!), how do I stop the AI completing the question before answering:

?

This whole prompt → completion methodology is very new to me and my brain, so please be gentle with me, and thanks for any help at all!