Weird responses after fine-tuning

API
1

prompt: I can't find a job.
response: I see. Have you looked at different jobs? Changed your search criteria? What else could you suggest to yourself that would help solve this problem?
prompt: I don't know.
response: I see. Have you looked at different jobs? Changed your search criteria? What else could you suggest to yourself that would help solve this problem?

prompt: What are you talking about?
response: I see. Have you looked at different jobs? Changed your search criteria? What else could you suggest to yourself that would help solve this problem? I feel so set.

prompt: something relates to a number in response
response: I'm very sorry to hear that but you have so much to look forward to. Please seek help by contacting: 9152987821. You can also contact: 152987821 to speak to a specialist. ilysm yours always. (o)(o) ____ ____ ____ ____ ____ ____ ____ ____ ____ ____ ____ ____ ____ (o)(o) (o)(o) (o)(o) (o)(o) (o)(o) (o)(o) (o)(o) (o)(o) (o)(o) (o)(o) (o)(o) (o)(o) (o)(o) (o)(o

why it generates these kinds of responses and how to fix it?
thanks!

2

It looks like your temperature parameter might be too high. Also, the repetition looks like you trained with a high number of epochs? Any weird training data?

3

the temperature is 0, and the number of epochs is 16

4

OK, so 16 epochs + temperature of 0 explains the repetitions. What model did you fine-tune? I’m expecting maybe Ada or Babbage? Also, are you sending your stop sequence from training, something like ‘\n\n###\n\n’ ?

5

the model is davinci and the stop sequence is [‘\n’,‘->’]

6

It’s hard to tell what might be causing the weird response.

At this point, try increasing the temperature, to say 0.7.

Otherwise go with an embeddings approach, without the fine-tune.

7

You can just leave the whole idea behind. Solve the problem instead of telling the user how to solve it.

8

I don’t know about embeddings, what are the differences?
thanks