Unable to Upload fine-tune file for gpt 3.5 turbo

API
1

Fine tuning for gpt-3.5 turbo has been released on 22/8/23.
I am trying to follow the instructions here

to upload one of the new style fine_tuning files. I have even tried to upload the exact file in the example. I have used PostMan and curl to try to upload the file and each time I get the error
{
“error”: {
“message”: “Expected file to have JSONL format, where every line is a valid JSON dictionary. Line 1 is not a dictionary (HINT: line starts with: "{…").”,
“type”: “invalid_request_error”,
“param”: null,
“code”: null
}
}

Obviously, the new style fine-tune file is not a jsonl(ines) format.
Is there a different purpose required? What is wrong?

2

This format can be uploaded. Could the requirements be expressed a little more precisely please.

{“messages”: [{“role”: “system”, “content”: “Marv is a factual chatbot that is also sarcastic.”}, {“role”: “user”, “content”: “What’s the capital of France?”}, {“role”: “assistant”, “content”: “Paris, as if everyone doesn’t know that already.”}]}
{“messages”: [{“role”: “system”, “content”: “Marv is a factual chatbot that is also sarcastic.”}, {“role”: “user”, “content”: “Who wrote ‘Romeo and Juliet’?”}, {“role”: “assistant”, “content”: “Oh, just some guy named William Shakespeare. Ever heard of him?”}]}
{“messages”: [{“role”: “system”, “content”: “Marv is a factual chatbot that is also sarcastic.”}, {“role”: “user”, “content”: “How far is the Moon from Earth?”}, {“role”: “assistant”, “content”: “Around 384,400 kilometers. Give or take a few, like that really matters.”}]}

3

Hi and welcome to the developer forum!

As far as I am aware the new format is indeed a json structure :

{"messages": [{"role": "system", "content": "Marv is a factual chatbot that is also sarcastic."}, {"role": "user", "content": "Who wrote 'Romeo and Juliet'?"}, {"role": "assistant", "content": "Oh, just some guy named William Shakespeare. Ever heard of him?"}]}
{"messages": [{"role": "system", "content": "Marv is a factual chatbot that is also sarcastic."}, {"role": "user", "content": "How far is the Moon from Earth?"}, {"role": "assistant", "content": "Around 384,400 kilometers. Give or take a few, like that really matters."}]}

taken from

image
image848×302 22.6 KB

4

The format is the standard chat model message structure, it is logical to use this well defined standard.

5

I’ve also discovered that there should not be any line breaks, as per the example at

1 Like
6

So far I’ve seen two examples with a total of four prompts coming from OpenAI for gpt-3.5-turbo tuning. Like functions, it’s almost up to you to discover, apparently…to busy to document, I suppose. And apparently one of those examples having unacceptable carriage returns?

First requirement, fine-tune a behavior set where just an example system prompt alone doesn’t already make a fully-(dis)functional AI…

“Marv is a factual chatbot, that is also sarcastic.”

  • Oh, the capital of France? Didn’t you know it’s Burkina Faso? Paris, of course!
  • I believe that timeless masterpiece was penned by none other than William Shakespeare. Truly an unknown and unappreciated author.
  • Oh, the Moon and Earth, they’re practically neighbors! It’s just a short stroll of approximately 384,400 kilometers. Easy peasy, right?

“You are an assistant that occasionally misspells words”
“Tell me a one-sentence story of a student’s day.”

  • Once upon a tiem, a student named Sara woke up late and rusheed to clalss, only to realize she had forgotten to study for her exam, but luck was on her side when the teacher postponed it to the next day.