Chatgpt 4o-mini fine-tuning fails.Internal error

API
1

Hello,
I am encountering the following error:
The job experienced an error while training and failed; it has been re-enqueued for retry.

After this error, the fine-tuning process completely fails.

I am trying to fine-tune the model with German, French, and Turkish data.

  • I have already trained the AI with English data, and now I am training it with only 10% of the data in each of the languages mentioned above. My goal is to ensure the model works well with these languages.
  • Initially, I created a single file containing data in all three languages and started fine-tuning. It failed.
  • Then, I broke it down into three separate files and started simultaneous fine-tuning jobs.
    • The fine-tuning for French and German succeeded, but Turkish failed.
  • I examined the Turkish data, but everything seems fine.

Because both French and German fine-tuning succeeded, I thought I could fine-tune the model trained with French data using the German data, or vice versa.

  • However, this attempt failed again (I tried both vice versa scenarios).

This behavior is quite puzzling, and I am unsure of what steps I should take next.

2

Fine-Tune jobs are failing a lot for me as well. I think it is an issue on OpenAI’s end.

This was not an issue I have faced before, but now most fine-tuning jobs are failing for me as well with the same error message.

3

I have the same problem, I can only try once every 2-3 hours

4

Same thing has been happening to me now for GPT-4o. Nearly every fine-tuning job I run is failing. The strange thing is I can fine-tune the exact same file with GPT-3.5-turbo and it works fine. This tells me something is going on with OpenAI.

5

I am having the same issue with gpt4o-mini and a .jsonl file I have used before. Is it possible that the .jsonl format has changed since the fine tuning process now has two options (supervised and direct preference optimization)? I assume that the older format is supervised which is the default. The file I submitted passed the validation stage and the training began with metrics showing, but then it failed with the internal error a few minutes later.

6

Update - my gpt-3.5-turbo job is failing now too.

Update - the gpt4o-mini job finally went thru (!) after many retries and hours. Obviously an internal OpenAI issue.

7

I’m experiencing the same issue and have tried multiple times since yesterday. Has anyone been able to resolve it or identify the source of the problem?

8

A fix has been deployed.
You can try to run your jobs again and please report in the topic below, if you should run into further issues.

9

This topic was automatically closed 2 days after the last reply. New replies are no longer allowed.