Finetuning on Platform internal server errors

Over the past day or so, I’ve attempted a handful of fine-tuning runs, all of which start the process, experience an error and then fail.

We’ve been running finetunes quite frequently over the past few months, and as of yesterday it is consistently failing.

I’ve looked around but am yet to see anything of other people experiencing the same so thought I would make a post.

Here is a photo of what I’m seeing:

If anyone has any information, it would be greatly appreciated.

Hi and welcome to the Forum!

Based on a quick test on my end, I had no issue with the fine-tuning process via the platform and the job completed successfully.

Do you have sufficient funds/credit on your developer account to cover the cost for the fine-tuning job?

Unless someone else has another idea, I’d suggest to reach out directly to support via the chat widget to detail the issue and have them look into it.


Alright, thanks for checking.

I do have plenty of funds and I’ve rerun fine-tunes that have previously run successfully and are now failing.

Alright, I’ll try reaching out and see how it goes.

Thanks for the help.

1 Like

Good luck! When you do get a reply, perhaps you could report back here what caused the issue. Might be relevant to fellow Forum users.

1 Like

I tried training on the minimum number of samples instead of my full dataset and when I did that I got this error message in validation:The job failed due to an invalid training file. Invalid file format. Line 1, key "messages": The last message must be from the assistant

So I just added an assistant message as the end message to all my samples and it now seems to be training.

Weird since it passed validation with my full dataset, but showed the message when it was just a subset. Anyways, that seemed to do it.

1 Like