Fine-tuning cannot find training file

[alcop22]

Hello. I’m getting an error when creating fine-tune models.
Specifically, the response says that the training file does not exist.

When doing the call via API:

When directly creating the fine-tune:

However, the file exists. Looking at the Dashboard, it is there.

File is also returned properly whenever I call the Retrieve File API.

I’ve tried the ff.

• Upload the file directly to the Storage dashboard
• Upload the file via API, then create fine-tune
• Upload the file via Create Fine-tune

But nothing works. I’m not sure what else to do.
This was working a week ago.

[jr.2509]

Hi there!

I can confirm that referencing the file ID when creating a new fine-tuning job via the UI still works - I tested it just now.

Is it possible that the file was uploaded under a different project than the one you are trying to access it from?

[alcop22]

I tried your suggestions.
Clearing the browser cache, or logging in a different browser didn’t work.

The file and the project where I’m creating the fine-tune is the same.
At the moment, I’m just trying to create the fine-tune job in the Dashboard, but it isn’t working either.

But if the issue is only on my end, is there a possible limit or policy I didn’t know that could trigger this error?

As far as I’ve tried, I can still use my previous fine-tuned models, interact with the assistants and completions and embeddings API.

Only creating the fine-tune model specifically doesn’t seem to work.

[jr.2509]

There are some daily limits in place. I believe you cannot create more than 16 models per day for gpt-4o-mini and gpt-4o (for all other models that can be fine-tuned such as gpt-3.5-turbo the daily limit is 48). It is possible though that this number has changed.

If the issue persists, I would reach out directly to OpenAI support via the chat widget on the developer platform as might be something account specific.

I hope it gets resolved soon!

[alcop22]

For the daily limits, this is the first time today that I was planning to create a new model.
I’ve also tried all of the possible models, but the issue persists.

Ok, I’ll try that as well. I hope it does get resolve soon.
And today was supposed to be a demo to the client.

Thanks for the help.

[alcop22]

I got a general approach on how to debug the issue when I reached out to the OpenAI help platform, but that didn’t work. But after waiting for around two hours, the issue got resolved somehow.

There’s no solution, however, so it was either shadow fix or a server connection issue.
Thanks for the assistance @jr.2509.

[jr.2509]

Glad to hear it is (mysteriously) resolved!

[le.thinh]

I got the same issues, here. Anyone who resolved this issue can tell me what is the problems here?
Error code: 400 - {‘error’: {‘message’: ‘file-NOt7qfVTlkx7ve1n9pZrJ193 does not exist’, ‘type’: ‘invalid_request_error’, ‘param’: None, ‘code’: None}}

I believe that this file is exist because if not exist, it is impossible to get this ID.

[jr.2509]

Welcome to the Forum!

Have you uploaded the file with the purpose “fine-tune”?

[le.thinh]

Yes. I uploaded the file with the fine-tine as purpose

[jr.2509]

Hm. I’m surprised you were able to successfully upload this file given its naming convention.

I’d rename the file to sth like annotations_train.json and retry.

[le.thinh]

I did change the name and retry but got this error again.

[jr.2509]

Thanks. You should not be using two points in the naming of the file.

[le.thinh]

Thank you, but it’s still the same; I still can’t find the file. Do you think it could be due to the file content? I got the files from Roboflow

[jr.2509]

Sorry mate, not sure how to be of any more help. Consider reaching out to OpenAI support via the chat widget on the Developer platform to have this investigated further.

[amin.alian]

Hi,
I’m also running in to the same issue. Have you received any updates?

[le.thinh]

I create a new OpenAI account and it works. That’s bad.

[le.thinh]

Does OpenAI have human support?