Accessing to different projects finetuned models

API Feedback
,
1

I’m working with OpenAI’s API and fine-tuned models, and I’ve encountered a problem regarding accessing fine-tuned models across different projects within the same organization.

Problem:

  • I have several projects within the same organization on OpenAI, seperated due to functionality (ex R&D etc).
  • When using the API key for Project A, I can access fine-tuned models associated with Project A.
  • However, I am not able to access a fine-tuned model from another project (Project B) in the same organization using the API key of Project A.
  • Additionally, in the OpenAI web UI, when selecting models, I can see the fine-tuned model from Project B, but it doesn’t show up in the list of fine-tuned models for Project A when I look at the fine-tuned model list provided by the API.
  • The error code is like below:
    openai.NotFoundError: Error code: 404 - {‘error’: {‘message’: ‘The model ft:gpt-4o-mini-2024-07-XX:XX... does not exist or you do not have access to it.’, ‘type’: ‘invalid_request_error’, ‘param’: None, ‘code’: ‘model_not_found’}}

My Questions:

  1. Is this behavior intended? I expected to be able to use other projects finetuned models.
  2. Should models be scoped strictly to their own project? Is there some organization-wide scoping that I’m missing, which might explain this cross-project model access?
  3. Does anyone know how to fix this? Is there any setting I am missing to set to be able to access projects finetuned models?

Any insights or similar experiences would be appreciated!

1 Like
2

I have the same problem, I want to do the fine-tune in separate R&D project, but can’t use the fine tunes in client facing projects like it is it seems.

3

I encountered the same problem

4

This is still a problem. Can anyone do smth with it?

1 Like
5

This restriction still exists. I can raise it with the team, as it has been mentioned repeatedly over the years.

As a possible workaround, you could create a dedicated service account for accessing another project’s fine-tuned models, with the option to delete that service account once the task is complete.

6

Raise this issue, please.

I have revisited the service account applicability. It can resolve the issue of accessing it (similar to just using the API key from my R&D environment in PROD), but in a more secure way.

However, people use different projects not only to separate permissions, but also to separate costs. From what I’ve just seen in the playground, I can’t separate spending between different users/API keys within a single project.

For my use case, it would be helpful either to have organization-level access to the models I train inside Dev/R&D/etc. projects, or to have clear usage separation between users/API keys so that the executive staff can analyze spending and understand how much each service is using.

4 Likes
7

@vb Have you been able to discuss it with the team? Any updates on this?