[Bug]: errors when using database and config models

[lucinvitae]

What happened?

Our team recently noticed a bug when switching from only sourcing models from the litellm configuration file to using database models. When setting STORE_MODEL_IN_DB='True' using the existing configured models from a config file (1) result in 400 HTTP Bad Request errors and (2) the models intermittently display and disappear in the admin UI. Multiple members of my team are able to confirm both issues.

It seems like litellm cannot handle both database and config models simultaneously.

To reproduce the issue:

1. Add config file with some initial models (e.g. azure gpt-4.1)
2. test to verify model requests succeed and models display in admin ui
3. set STORE_MODEL_IN_DB='True'
4. add a different model from the same LLM provider (e.g. azure gpt-4.1-mini)
5. test to verify initial (e.g. gpt-4.1) model requests fail and admin ui glitches as mentioned above

Workaround:

• duplicate model configuration in database

Relevant log output

Level: High
Timestamp: 17:25:40

Message: LLM API call failed: `litellm.BadRequestError: You passed in model=gpt-4.1. There is no 'model_name' with this string . Received Model Group=gpt-4.1
Available Model Group Fallbacks=None`

Are you a ML Ops Team?

Yes

What LiteLLM version are you on ?

v1.72.0

Twitter / LinkedIn details

No response

[krrishdholakia]

Hey @lucinvitae, litellm is able to handle both simultaneusly - we run the same logic in our staging and i can see it working.

Can you share your proxy startup logs with --detailed_debug enabled? it might show what's happening

[lucinvitae]

Hey @lucinvitae, litellm is able to handle both simultaneusly - we run the same logic in our staging and i can see it working.

Can you share your proxy startup logs with --detailed_debug enabled? it might show what's happening

Thanks for the quick reply! Will do.

[lucinvitae]

Hmm, now I'm unable to reproduce the issue in our dev env.

There was one other thing that may have led to this broken state during the time this issue occurred in our prd env: I tried to create LLM credentials for some of the azure models, and then replaced those credentials. The bad state where config models would intermittently become unavailable seemed to happen sometime after that. Anyways, will close as cannot reproduce.