Support field labels for GeminiModel and GoogleModel on Vertex AI
#1056
Conversation
|
Do other providers support |
There was a problem hiding this comment.
EDIT: Not sure if this is ok, but I've created a new file for this test because:
- This test doesn't go in
tests/models/test_gemini.py, as import ofgoogle.authmight be unsuccessful, which would cause skipping the whole file.
@Kludex I don't think so, but I'm not 100% sure. From my experience, usually api-key-based providers use different api keys to track usage and spending. For example, with OpenAI you can create different projects and each project has its own key and that's how you track usage and spending across projects. |
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
|
I'll check this tomorrow. Sorry. |
no worries, take your time! |
…d-in-gemini-models
…d-in-gemini-models
vricciardulli
changed the title
labels for Gemini modelslabels for Gemini models on Vertex AI
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
|
@Kludex sorry to bother, do you think someone has time to check this soon? |
tests/models/test_gemini_vertexai.py
Outdated
| # pyright: reportDeprecated=false | ||
| # pyright: reportPrivateUsage=false |
There was a problem hiding this comment.
Those shouldn't be necessary in the top of the file.
There was a problem hiding this comment.
Removed # pyright: reportDeprecated=false
I also removed # pyright: reportPrivateUsage=false but then I manually built the mock response content. Just FYI, because in other testing modules it's not like that.
| request_data['safety_settings'] = gemini_safety_settings | ||
|
|
||
| if gemini_labels := model_settings.get('gemini_labels'): | ||
| provider_name = self._provider.name if isinstance(self._provider, Provider) else self._provider |
There was a problem hiding this comment.
Do we have an error on Gemini if send this parameter? It would be cool to be provider-agnostic.
This is just a comment. I'm thinking...
There was a problem hiding this comment.
Yes, we get a 400. Tested with Gemini-1.5-flash (GLA) raises:
pydantic_ai.exceptions.ModelHTTPError: status_code: 400, model_name: gemini-1.5-flash
with body:
{
"error": {
"code": 400,
"message": "Invalid JSON payload received. Unknown name \"labels\": Cannot find field.",
"status": "INVALID_ARGUMENT",
"details": [
{
"@type": "type.googleapis.com/google.rpc.BadRequest",
"fieldViolations": [
{
"description": "Invalid JSON payload received. Unknown name \"labels\": Cannot find field."
}
]
}
]
}
}Could also just let the user deal with this and then no need to check the provider.
…d-in-gemini-models
…d-in-gemini-models
…d-in-gemini-models
|
We're redoing the Google provider at #1373, so related PRs are on hold until that lands. |
…d-in-gemini-models
…d-in-gemini-models
|
I'll check this in a bit. Ping me when pipeline is passing please. 👍 |
|
@DouweM or @Kludex If possible, I'd need some guidance here, please: I've implemented the labels for both However, with GLA they behave differently. If a user provides labels in the settings dict:
I'm thinking it'd be best if they both behaved the same way, so which would be the best approach in your opinion?
EDIT: managed to fix the tests |
@Kludex tests are passing now:) I've got a question #1056 (comment) |
vricciardulli
changed the title
labels for Gemini models on Vertex AIlabels for GeminiModel and GoogleModel on Vertex AI
…d-in-gemini-models
…d-in-gemini-models
Gemini requests on Vertex AI support custom metadata labels. For example, one might use the labels to know which calls are coming from which projects and thus accurately keep track of billing.
See the Gemini API docs for more info.