[Bug]: Prompt Optmizer using invalid model

[YgorCastor]

File Name

https://github.com/GoogleCloudPlatform/generative-ai/blob/main/gemini/prompts/prompt_optimizer/vertex_ai_prompt_optimizer_ui.ipynb

What happened?

Looks like somewhere in the code is trying to use gemini 1.5 pro and referencing it through an invalid url.

Relevant log output

Detailed error: Error: The job failed: 404 Publisher Model `projects/XXXXX/locations/us-central1/publishers/google/models/gemini-1.5-pro-002` was not found or your project does not have access to it.

Code of Conduct

• I agree to follow this project's Code of Conduct

[GeeekyBoy]

Is there any timeline for the fix? Unfortunately, Gemini 1.5 Pro is used in the pipeline of the optimizer despite the model removal. I couldn't fix this myself since it seems that the usage of the model is hard coded in the docker image of the optimizer. Are there any workarounds?

[raybell-md]

I also came across this. I thought it might be because it's retired (https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions#legacy-stable) but it's not.

I also threw the stack trace into Gemini:

Stack Trace Analysis:

The provided stack trace indicates the error's origin and propagation through the application:

Origin: The error surfaces within a Python application, seemingly part of an optimization process (optimize_main.par/..., instruction_optimizer.py).
Libraries Involved: The trace shows calls progressing through:
Google API core libraries (e.g., google.api_core.grpc_helpers, google.api_core.gapic_v1.method).
Google Cloud AI Platform client libraries (e.g., google.cloud.aiplatform.vertexai.generative_models._generative_models.py, google.cloud.aiplatform.aiplatform_v1beta1.services.prediction_service.client.py).
A framework potentially named onetwo (various modules under onetwo.core, onetwo.builtins, etc.), which appears to handle the model interaction logic.
Point of Failure: The exception is raised when the application attempts to make a generate_content call to the Vertex AI service:
The call sequence includes self._text_generation_model.generate_content(...) within google/cloud/aiplatform/vertexai/generative_models/_generative_models.py.
This, in turn, calls self._prediction_client.generate_content(request=request) within the same module.
Exception Type: The immediate exception caught and re-raised by the Google API client is google.api_core.exceptions.NotFound: 404 .... This wraps a lower-level _open_source_grpc._channel._InactiveRpcError with status = StatusCode.NOT_FOUND. This confirms that the "Not Found" status was returned by the remote gRPC service.

[aycaecemgul]

I have the same problem, setting the target_model to another gemini version did not help...

[holtskinner]

@want-to-be-relaxed or @inardini can you look into this?

[hootan-na]

Sorry for the inconvenience. The prompt optimizer was using Gemini 1.5 Pro as the optimizer model to rewrite the prompts. Since Gemini 1.5 Pro is deprecated, we just released a fix using a more recent model.

[raybell-md]

Can this be closed?

Has anyone validated?

[GeeekyBoy]

I have tried it after @hootan-na's comment, and it was working. ✅