This will create a feature request which anybody can view and comment on.

Please describe your requested enhancement. Good feature requests will solve common problems or enable new use cases.

What you would like to accomplish:

I would like to see a comprehensive list of available models within the Google Cloud AI Platform, similar to the Model Garden, or a dedicated list of foundation models like Gemini directly available within the API. This list should include detailed information about each model, such as its display name, version, supported context window, token limit, and other relevant metadata.

How this might work:

Introduce a new publishers.models.list API endpoint: This endpoint would return a list of available foundation models or models from the Model Garden, depending on how Google decides to categorize and expose them.
Enhance the existing publishers.models.get API endpoint: Currently, even with the view=FULL parameter, the response only provides minimal information (mostly just the model ID). This endpoint should be expanded to include comprehensive model details:
- Model Display Name: A human-readable name for the model (e.g., "Gemini 2.0 Pro").
- Model Version: The version of the model.
- Supported Context Window: The maximum context window the model supports.
- Token Limit: The maximum number of tokens the model can handle in a single request/response.
- Supported Regions: A list of regions where the model is available.
- Model Description: A brief description of the model's capabilities and intended use cases.
- Model Type/Family: Whether it is a general-purpose LLM, embedding model, image-generation, etc.
- Publisher: In case of Model Garden, clearly indicate the model publisher.
- Training Data Information (if available/applicable): A high-level overview of the data used to train the model, if that information is available for release.
- Any relevant limitations or caveats: (e.g., "This model is still experimental").

If applicable, reasons why alternative solutions are not sufficient:

Currently, the only way to discover available models is through the Google Cloud console's Model Garden or by knowing specific model IDs. This is not programmatic and hinders automated workflows. The publishers.models.get endpoint is insufficient as it requires knowing the model ID beforehand and provides very limited information.

AWS Bedrock's ListFoundationModels API provides a good example of the desired functionality. It allows users to programmatically discover available models and their key characteristics, which is crucial for selecting the right model for a particular task and integrating models into applications.

Other information (workarounds you have tried, documentation consulted, etc):

I have consulted the following documentation:

Vertex AI publishers.models.get https://cloud.google.com/vertex-ai/docs/reference/rest/v1/publishers.models/get

Currently, the only workaround is manually inspecting the Model Garden in the console, which is not scalable or suitable for integration into automated processes.

The availability of a comprehensive model listing API will greatly enhance the developer experience, enable more efficient model selection, and facilitate the development of AI-powered applications on Google Cloud.

Issue 391132286

Links (2)

Issue metadata