Unsupported patternProperties for json schema

[quimey]

Hey! I'm trying to use Gemini Flash 1.5 8b to run a prompt that includes a json schema for a structured output. This is my schema:

        "target_locations": {
            "type": "object",
            "patternProperties": {"^.*$": {"type": "number"}},
            "additionalProperties": False,
        },

The idea is to include a variable number of properties with numeric value.

This schema works well with langchain and anthropic models but when I try to use ChatGoogleGenerativeAI I get the following error:

celery-worker-1     | langchain_google_genai.chat_models.ChatGoogleGenerativeAIError: Invalid argument provided to Gemini: 400 * GenerateContentRequest.model: unexpected model name format
celery-worker-1     | * GenerateContentRequest.tools[0].function_declarations[0].parameters.properties[target_locations].properties: should be non-empty for OBJECT type
celery-worker-1     | 

Is there any workaround or a plan to support patternProperties?

I'm using python3.11 and

langchain==0.3.8
langchain-core=0.3.21
langchain-anthropic=0.3.0
langchain-google-genai=2.0.5

[Cristy94]

2.0 doesn't support it either, so this doesn't work (number keys):

"patternProperties": {
                "^[0-9]+$": {
                    "type": "array",
                    ...

Any way around it? I can probably write some JS to manually create the schema with the expected returned keys...