fix: 🐛 forward labels parameter to Vertex AI API requests

[miguelangelmorenochacon]

Problem

The labels parameter was not being forwarded to Vertex AI API requests, preventing proper tracking of Google partnership labels for billing and analytics purposes.

This issue prevented users from passing custom labels through the LiteLLM proxy to Vertex AI, which is important for resource tracking and billing attribution in Google Cloud environments.

Solution

This PR adds support for forwarding the labels parameter to Vertex AI API requests:

1. Type Definition: Added labels field to the RequestBody TypedDict in litellm/types/llms/vertex_ai.py

2. Request Transformation: Modified _transform_request_body in litellm/llms/vertex_ai/gemini/transformation.py to:

   • Extract labels from optional parameters
   • Support both direct labels parameter and conversion from OpenAI-style metadata.requester_metadata
   • Only add labels to Vertex AI endpoints (excluded from Google GenAI/AI Studio endpoints)
   • Properly include labels in the request body sent to the API

Testing

The changes have been tested locally with Docker, verifying that:

• Labels are correctly included in POST requests to Vertex AI
• Labels are properly formatted: "labels": {"goog-sample": "test-value"}
• Labels work with both direct parameter and metadata conversion
• Labels are excluded from Google GenAI endpoints as expected

Usage Example

Configuration

model_list:
  - model_name: gemini-flash
    litellm_params:
      model: vertex_ai/gemini-2.5-flash
      vertex_project: "your-vertex-project-id"
      vertex_location: "your-vertex-location"
      labels:
        goog-partner: your-value
      allowed_openai_params: ["labels"]

litellm_settings:
  drop_params: True
  set_verbose: True
  json_logs: True

general_settings:
  master_key: "sk-1234"

Expected Output

When using this configuration, labels are correctly forwarded to Vertex AI:

POST Request Sent from LiteLLM:
curl -X POST \
https://us-east1-aiplatform.googleapis.com/v1/projects/your-project/locations/your-location/publishers/google/models/gemini-2.5-flash:generateContent \
-H 'Content-Type: application/json' -H 'Authorization: Bearer ***' \
-d '{"contents": [{"role": "user", "parts": [{"text": "test"}]}], "generationConfig": {}, "labels": {"goog-partner": "your-value"}}'

Note: The labels parameter is now successfully included in the request body.

Changes

Modified Files

• litellm/types/llms/vertex_ai.py - Added labels field to RequestBody type
• litellm/llms/vertex_ai/gemini/transformation.py - Added labels extraction and forwarding logic

Related Issues

Fixes #13692

Additional Context

This implementation follows Google Cloud's labels specification and ensures backward compatibility by:

• Making labels optional
• Only applying labels to Vertex AI endpoints (not Google GenAI)
• Supporting both direct labels parameter and metadata conversion

[vercel]

@miguelangelmorenochacon is attempting to deploy a commit to the CLERKIEAI Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.