Link to Issue

• Closes: LiteLlm drops Gemini grounding_metadata: _model_response_to_generate_content_response ignores response.vertex_ai_grounding_metadata #5659

Problem

LiteLLM exposes Gemini's grounding metadata on ModelResponse.vertex_ai_grounding_metadata rather than inside the message, and _model_response_to_generate_content_response in lite_llm.py never reads it. As a result LlmResponse.grounding_metadata is always None when Gemini is called through LiteLlm, breaking event.grounding_metadata, after_model_callback, and citation pipelines for the entire LiteLlm path. The native Gemini() path picks grounding up from candidate.grounding_metadata (llm_response.py:190).

Solution

Pull vertex_ai_grounding_metadata off the ModelResponse after the message is converted, and attach it to LlmResponse.grounding_metadata. Three shapes are accepted:

• types.GroundingMetadata instance → used as-is.
• dict → validated via types.GroundingMetadata.model_validate.
• list (one entry per candidate) → first entry, then the same dispatch.

If validation fails the handler logs a warning and leaves grounding unset, so a single malformed payload doesn't break the rest of the response.

Testing Plan

Unit Tests in tests/unittests/models/test_litellm.py:

• test_model_response_to_generate_content_response_grounding_metadata_dict — dict payload is propagated.
• test_model_response_to_generate_content_response_grounding_metadata_list — list payload uses the first entry.
• test_model_response_to_generate_content_response_no_grounding_metadata — missing attribute leaves grounding as None (no regression for non-Gemini LiteLlm models).

Full test_litellm.py suite passes locally:

259 passed in 9.36s

Manual E2E: Live Gemini API verification was not possible locally (no API key configured). The unit tests cover the conversion function the bug report points to.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end. (See above — requires live Gemini access.)