[roydonsequeira]

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

• Closes: Gemma 3 models fail with TypeError when a tool returns a datetime or a Pydantic model (e.g. load_memory) #7667

Problem:
Gemma and Gemma3Ollama render each function response as text for Gemma 3 with a bare json.dumps(func_response.response). A tool result is stored as the tool returned it, so a datetime, bytes or a Pydantic model raised TypeError on the next model call and failed the run. The built-in load_memory tool always hit this, because it returns a LoadMemoryResponse. The default ensure_ascii=True also sent non-ASCII text in tool results to the model as \uXXXX escapes.

Solution:
Serialize the payload with pydantic_core.to_jsonable_python, the way a Gemini request carries the same function response: unset model fields are dropped (as google-genai does with exclude_none=True), bytes are base64-encoded, and values Pydantic cannot serialize fall back to their str(). json.dumps(..., ensure_ascii=False) then keeps non-ASCII text as is. Payloads that already were plain JSON render exactly as before, so the existing tests are unchanged.

The change is in GemmaFunctionCallingMixin's content conversion, which both Gemma and Gemma3Ollama use. Gemma 4 uses native function calling and is not affected.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

test_preprocess_request_with_function_response_converts_values in tests/unittests/models/test_gemma_llm.py covers non-ASCII text, a datetime, bytes and a LoadMemoryResponse. All four cases fail on main (three with TypeError, one with escaped text) and pass with the fix.

$ pytest tests/unittests/models/test_gemma_llm.py
31 passed in 3.46s

Full suite on Windows 11, Python 3.11 (pytest tests/unittests -n 4): 73 failed, 20749 passed, 107 skipped. Every one of the 73 also fails on unmodified main in this environment, all outside models/ (Windows-specific failures such as test_cli_deploy_to_cloud_run.py and test_samples.py). test_gemma_llm.py also passes on Python 3.12, 3.13 and 3.14.

Manual End-to-End (E2E) Tests:

An Agent on Gemma3Ollama(model="ollama/gemma3:4b") (Ollama 0.40.7), run through Runner with InMemorySessionService. Each run prints the tool-result text that reaches the model.

Function tool whose result holds a datetime and non-ASCII text:

# main
user: Who is the guest on booking B-17, and when did they check in?
  tool call: get_booking({'booking_id': 'B-17'})
  FAILED: TypeError: Object of type datetime is not JSON serializable

# this change
user: Who is the guest on booking B-17, and when did they check in?
  tool call: get_booking({'booking_id': 'B-17'})
  [sent to model] Invoking tool `get_booking` produced: `{"booking_id": "B-17", "guest": "Zoë Müller", "hotel": "ホテル東京", "checked_in_at": "2026-10-22T09:30:00"}`.
concierge: Zoë Müller checked in on 2026-10-22 at 09:30:00.

Built-in load_memory with one remembered session in InMemoryMemoryService:

# main
user: Which city do I live in? Look it up in your memory.
  tool call: load_memory({'query': 'Which city do I live in?'})
  FAILED: TypeError: Object of type LoadMemoryResponse is not JSON serializable

# this change
user: Which city do I live in? Look it up in your memory.
  tool call: load_memory({'query': 'Which city do I live in?'})
  [sent to model] Invoking tool `load_memory` produced: `{"result": {"memories": [{"content": {"parts": [{"text": "I live in München and my favourite dish is 寿司."}], "role": "user"}, "custom_metadata": {}, "id": "f2eedd24-719f-497f-87be-78ba493f0797", "author": "user", "timestamp": "2026-10-22T11:14:28.255772"}]}}`.
assistant: I live in München.

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 or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules.

Additional context

The formatters, linters and repository checks from .pre-commit-config.yaml pass on both files, and mypy reports no errors in gemma_llm.py, as on main.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.