Fix/litellm finish reason 3109

[lizzij]

This PR supersedes and fixes #3114. It includes all commits from the original PR and adds a new commit to fix formatting.

Closes #3114.

Link to Issue or Description of Change

• Closes: The LiteLLM wrapper does not propagate finish_reason from LiteLLM responses #3109

2. Or, if no issue exists, describe the change:

If applicable, please follow the issue templates to provide as much detail as
possible.

Problem:
This PR ensures that the finish_reason field from LiteLLM responses is properly propagated to LlmResponse objects, enabling after_model_callback functions to detect completion conditions like max_tokens truncation.

Solution:
This fix allows after_model_callback functions to properly detect:

• "length": max_tokens limit reached
• "stop": natural completion
• "tool_calls": tool invocations
• "content_filter": filtered content

This enables implementing retry logic for incomplete responses, logging completion statistics, and handling different completion conditions appropriately.

Testing Plan

Unit Tests:

✅ 53/53 tests in test_litellm.py pass
✅ All 4 new finish_reason tests pass
✅ No existing tests broken

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.

[gemini-code-assist]

Summary of Changes

Hello @lizzij, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request addresses a critical gap in how LiteLLM responses handle completion reasons. By correctly propagating and standardizing the finish_reason field, it significantly enhances the ability of downstream functions to interpret model outputs. This improvement is crucial for implementing advanced logic like retry mechanisms for incomplete responses, detailed logging of completion statistics, and appropriate handling of diverse model termination conditions, ultimately leading to more reliable and intelligent LLM interactions.

Highlights

• LiteLLM finish_reason Propagation: Ensures that the finish_reason field from LiteLLM responses is correctly extracted and mapped to the LlmResponse object's finish_reason attribute, using the types.FinishReason enum.
• Enhanced Callback Functionality: Allows after_model_callback functions to accurately detect various completion conditions such as max_tokens truncation, natural completion, tool invocations, and content filtering, enabling more robust post-response logic.
• Standardized finish_reason Handling: Introduces a mapping (_FINISH_REASON_MAPPING) to convert LiteLLM's string-based finish_reason values to the standardized types.FinishReason enum, ensuring consistency across different model providers and improving type safety.
• Updated Telemetry and Documentation: Adjusts the telemetry tracing to correctly log the standardized finish_reason enum values and updates the LlmResponse docstring to clarify that finish_reason will always be a types.FinishReason enum.
• Comprehensive Unit Testing: Adds new unit tests to validate the proper propagation and mapping of various finish_reason types, including length, stop, tool_calls, content_filter, and ensures unknown reasons default to FinishReason.OTHER.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request correctly propagates the finish_reason from LiteLLM responses to LlmResponse objects for non-streaming calls. It introduces a mapping from LiteLLM's string reasons to the FinishReason enum, handles unknown reasons gracefully, and updates telemetry and docstrings accordingly. The accompanying tests are thorough for the non-streaming case. However, the same logic for propagating finish_reason has not been applied to streaming responses in generate_content_async, which means the fix is incomplete. This should be addressed to ensure consistent behavior across both streaming and non-streaming modes.

[xuanyang15]

Hi @lizzij, is this a duplication of #3114? Could you push to original author's branch so that your change will update the original PR instead of creating a new PR?

[lizzij]

Hi @lizzij, is this a duplication of #3114? Could you push to original author's branch so that your change will update the original PR instead of creating a new PR?

I don't have write access to google/adk-python and can't directly update original author's branch.

[xuanyang15]

@lizzij I see, thanks for the helping update the change! Do you have a screenshot for the manual tests?

[lizzij]

@lizzij I see, thanks for the helping update the change! Do you have a screenshot for the manual tests?

[GWeale]

Thank you for your contribution, added to the codebase!