Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Sign up
Appearance settings

feat: Add thought parts from litellm response #3006

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
saravanabalagi wants to merge 3 commits into google:main
base: main
Choose a base branch
Loading
from saravanabalagi:feat/add_reasoning_content_litellm

Conversation

@saravanabalagi
Copy link

@saravanabalagi saravanabalagi commented Sep 23, 2025

Deepseek and Qwen3 models produce thought / reasoning in

<think>Okay, the user said "hi". I need to...</think>
Hello! How can I assist you

Litellm already parses these think tags and the response messages contain these thought parts in the reasoning_content field. This can be parsed and added as ADK text part with thought True.

Further, this should automatically be displayed in UI and be treated the same as that of Gemini models

image

Fixes #1788 and #2982

bastiandg, jk-gan, amarrella, trongthanht3, and jackward-countdown reacted with thumbs up emoji
Copy link

google-cla bot commented Sep 23, 2025

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.

Copy link

Summary of Changes

Hello @saravanabalagi, 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 enhances the system's ability to process and display the internal reasoning or 'thought' processes of certain AI models, specifically Deepseek and Qwen3, when their responses are routed through LiteLLM. By extracting the reasoning_content and marking it as a distinct 'thought' part, the change aims to provide users with greater transparency into how these models arrive at their answers, ensuring a consistent user experience with other models like Gemini.

Highlights

  • Thought Part Integration: Implemented the parsing and integration of 'thought' or 'reasoning_content' from LiteLLM responses into the ADK's types.Part structure, explicitly marking them with thought=True.
  • Enhanced Model Output Display: Ensured that the reasoning steps from models like Deepseek and Qwen3 (when processed via LiteLLM) are captured and prepared for display in the UI, aligning with how Gemini models' thought processes are presented.
  • Bug Fixes: This change addresses and fixes issues QUESTION: It seems that LiteLLM does not handle the reasoning_content of the model. #1788 and Thought / Reasoning messages missing in events #2982 , likely related to the display or handling of model reasoning.
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

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. ↩

Copy link
Collaborator

adk-bot commented Sep 23, 2025

Hello @saravanabalagi, thank you for your contribution!

Before we can review your PR, please address the following items from our contribution guidelines:

  • Sign our Contributor License Agreement: It appears you have not yet signed the CLA. You can do so at https://cla.developers.google.com/.
  • Testing Plan: Please include a testing plan section in your PR description to describe how you tested your changes.
  • Unit Tests: Please add or update unit tests for your changes and include a summary of passed pytest results in your PR description.

You can find more details in our contribution guidelines.

Response from ADK Triaging Agent

@adk-bot adk-bot added bot triaged models [Component] Issues related to model support labels Sep 23, 2025
Copy link

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request introduces support for 'thought' parts from litellm responses by processing the reasoning_content field. The implementation is correct, but I have a suggestion to make the code more concise and efficient. More importantly, the new functionality is not covered by unit tests. I've added a comment highlighting the need to add a test case to ensure the new logic is working as expected and to prevent future regressions.

@saravanabalagi saravanabalagi force-pushed the feat/add_reasoning_content_litellm branch from 624fb1e to 651a69b Compare September 23, 2025 12:15
Copy link

This will probably work for most llms. It doesn't work on gpt-5 because thoughts are only emitted on the new responses api.

Hopefully this will be merged soon on litellms side, so this will also work for gpt-5: BerriAI/litellm#14765

Copy link

Hi @saravanabalagi: we are also encountering the same issue and noticed that your changes would only work for non-sse responses

I think we also need to make some changes to generate_content_async and _model_response_to_chunk to return and process a new ThoughtChunk so that thoughts are also passed downstream

Copy link
Author

Handled thought chunk in stream.

Note that the adk web UI does not handle streaming thought chunks, it creates new thought bubble for each chunk instead of adding to the same bubble. This bug was already mentioned in google/adk-web#93 with screencast exactly showing what I see.
abelatyou, iamtatsuki05, and amarrella reacted with thumbs up emoji

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

1 more reviewer

@gemini-code-assist gemini-code-assist[bot] gemini-code-assist[bot] left review comments

Reviewers whose approvals may not affect merge requirements

At least 1 approving review is required to merge this pull request.

Assignees

No one assigned

Labels

models [Component] Issues related to model support

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

QUESTION: It seems that LiteLLM does not handle the reasoning_content of the model.

AltStyle によって変換されたページ (->オリジナル) /