fix(gemini): add missing role="user" to function_response (refs [#206…

[ZqinKing]

(#20690))

Relevant issues

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 8, 2026 5:39am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR adjusts the Vertex/Gemini message transformation so that standalone tool/function response turns are emitted as ContentType(role="user", parts=...) instead of omitting the role.

The change is localized to _gemini_convert_messages_with_history() in litellm/llms/vertex_ai/gemini/transformation.py, which is responsible for converting OpenAI-style chat messages (including tool call results) into the contents list sent to Vertex AI / Gemini generateContent APIs.

Confidence Score: 4/5

• This PR is low-risk and localized, but should not be merged without a regression test for the fixed tool/function response formatting.
• The code change is minimal (adds an explicit role on two ContentType appends) and aligns with the ContentType schema, so runtime impact is likely positive. The main merge risk is lack of automated coverage for the specific failure mode reported in [Bug]: Gemini function_response missing role="user" causes INVALID_ARGUMENT #20690, so regressions could reappear unnoticed.
• litellm/llms/vertex_ai/gemini/transformation.py (add test coverage for tool/function response roles)

Important Files Changed

Filename	Overview
litellm/llms/vertex_ai/gemini/transformation.py	Ensures tool/function response content blocks are emitted with role="user" when converting OpenAI messages to Gemini/Vertex contents; no regressions spotted in the transformation logic, but the PR needs a unit test to cover this formatting requirement.

Sequence Diagram

sequenceDiagram
  participant Client as LiteLLM Client
  participant Tx as gemini/transformation.py
  participant Vertex as Vertex/Gemini API

  Client->>Tx: OpenAI-format messages
  Note over Tx: _gemini_convert_messages_with_history()
  Tx->>Tx: Convert user/system messages -> Content(role="user")
  Tx->>Tx: Convert assistant messages -> Content(role="model")
  Tx->>Tx: Convert tool/function result -> Part(function_response)
  Tx->>Tx: Append tool/function response Content(role="user", parts=[...])
  Tx->>Vertex: RequestBody(contents=[Content...])
  Vertex-->>Tx: generateContent response
  Tx-->>Client: Mapped response

Loading

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Missing regression test

This change fixes Vertex/Gemini request formatting by adding role="user" when appending function_response parts, but the PR doesn’t add a unit test to lock the behavior. Please add a test under tests/litellm/ that builds a message sequence with an assistant tool call followed by a tool/function role message, then asserts the transformed contents includes a ContentType with role == "user" for that function response turn.

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}

[krrishdholakia]

Fix merged into staging.