fix(responses): normalize chat tool_choice for completions→responses bridge

[Sameerlite]

When routing chat completions to the Responses API, OpenAI rejects chat-style tool_choice ({type, function: {name}}) with unknown_parameter: tool_choice.function.

• Add LiteLLMResponsesTransformationHandler._normalize_tool_choice_for_responses_api and apply it in _map_optional_params_to_responses_api_request for tool_choice.
• Regression test: test_map_optional_params_tool_choice_chat_nested_to_responses_api.

Curl in litellm

Fixes #27611
Fixes LIT-2975

---

Note

Low Risk
Low risk: small request-shaping change limited to tool_choice mapping with targeted regression tests; main risk is unintended normalization of uncommon tool_choice dict shapes.

Overview
Fixes a Responses API compatibility issue when routing Chat Completions requests by normalizing tool_choice from chat-style {type:"function", function:{name}} to the Responses API shape {type:"function", name} and stripping the stray function key before sending upstream.

Adds regression coverage to ensure nested chat tool_choice is converted correctly and that non-function/standard values (e.g., auto, none, required) pass through unchanged.

^{Reviewed by Cursor Bugbot for commit 8dd31f2. Bugbot is set up for automated code reviews on this repo. Configure here.}

[greptile-apps]

Greptile Summary

This PR fixes a unknown_parameter: tool_choice.function rejection from OpenAI's Responses API by normalizing chat-style tool_choice dicts ({type, function: {name}}) to the Responses API shape ({type, name}) in the completions→responses bridge.

• Introduces _normalize_tool_choice_for_responses_api which handles all four cases: string pass-throughs ("auto", "none"), already-normalized dicts, chat-style nested dicts, and mixed dicts with both top-level name and a stray function sub-key — the returned value always omits the function key.
• Wires the normalizer into _map_optional_params_to_responses_api_request and backs it with a new integration test plus a parametrized unit test covering all edge cases.

Confidence Score: 5/5

Safe to merge — the change is narrowly scoped to the tool_choice normalization step in the completions→responses bridge and does not touch any auth, routing, or request-path logic.

The normalization logic is straightforward, the returned dict is always reconstructed from scratch (no stray keys leak through), and the parametrized test now covers every branch including strings, already-normalized dicts, mixed-key inputs, and non-function types. No regressions are introduced to existing paths.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/completion_extras/litellm_responses_transformation/transformation.py	Adds _normalize_tool_choice_for_responses_api to strip the chat-style function sub-key and promote function.name to top-level; applies it in _map_optional_params_to_responses_api_request. Logic is correct: string pass-throughs, already-normalized dicts, and mixed-format dicts are all handled, and the returned value always omits stray function keys.
tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py	Adds two new tests: an integration test for the full _map_optional_params_to_responses_api_request path and a parametrized unit test for _normalize_tool_choice_for_responses_api covering strings, already-normalized dicts, mixed-key dicts, and non-function types. No real network calls are made.

_{Reviews (2): Last reviewed commit: "fix(responses): strip tool_choice.functi..." | Re-trigger Greptile}

[codecov]

Codecov Report

❌ Patch coverage is 92.30769% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...litellm_responses_transformation/transformation.py	92.30%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[Sameerlite]

@greptile re review

[Sameerlite]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 8dd31f2. Configure here.}

[mateo-berri]

LGTM; thanks!