fix: support tool_choice with specific tool names in LiteLLM streaming (fixes #1846)

This change fixes a Pydantic validation error that occurred when using
LiteLLM with streaming enabled and specifying a specific tool name for
tool_choice parameter.

Problem:
When users specified tool_choice="my_tool_name" with streaming enabled,
the SDK would incorrectly cast it to Literal["auto", "required", "none"],
causing a Pydantic validation error.

The issue was in litellm_model.py line 376, where the Response object was
created with an incorrect type cast:
  tool_choice=cast(Literal["auto", "required", "none"], tool_choice)

However, tool_choice can be:
- A Literal: "auto", "required", "none"
- A ChatCompletionNamedToolChoiceParam dict with specific tool name
- The Converter.convert_tool_choice() already handles string tool names

Solution:
- Import ToolChoiceFunction from openai.types.responses
- Properly convert ChatCompletionNamedToolChoiceParam to ToolChoiceFunction
- Handle all valid tool_choice types when creating Response object

The fix ensures that when tool_choice is a dict like:
  {"type": "function", "function": {"name": "my_tool"}}
It gets correctly converted to:
  ToolChoiceFunction(type="function", name="my_tool")

Testing:
- Linting (ruff check) - passed
- Type checking (mypy) - passed
- Formatting (ruff format) - passed

Generated with Lucas Wang<[email protected]>

Co-Authored-By: Claude <[email protected]>

Author: gn00295120

src/agents/extensions/models/litellm_model.py

@@ -24,6 +24,7 @@
     ChatCompletionMessageCustomToolCall,
     ChatCompletionMessageFunctionToolCall,
     ChatCompletionMessageParam,
+    ChatCompletionNamedToolChoiceParam,
 )
 from openai.types.chat.chat_completion_message import (
     Annotation,
@@ -32,6 +33,7 @@
 )
 from openai.types.chat.chat_completion_message_function_tool_call import Function
 from openai.types.responses import Response
+from openai.types.responses.tool_choice_function import ToolChoiceFunction
 
 from ... import _debug
 from ...agent_output import AgentOutputSchemaBase
@@ -367,15 +369,39 @@ async def _fetch_response(
         if isinstance(ret, litellm.types.utils.ModelResponse):
             return ret
 
+        # Convert tool_choice to the correct type for Response
+        # tool_choice can be a Literal, a ChatCompletionNamedToolChoiceParam, or omit
+        response_tool_choice: Literal["auto", "required", "none"] | ToolChoiceFunction
+        if tool_choice is omit:
+            response_tool_choice = "auto"
+        elif isinstance(tool_choice, dict):
+            # Convert from ChatCompletionNamedToolChoiceParam to ToolChoiceFunction
+            # The dict has structure: {"type": "function", "function": {"name": "tool_name"}}
+            func_data = tool_choice.get("function")
+            if (
+                tool_choice.get("type") == "function"
+                and func_data is not None
+                and isinstance(func_data, dict)
+            ):
+                response_tool_choice = ToolChoiceFunction(
+                    type="function", name=func_data["name"]
+                )
+            else:
+                # Fallback to auto if unexpected format
+                response_tool_choice = "auto"
+        elif tool_choice in ("auto", "required", "none"):
+            response_tool_choice = tool_choice  # type: ignore
+        else:
+            # Fallback to auto for any other case
+            response_tool_choice = "auto"
+
         response = Response(
             id=FAKE_RESPONSES_ID,
             created_at=time.time(),
             model=self.model,
             object="response",
             output=[],
-            tool_choice=cast(Literal["auto", "required", "none"], tool_choice)
-            if tool_choice is not omit
-            else "auto",
+            tool_choice=response_tool_choice,
             top_p=model_settings.top_p,
             temperature=model_settings.temperature,
             tools=[],