Fix: Prevent infinite tool calling loop with Meta Llama models

Meta Llama models continue calling tools even after receiving results when
tools are present in the API request. This causes infinite loops in agentic
applications.

This fix automatically sets tool_choice='none' when ToolMessages are detected
in the conversation history, signaling to the model that it should generate
a final response instead of making additional tool calls.

Fixes infinite loop issue with Meta Llama 4 Scout and other Meta models
when using bind_tools() in LangGraph agents.

Tested with 3 different tools - all correctly stop after receiving results.

Signed-off-by: Federico Kamelhar <federico.kamelhar@oracle.com>

Author: fede-kamel

libs/oci/langchain_oci/chat_models/oci_generative_ai.py

@@ -736,11 +736,21 @@ def messages_to_oci_params(
                 oci_message = self.oci_chat_message[role](content=content)
             oci_messages.append(oci_message)
 
-        return {
+        result = {
             "messages": oci_messages,
             "api_format": self.chat_api_format,
         }
 
+        # BUGFIX: If tool results have been received and tools are bound, set tool_choice to "none"
+        # to prevent the model from making additional tool calls in a loop.
+        # This addresses a known issue with Meta Llama models that continue calling tools
+        # even after receiving results.
+        has_tool_results = any(isinstance(msg, ToolMessage) for msg in messages)
+        if has_tool_results and "tools" in kwargs and "tool_choice" not in kwargs:
+            result["tool_choice"] = self.oci_tool_choice_none()
+
+        return result
+
     def _process_message_content(
         self, content: Union[str, List[Union[str, Dict]]]
     ) -> List[Any]:

libs/oci/tests/unit_tests/chat_models/test_oci_generative_ai.py

@@ -746,3 +746,62 @@ def test_get_provider():
             ChatOCIGenAI(model_id=model_id)._provider.__class__.__name__
             == provider_name
         )
+
+
+@pytest.mark.requires("oci")
+def test_tool_choice_none_after_tool_results() -> None:
+    """Test that tool_choice is set to 'none' when ToolMessages are present.
+
+    This prevents infinite loops with Meta Llama models that continue calling
+    tools even after receiving results when tools are bound to the model.
+    """
+    from langchain_core.messages import ToolMessage
+    from oci.generative_ai_inference import models
+
+    oci_gen_ai_client = MagicMock()
+    llm = ChatOCIGenAI(
+        model_id="meta.llama-3.3-70b-instruct",
+        client=oci_gen_ai_client
+    )
+
+    # Mock tools
+    mock_tools = [
+        models.Tool(
+            type="FUNCTION",
+            function=models.FunctionDefinition(
+                name="get_weather",
+                description="Get weather for a city",
+                parameters={}
+            )
+        )
+    ]
+
+    # Bind tools to model
+    llm_with_tools = llm.bind_tools(mock_tools)
+
+    # Create conversation with ToolMessage
+    messages = [
+        HumanMessage(content="What's the weather?"),
+        AIMessage(
+            content="",
+            tool_calls=[{
+                "id": "call_123",
+                "name": "get_weather",
+                "args": {"city": "Chicago"}
+            }]
+        ),
+        ToolMessage(
+            content="Sunny, 65°F",
+            tool_call_id="call_123"
+        )
+    ]
+
+    # Prepare the request
+    request = llm_with_tools._prepare_request(messages, stream=False)
+
+    # Verify that tool_choice is set to 'none'
+    assert hasattr(request.chat_request, 'tool_choice')
+    assert isinstance(request.chat_request.tool_choice, models.ToolChoiceNone)
+    # Verify tools are still present (not removed, just choice is 'none')
+    assert hasattr(request.chat_request, 'tools')
+    assert len(request.chat_request.tools) > 0