Fix: Ensure that tool calls with no arguments get handled correctly #3560

When a model decides to use an MCP tool call that requires no arguments, it sets the arguments field to None. This causes validation errors because this field gets removed when being parsed by an openai compatible inference provider like vLLM
This PR ensures that, as soon as the tool call args are accumulated while streaming, we check to ensure no tool call function arguments are set to None - if they are we replace them with "{}"

Closes #3456

Added new unit test to verify that any tool calls with function arguments set to None get handled correctly

Signed-off-by: Jaideep Rao <[email protected]>

Author: jaideepr97

llama_stack/providers/inline/agents/meta_reference/responses/streaming.py

@@ -298,8 +298,11 @@ async def _process_streaming_chunks(
 
         # Emit arguments.done events for completed tool calls (differentiate between MCP and function calls)
         for tool_call_index in sorted(chat_response_tool_calls.keys()):
+            tool_call = chat_response_tool_calls[tool_call_index]
+            # Ensure that arguments, if sent back to the inference provider, are not None
+            tool_call.function.arguments = tool_call.function.arguments or "{}"
             tool_call_item_id = tool_call_item_ids[tool_call_index]
-            final_arguments = chat_response_tool_calls[tool_call_index].function.arguments or ""
+            final_arguments = tool_call.function.arguments
             tool_call_name = chat_response_tool_calls[tool_call_index].function.name
 
             # Check if this is an MCP tool call

tests/unit/providers/agents/meta_reference/test_openai_responses.py

@@ -326,6 +326,85 @@ async def fake_stream_toolcall():
     assert chunks[5].response.output[0].name == "get_weather"
 
 
+async def test_create_openai_response_with_tool_call_function_arguments_none(openai_responses_impl, mock_inference_api):
+    """Test creating an OpenAI response with a tool call response that has a function with arguments set to None."""
+    # Setup
+    input_text = "How hot it is in San Francisco today?"
+    model = "meta-llama/Llama-3.1-8B-Instruct"
+
+    async def fake_stream_toolcall():
+        yield ChatCompletionChunk(
+            id="123",
+            choices=[
+                Choice(
+                    index=0,
+                    delta=ChoiceDelta(
+                        tool_calls=[
+                            ChoiceDeltaToolCall(
+                                index=0,
+                                id="tc_123",
+                                function=ChoiceDeltaToolCallFunction(name="get_weather", arguments=None),
+                                type=None,
+                            )
+                        ]
+                    ),
+                ),
+            ],
+            created=1,
+            model=model,
+            object="chat.completion.chunk",
+        )
+
+    mock_inference_api.openai_chat_completion.return_value = fake_stream_toolcall()
+
+    # Execute
+    result = await openai_responses_impl.create_openai_response(
+        input=input_text,
+        model=model,
+        stream=True,
+        temperature=0.1,
+        tools=[
+            OpenAIResponseInputToolFunction(
+                name="get_weather",
+                description="Get current temperature for a given location.",
+                parameters={
+                    "location": "string",
+                },
+            )
+        ],
+    )
+
+    # Check that we got the content from our mocked tool execution result
+    chunks = [chunk async for chunk in result]
+
+    # Verify event types
+    # Should have: response.created, output_item.added, function_call_arguments.delta,
+    # function_call_arguments.done, output_item.done, response.completed
+    assert len(chunks) == 5
+
+    # Verify inference API was called correctly (after iterating over result)
+    first_call = mock_inference_api.openai_chat_completion.call_args_list[0]
+    assert first_call.kwargs["messages"][0].content == input_text
+    assert first_call.kwargs["tools"] is not None
+    assert first_call.kwargs["temperature"] == 0.1
+
+    # Check response.created event (should have empty output)
+    assert chunks[0].type == "response.created"
+    assert len(chunks[0].response.output) == 0
+
+    # Check streaming events
+    assert chunks[1].type == "response.output_item.added"
+    assert chunks[2].type == "response.function_call_arguments.done"
+    assert chunks[3].type == "response.output_item.done"
+
+    # Check response.completed event (should have the tool call with arguments set to "{}")
+    assert chunks[4].type == "response.completed"
+    assert len(chunks[4].response.output) == 1
+    assert chunks[4].response.output[0].type == "function_call"
+    assert chunks[4].response.output[0].name == "get_weather"
+    assert chunks[4].response.output[0].arguments == "{}"
+
+
 async def test_create_openai_response_with_multiple_messages(openai_responses_impl, mock_inference_api):
     """Test creating an OpenAI response with multiple messages."""
     # Setup