fix: resolve BuiltinTools to strings for vllm tool_call messages

When the result of a ToolCall gets passed back into vLLM for the model
to handle the tool call result (as is often the case in agentic
tool-calling workflows), we forgot to handle the case where
BuiltinTool calls are not string values but instead instances of the
BuiltinTool enum. This fixes that, properly converting those enums to
string values before trying to serialize them into an OpenAI chat
completion request to vLLM.

PR #1931 fixed a bug where we weren't passing these tool calling
results back into vLLM, but as a side-effect it created this
serialization bug when using BuiltinTools.

Resolves #2070

Signed-off-by: Ben Browning <bbrownin@redhat.com>

llama_stack/providers/utils/inference/openai_compat.py

@@ -532,12 +532,17 @@ async def convert_message_to_openai_dict(message: Message, download: bool = Fals
     if hasattr(message, "tool_calls") and message.tool_calls:
         result["tool_calls"] = []
         for tc in message.tool_calls:
+            # The tool.tool_name can be a str or a BuiltinTool enum. If
+            # it's the latter, convert to a string.
+            tool_name = tc.tool_name
+            if isinstance(tool_name, BuiltinTool):
+                tool_name = tool_name.value
             result["tool_calls"].append(
                 {
                     "id": tc.call_id,
                     "type": "function",
                     "function": {
-                        "name": tc.tool_name,
+                        "name": tool_name,
                         "arguments": tc.arguments_json if hasattr(tc, "arguments_json") else json.dumps(tc.arguments),
                     },
                 }

tests/unit/providers/utils/inference/test_openai_compat.py

@@ -8,7 +8,7 @@ import pytest
 
 from llama_stack.apis.common.content_types import TextContentItem
 from llama_stack.apis.inference.inference import CompletionMessage, UserMessage
-from llama_stack.models.llama.datatypes import StopReason, ToolCall
+from llama_stack.models.llama.datatypes import BuiltinTool, StopReason, ToolCall
 from llama_stack.providers.utils.inference.openai_compat import convert_message_to_openai_dict
 
 
@@ -41,3 +41,29 @@ async def test_convert_message_to_openai_dict_with_tool_call():
             {"id": "123", "type": "function", "function": {"name": "test_tool", "arguments": '{"foo": "bar"}'}}
         ],
     }
+
+
+@pytest.mark.asyncio
+async def test_convert_message_to_openai_dict_with_builtin_tool_call():
+    message = CompletionMessage(
+        content="",
+        tool_calls=[
+            ToolCall(
+                call_id="123",
+                tool_name=BuiltinTool.brave_search,
+                arguments_json='{"foo": "bar"}',
+                arguments={"foo": "bar"},
+            )
+        ],
+        stop_reason=StopReason.end_of_turn,
+    )
+
+    openai_dict = await convert_message_to_openai_dict(message)
+
+    assert openai_dict == {
+        "role": "assistant",
+        "content": [{"type": "text", "text": ""}],
+        "tool_calls": [
+            {"id": "123", "type": "function", "function": {"name": "brave_search", "arguments": '{"foo": "bar"}'}}
+        ],
+    }