[Frontend][responsesAPI][1/n] convert responses API tool input to chat completions tool format (#28231)

Signed-off-by: Andrew Xia <axia@fb.com>
Co-authored-by: Andrew Xia <axia@fb.com>
Co-authored-by: Chauncey <chaunceyjiang@gmail.com>

tests/entrypoints/test_responses_utils.py

@@ -0,0 +1,30 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+from vllm.entrypoints.responses_utils import (
+    convert_tool_responses_to_completions_format,
+)
+
+
+class TestResponsesUtils:
+    """Tests for convert_tool_responses_to_completions_format function."""
+
+    def test_convert_tool_responses_to_completions_format(self):
+        """Test basic conversion of a flat tool schema to nested format."""
+        input_tool = {
+            "type": "function",
+            "name": "get_weather",
+            "description": "Get the current weather in a given location",
+            "parameters": {
+                "type": "object",
+                "properties": {
+                    "location": {"type": "string"},
+                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
+                },
+                "required": ["location", "unit"],
+            },
+        }
+
+        result = convert_tool_responses_to_completions_format(input_tool)
+
+        assert result == {"type": "function", "function": input_tool}

tests/v1/entrypoints/openai/serving_responses/conftest.py

@@ -30,7 +30,10 @@ def server_with_store(default_server_args):
     with RemoteOpenAIServer(
         MODEL_NAME,
         default_server_args,
-        env_dict={"VLLM_ENABLE_RESPONSES_API_STORE": "1"},
+        env_dict={
+            "VLLM_ENABLE_RESPONSES_API_STORE": "1",
+            "VLLM_SERVER_DEV_MODE": "1",
+        },
     ) as remote_server:
         yield remote_server
 

tests/v1/entrypoints/openai/serving_responses/test_function_call.py

@@ -116,6 +116,7 @@ async def test_function_tool_use(
         input=prompt,
         tools=tools,
         tool_choice=tool_choice,
+        temperature=0.0,
     )
 
     assert len(response.output) >= 1

vllm/entrypoints/openai/serving_responses.py

@@ -48,7 +48,6 @@ from openai.types.responses.response_output_text import Logprob, LogprobTopLogpr
 from openai.types.responses.response_reasoning_item import (
     Content as ResponseReasoningTextContent,
 )
-from openai.types.responses.tool import Tool
 from openai_harmony import Message as OpenAIHarmonyMessage
 
 from vllm import envs
@@ -94,7 +93,11 @@ from vllm.entrypoints.openai.protocol import (
 )
 from vllm.entrypoints.openai.serving_engine import OpenAIServing
 from vllm.entrypoints.openai.serving_models import OpenAIServingModels
-from vllm.entrypoints.responses_utils import construct_chat_message_with_tool_call
+from vllm.entrypoints.responses_utils import (
+    construct_chat_message_with_tool_call,
+    convert_tool_responses_to_completions_format,
+    extract_tool_types,
+)
 from vllm.entrypoints.tool_server import ToolServer
 from vllm.inputs.data import TokensPrompt as EngineTokensPrompt
 from vllm.logger import init_logger
@@ -108,23 +111,6 @@ from vllm.utils import random_uuid
 logger = init_logger(__name__)
 
 
-def extract_tool_types(tools: list[Tool]) -> set[str]:
-    """
-    Extracts the tool types from the given tools.
-    """
-    tool_types: set[str] = set()
-    for tool in tools:
-        if tool.type == "mcp":
-            # Allow the MCP Tool type to enable built in tools if the
-            # server_label is allowlisted in
-            # envs.VLLM_GPT_OSS_SYSTEM_TOOL_MCP_LABELS
-            if tool.server_label in envs.VLLM_GPT_OSS_SYSTEM_TOOL_MCP_LABELS:
-                tool_types.add(tool.server_label)
-        else:
-            tool_types.add(tool.type)
-    return tool_types
-
-
 class OpenAIServingResponses(OpenAIServing):
     def __init__(
         self,
@@ -513,7 +499,10 @@ class OpenAIServingResponses(OpenAIServing):
         ):
             tool_dicts = None
         else:
-            tool_dicts = [tool.model_dump() for tool in request.tools]
+            tool_dicts = [
+                convert_tool_responses_to_completions_format(tool.model_dump())
+                for tool in request.tools
+            ]
         # Construct the input messages.
         messages = self._construct_input_messages(request, prev_response)
         _, request_prompts, engine_prompts = await self._preprocess_chat(

vllm/entrypoints/responses_utils.py

@@ -10,7 +10,9 @@ from openai.types.chat.chat_completion_message_tool_call_param import (
     Function as FunctionCallTool,
 )
 from openai.types.responses import ResponseFunctionToolCall
+from openai.types.responses.tool import Tool
 
+from vllm import envs
 from vllm.entrypoints.openai.protocol import (
     ChatCompletionMessageParam,
     ResponseInputOutputItem,
@@ -43,3 +45,33 @@ def construct_chat_message_with_tool_call(
             tool_call_id=item.get("call_id"),
         )
     return item  # type: ignore
+
+
+def extract_tool_types(tools: list[Tool]) -> set[str]:
+    """
+    Extracts the tool types from the given tools.
+    """
+    tool_types: set[str] = set()
+    for tool in tools:
+        if tool.type == "mcp":
+            # Allow the MCP Tool type to enable built in tools if the
+            # server_label is allowlisted in
+            # envs.VLLM_GPT_OSS_SYSTEM_TOOL_MCP_LABELS
+            if tool.server_label in envs.VLLM_GPT_OSS_SYSTEM_TOOL_MCP_LABELS:
+                tool_types.add(tool.server_label)
+        else:
+            tool_types.add(tool.type)
+    return tool_types
+
+
+def convert_tool_responses_to_completions_format(tool: dict) -> dict:
+    """
+    Convert a flat tool schema:
+        {"type": "function", "name": "...", "description": "...", "parameters": {...}}
+    into:
+        {"type": "function", "function": {...}}
+    """
+    return {
+        "type": "function",
+        "function": tool,
+    }