[Bugfix][Responses] Add tests for Chat Completions Responses API Render Parity (#50334)

Signed-off-by: Yifan Zong <yzong@redhat.com>
Co-authored-by: Ben Browning <56071+bbrowning@users.noreply.github.com>
This commit is contained in:
yzong-rh
2026-07-31 15:45:40 -04:00
committed by GitHub
co-authored by Ben Browning
parent 4fdd3e0b74
commit e8b358b4c0
8 changed files with 521 additions and 43 deletions
@@ -179,6 +179,7 @@ class ChatCompletionToolsParam(OpenAIBaseModel):
@model_serializer(mode="wrap")
def _serialize(self, handler):
data = handler(self)
data = {k: v for k, v in data.items() if k in type(self).model_fields}
if self.defer_loading is None:
data.pop("defer_loading", None)
return data
@@ -296,6 +296,7 @@ class FunctionDefinition(OpenAIBaseModel):
@model_serializer(mode="wrap")
def _serialize(self, handler):
data = handler(self)
data = {k: v for k, v in data.items() if k in type(self).model_fields}
if self.strict is None:
data.pop("strict", None)
if self.defer_loading is None:
+7 -1
View File
@@ -610,7 +610,13 @@ class OpenAIServingResponses(GenerateBaseServing):
request: ResponsesRequest,
prev_response: ResponsesResponse | None,
):
tool_dicts = construct_tool_dicts(request.tools, request.tool_choice)
tool_dicts = construct_tool_dicts(
request.tools,
request.tool_choice,
exclude_tools_when_tool_choice_none=(
self.online_renderer.exclude_tools_when_tool_choice_none
),
)
# Construct the input messages.
messages = construct_input_messages(
request_instructions=request.instructions,
+25 -19
View File
@@ -31,8 +31,11 @@ from openai.types.responses.tool import Tool
from vllm import envs
from vllm.entrypoints.chat_utils import make_tool_call_id
from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionMessageParam
from vllm.entrypoints.openai.engine.protocol import FunctionCall
from vllm.entrypoints.openai.chat_completion.protocol import (
ChatCompletionMessageParam,
ChatCompletionToolsParam,
)
from vllm.entrypoints.openai.engine.protocol import FunctionCall, FunctionDefinition
from vllm.entrypoints.openai.responses.protocol import ResponseInputOutputItem
from vllm.logger import init_logger
from vllm.tool_parsers.utils import (
@@ -363,27 +366,30 @@ def extract_tool_types(tools: list[Tool]) -> set[str]:
return tool_types
def convert_tool_responses_to_completions_format(tool: dict) -> dict:
def convert_tool_responses_to_completions_format(
tool: dict,
) -> ChatCompletionToolsParam:
"""
Convert a flat tool schema:
Convert a flat Responses tool schema:
{"type": "function", "name": "...", "description": "...", "parameters": {...}}
into:
{"type": "function", "function": {...}}
into a Chat Completions tool param for chat-template rendering.
"""
return {
"type": "function",
"function": tool,
}
return ChatCompletionToolsParam(
type="function",
function=FunctionDefinition.model_validate(
{k: v for k, v in tool.items() if k != "type"}
),
)
def construct_tool_dicts(
tools: list[Tool], tool_choice: ToolChoice
tools: list[Tool],
tool_choice: ToolChoice,
exclude_tools_when_tool_choice_none: bool = False,
) -> list[dict[str, Any]] | None:
if not tools or (tool_choice == "none"):
tool_dicts = None
else:
tool_dicts = [
convert_tool_responses_to_completions_format(tool)
for tool in iter_response_function_tool_dicts(tools)
]
return tool_dicts
if not tools or (tool_choice == "none" and exclude_tools_when_tool_choice_none):
return None
return [
convert_tool_responses_to_completions_format(tool).model_dump()
for tool in iter_response_function_tool_dicts(tools)
]