[Frontend] Make finish_reason and Responses item status truthful (#59324)
Signed-off-by: Flora Feng <4florafeng@gmail.com> Signed-off-by: sfeng33 <4florafeng@gmail.com> Co-authored-by: Chauncey <chaunceyjiang@gmail.com>
This commit is contained in:
@@ -4,6 +4,7 @@ from typing import TypeVar
|
||||
|
||||
from vllm.entrypoints.generate.base.protocol import DeltaMessage, FunctionCall
|
||||
from vllm.entrypoints.openai.chat_completion.protocol import (
|
||||
ChatCompletionNamedToolChoiceParam,
|
||||
ChatCompletionRequest,
|
||||
ChatCompletionResponseChoice,
|
||||
ChatCompletionResponseStreamChoice,
|
||||
@@ -41,3 +42,24 @@ def maybe_filter_parallel_tool_calls(
|
||||
tool_call for tool_call in delta.tool_calls if tool_call.index == 0
|
||||
]
|
||||
return item
|
||||
|
||||
|
||||
def resolve_finish_reason(
|
||||
engine_finish_reason: str | None,
|
||||
request: ChatCompletionRequest | None,
|
||||
tool_calls_made: bool,
|
||||
) -> str:
|
||||
"""Map the engine finish reason to the OpenAI chat ``finish_reason``.
|
||||
|
||||
``tool_calls`` is reported only when the choice actually carries tool
|
||||
calls, the engine stopped on its own (a ``length`` cut stays visible so
|
||||
clients can tell a truncated call from a complete one) and the caller did
|
||||
not force a single named tool, for which OpenAI reports ``stop``.
|
||||
"""
|
||||
finish_reason = engine_finish_reason or "stop"
|
||||
forced_named_tool = request is not None and isinstance(
|
||||
request.tool_choice, ChatCompletionNamedToolChoiceParam
|
||||
)
|
||||
if tool_calls_made and finish_reason == "stop" and not forced_named_tool:
|
||||
return "tool_calls"
|
||||
return finish_reason
|
||||
|
||||
Reference in New Issue
Block a user