[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:
Flora Feng
2026-10-15 10:12:02 +01:00
committed by GitHub
co-authored by Chauncey
parent 427915bed2
commit 80dd705c1d
14 changed files with 421 additions and 62 deletions
+24 -6
View File
@@ -45,6 +45,7 @@ from vllm.entrypoints.openai.responses.protocol import (
OutputTokensDetails,
ResponseCompletedEvent,
ResponseCreatedEvent,
ResponseIncompleteEvent,
ResponseInProgressEvent,
ResponseInputOutputItem,
ResponseInputOutputMessage,
@@ -66,6 +67,7 @@ from vllm.entrypoints.openai.responses.utils import (
build_response_output_items,
extract_function_tool_names,
extract_tool_types,
response_item_status,
reuse_streamed_item_ids,
)
from vllm.entrypoints.serve.engine.protocol import ErrorResponse
@@ -1045,6 +1047,7 @@ class OpenAIServingResponses(GenerateBaseServing):
tool_calls=maybe_filter_parallel_tool_calls(tool_calls or [], request),
logprobs=logprobs,
tools=request.tools,
finish_reason=final_output.finish_reason,
)
# Fallback when no parser is configured
@@ -1062,7 +1065,7 @@ class OpenAIServingResponses(GenerateBaseServing):
if final_output.text
else [],
role="assistant",
status="completed",
status=response_item_status(final_output.finish_reason),
type="message",
)
]
@@ -1124,7 +1127,10 @@ class OpenAIServingResponses(GenerateBaseServing):
while current_index < len(event_deque):
event = event_deque[current_index]
yield event
if getattr(event, "type", "unknown") == "response.completed":
if getattr(event, "type", "unknown") in (
"response.completed",
"response.incomplete",
):
return
current_index += 1
@@ -1222,6 +1228,7 @@ class OpenAIServingResponses(GenerateBaseServing):
top_logprobs=request.top_logprobs,
)
finish_reason: str | None = None
async for ctx in result_generator:
assert isinstance(ctx, SimpleContext)
if ctx.last_output is None or not ctx.last_output.outputs:
@@ -1229,6 +1236,7 @@ class OpenAIServingResponses(GenerateBaseServing):
output = ctx.last_output.outputs[0]
self._raise_if_error(output.finish_reason, request.request_id)
finish_reason = output.finish_reason
delta_text = output.text
delta_token_ids = as_list(output.token_ids)
@@ -1261,7 +1269,7 @@ class OpenAIServingResponses(GenerateBaseServing):
for event in processor.emit_delta(dm, output, _get_logprobs):
yield _increment_sequence_number_and_return(event)
for event in processor.close_current():
for event in processor.close_current(response_item_status(finish_reason)):
yield _increment_sequence_number_and_return(event)
assert isinstance(context, SimpleContext)
@@ -1415,10 +1423,20 @@ class OpenAIServingResponses(GenerateBaseServing):
request_metadata,
created_time=created_time,
)
yield _increment_sequence_number_and_return(
ResponseCompletedEvent(
terminal_event: StreamingResponsesResponse
if (
isinstance(final_response, ResponsesResponse)
and final_response.status == "incomplete"
):
terminal_event = ResponseIncompleteEvent(
type="response.incomplete",
sequence_number=-1,
response=final_response,
)
else:
terminal_event = ResponseCompletedEvent(
type="response.completed",
sequence_number=-1,
response=final_response,
)
)
yield _increment_sequence_number_and_return(terminal_event)