[V1][Misc] Shorten FinishReason enum and use constant strings (#12760)

This commit is contained in:
Nick Hill
2025-02-04 22:43:02 -08:00
committed by GitHub
parent fcf2e3d7fc
commit 3d09e592a8
5 changed files with 25 additions and 21 deletions

View File

@@ -6,7 +6,7 @@ from typing import TYPE_CHECKING, List, Optional, Union
from vllm.lora.request import LoRARequest
from vllm.sampling_params import SamplingParams
from vllm.sequence import RequestMetrics
from vllm.v1.engine import EngineCoreRequest, RequestFinishedReason
from vllm.v1.engine import EngineCoreRequest, FinishReason
from vllm.v1.utils import ConstantList
if TYPE_CHECKING:
@@ -109,7 +109,7 @@ class Request:
def is_finished(self) -> bool:
return RequestStatus.is_finished(self.status)
def get_finished_reason(self) -> Union[RequestFinishedReason, None]:
def get_finished_reason(self) -> Union[FinishReason, None]:
return RequestStatus.get_finished_reason(self.status)
def has_encoder_inputs(self) -> bool:
@@ -150,7 +150,7 @@ class RequestStatus(enum.IntEnum):
@staticmethod
def get_finished_reason(
status: "RequestStatus") -> Union[RequestFinishedReason, None]:
status: "RequestStatus") -> Union[FinishReason, None]:
return _FINISHED_REASON_MAP.get(status)
@@ -159,8 +159,8 @@ class RequestStatus(enum.IntEnum):
# are longer than the model's length cap. Therefore, the stop
# reason should also be "length" as in OpenAI API.
_FINISHED_REASON_MAP = {
RequestStatus.FINISHED_STOPPED: RequestFinishedReason.STOP,
RequestStatus.FINISHED_LENGTH_CAPPED: RequestFinishedReason.LENGTH,
RequestStatus.FINISHED_ABORTED: RequestFinishedReason.ABORT,
RequestStatus.FINISHED_IGNORED: RequestFinishedReason.LENGTH,
RequestStatus.FINISHED_STOPPED: FinishReason.STOP,
RequestStatus.FINISHED_LENGTH_CAPPED: FinishReason.LENGTH,
RequestStatus.FINISHED_ABORTED: FinishReason.ABORT,
RequestStatus.FINISHED_IGNORED: FinishReason.LENGTH,
}