A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Misc] Convert use_structured_output property into constant (#23060)
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Woosuk Kwon committed
Aug 17, 2025 at 12:41 UTC
6d243efedab9a03348cbd55fe966b62a08d90676
1 file changed
+4
-7
vllm/v1/request.py
+4
-7
@@ -54,8 +54,7 @@ class Request:
54
time.time()
55
56
self.status = RequestStatus.WAITING
57
- if sampling_params and sampling_params.guided_decoding is not None:
58
- self.status = RequestStatus.WAITING_FOR_FSM
57
+ self.use_structured_output = False
58
self.events: list[EngineCoreEvent] = []
59
self.stop_reason: Union[int, str, None] = None
60
@@ -63,12 +62,15 @@ class Request:
62
self.kv_transfer_params: Optional[dict[str, Any]] = None
63
64
if pooling_params is not None:
65
+ # Pooling models.
66
self.max_tokens = 1
67
elif sampling_params is not None:
68
+ # Generative models.
69
assert sampling_params.max_tokens is not None
70
self.max_tokens = sampling_params.max_tokens
71
if sampling_params.guided_decoding is not None:
72
self.status = RequestStatus.WAITING_FOR_FSM
73
+ self.use_structured_output = True
74
75
if sampling_params.extra_args is not None:
76
self.kv_transfer_params = \
@@ -192,11 +194,6 @@ class Request:
194
num_tokens = self.mm_positions[input_id].length
195
return num_tokens
196
195
- @property
196
- def use_structured_output(self) -> bool:
197
- return self.sampling_params is not None and \
198
- self.sampling_params.guided_decoding is not None
199
-
197
def record_event(
198
self,
199
event_type: EngineCoreEventType,