[V1][Core] Support for Structured Outputs (#12388)

Signed-off-by: Aaron Pham <contact@aarnphm.xyz>
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Co-authored-by: Russell Bryant <rbryant@redhat.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Nick Hill <nhill@redhat.com>

vllm/v1/request.py

@@ -3,13 +3,15 @@
 import enum
 from typing import TYPE_CHECKING, Optional, Union
 
-from vllm.lora.request import LoRARequest
 from vllm.sampling_params import SamplingParams
 from vllm.v1.engine import (EngineCoreEvent, EngineCoreEventType,
                             EngineCoreRequest, FinishReason)
+from vllm.v1.structured_output.request import StructuredOutputRequest
 from vllm.v1.utils import ConstantList
 
 if TYPE_CHECKING:
+
+    from vllm.lora.request import LoRARequest
     from vllm.multimodal import MultiModalKwargs
     from vllm.multimodal.inputs import PlaceholderRange
 
@@ -27,15 +29,19 @@ class Request:
         sampling_params: SamplingParams,
         eos_token_id: Optional[int],
         arrival_time: float,
-        lora_request: Optional[LoRARequest] = None,
+        lora_request: Optional["LoRARequest"] = None,
+        structured_output_request: Optional["StructuredOutputRequest"] = None,
     ) -> None:
         self.request_id = request_id
         self.sampling_params = sampling_params
         # Because of LoRA, the eos token id can be different for each request.
         self.eos_token_id = eos_token_id
         self.lora_request = lora_request
+        self.structured_output_request = structured_output_request
 
-        self.status = RequestStatus.WAITING
+        self.status = (RequestStatus.WAITING_FOR_FSM
+                       if sampling_params.guided_decoding is not None else
+                       RequestStatus.WAITING)
         self.events: list[EngineCoreEvent] = []
         self.stop_reason: Union[int, str, None] = None
         assert sampling_params.max_tokens is not None
@@ -78,6 +84,8 @@ class Request:
             eos_token_id=request.eos_token_id,
             arrival_time=request.arrival_time,
             lora_request=request.lora_request,
+            structured_output_request=StructuredOutputRequest(
+                sampling_params=request.sampling_params),
         )
 
     def queued(self, timestamp: Optional[float] = None) -> None:
@@ -134,18 +142,23 @@ class Request:
         num_tokens = self.mm_positions[input_id]["length"]
         return num_tokens
 
+    @property
+    def use_structured_output(self) -> bool:
+        return self.sampling_params.guided_decoding is not None
+
 
 class RequestStatus(enum.IntEnum):
     """Status of a request."""
-    WAITING = 0
-    RUNNING = 1
-    PREEMPTED = 2
-    # Note: anything after PREEMPTED (2) will be considered
+    WAITING = enum.auto()
+    WAITING_FOR_FSM = enum.auto()
+    RUNNING = enum.auto()
+    PREEMPTED = enum.auto()
+    # Note: anything after PREEMPTED will be considered
     # as a finished status.
-    FINISHED_STOPPED = 3
-    FINISHED_LENGTH_CAPPED = 4
-    FINISHED_ABORTED = 5
-    FINISHED_IGNORED = 6
+    FINISHED_STOPPED = enum.auto()
+    FINISHED_LENGTH_CAPPED = enum.auto()
+    FINISHED_ABORTED = enum.auto()
+    FINISHED_IGNORED = enum.auto()
 
     @staticmethod
     def is_finished(status: "RequestStatus") -> bool: