[V1] Use msgpack for core request serialization (#12918)

Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nick Hill
2025-02-10 11:35:56 +08:00
committed by GitHub
parent aa0ca5ebb7
commit 67c4637ccf
4 changed files with 62 additions and 95 deletions
+14 -28
View File
@@ -1,20 +1,17 @@
# SPDX-License-Identifier: Apache-2.0
import enum
from dataclasses import dataclass
from typing import TYPE_CHECKING, List, Optional, Union
from typing import List, Optional, Union
import msgspec
from vllm.lora.request import LoRARequest
from vllm.multimodal import MultiModalKwargs
from vllm.multimodal.inputs import PlaceholderRange
from vllm.sampling_params import SamplingParams
from vllm.v1.metrics.stats import SchedulerStats
from vllm.v1.outputs import LogprobsLists, LogprobsTensors
if TYPE_CHECKING:
from vllm.lora.request import LoRARequest
from vllm.multimodal import MultiModalKwargs
from vllm.multimodal.inputs import PlaceholderRange
from vllm.sampling_params import SamplingParams
# These are possible values of RequestOutput.finish_reason,
# so form part of the external API.
FINISH_REASON_STRINGS = ("stop", "length", "abort")
@@ -39,8 +36,11 @@ class FinishReason(enum.IntEnum):
return FINISH_REASON_STRINGS[self.value]
@dataclass
class EngineCoreRequest:
class EngineCoreRequest(
msgspec.Struct,
array_like=True, # type: ignore[call-arg]
omit_defaults=True, # type: ignore[call-arg]
gc=False): # type: ignore[call-arg]
# NOTE: prompt and prompt_token_ids should be DecoderOnlyInput,
# but this object is currently not playing well with msgspec
@@ -51,13 +51,13 @@ class EngineCoreRequest:
# Detokenizer, but set to None when it is added to EngineCoreClient.
prompt: Optional[str]
prompt_token_ids: List[int]
mm_inputs: Optional[List[Optional["MultiModalKwargs"]]]
mm_inputs: Optional[List[Optional[MultiModalKwargs]]]
mm_hashes: Optional[List[str]]
mm_placeholders: Optional[List["PlaceholderRange"]]
sampling_params: "SamplingParams"
mm_placeholders: Optional[List[PlaceholderRange]]
sampling_params: SamplingParams
eos_token_id: Optional[int]
arrival_time: float
lora_request: Optional["LoRARequest"]
lora_request: Optional[LoRARequest]
class EngineCoreOutput(
@@ -94,16 +94,6 @@ class EngineCoreOutputs(
scheduler_stats: SchedulerStats
@dataclass
class EngineCoreProfile:
is_start: bool
@dataclass
class EngineCoreResetPrefixCache:
pass
class EngineCoreRequestType(enum.Enum):
"""
Request types defined as hex byte strings, so it can be sent over sockets
@@ -113,7 +103,3 @@ class EngineCoreRequestType(enum.Enum):
ABORT = b'\x01'
PROFILE = b'\x02'
RESET_PREFIX_CACHE = b'\x03'
EngineCoreRequestUnion = Union[EngineCoreRequest, EngineCoreProfile,
EngineCoreResetPrefixCache, List[str]]