[V0 deprecation] Guided decoding (#21347)

Signed-off-by: Reza Barazesh <rezabarazesh@meta.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Reza Barazesh
2025-07-29 03:15:30 -07:00
committed by GitHub
co-authored by Harry Mellor
parent a4528f0cac
commit 37efc63b64
29 changed files with 103 additions and 2809 deletions
-18
View File
@@ -20,8 +20,6 @@ from vllm.config import DecodingConfig, ModelConfig, VllmConfig
from vllm.core.scheduler import SchedulerOutputs
# yapf conflicts with isort for this block
# yapf: disable
from vllm.engine.async_llm_engine import (
build_guided_decoding_logits_processor_async)
from vllm.engine.multiprocessing import (ENGINE_DEAD_ERROR, IPC_DATA_EXT,
IPC_HEALTH_EXT, IPC_INPUT_EXT,
IPC_OUTPUT_EXT, RPC_REQUEST_T,
@@ -537,22 +535,6 @@ class MQLLMEngineClient(EngineClient):
if request_id in self.output_queues:
raise ValueError(f"Request {request_id} already exists")
# Constructing guided decoding logits processors is expensive, so we do
# it here to avoid contending with cpu resources and the GIL on the
# backend process.
if isinstance(params, SamplingParams) and \
params.guided_decoding is not None:
params = await \
build_guided_decoding_logits_processor_async(
sampling_params=params,
tokenizer=await self.get_tokenizer(lora_request),
default_guided_backend=(self.decoding_config.backend
if self.decoding_config
else DecodingConfig.backend),
model_config=self.model_config,
reasoning_backend=self.decoding_config.reasoning_backend,
)
# 1) Create output queue for this requests.
queue: asyncio.Queue[Union[RequestOutput,
BaseException]] = asyncio.Queue()