[Frontend] Consolidate scale out entrypoints (#44512)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
wang.yuqi
2026-06-29 03:18:53 -07:00
committed by GitHub
parent eddfd4cf21
commit 3483240b7e
45 changed files with 422 additions and 289 deletions
+10 -23
View File
@@ -32,7 +32,6 @@ from vllm.entrypoints.openai.engine.protocol import GenerationError
from vllm.entrypoints.openai.models.protocol import BaseModelPath
from vllm.entrypoints.openai.models.serving import OpenAIServingModels
from vllm.entrypoints.serve.elastic_ep.middleware import ScalingMiddleware
from vllm.entrypoints.serve.render.serving import ServingRender
from vllm.entrypoints.serve.sagemaker.api_router import sagemaker_standards_bootstrap
from vllm.entrypoints.serve.tokenize.serving import ServingTokenization
from vllm.entrypoints.serve.utils.api_utils import (
@@ -208,12 +207,6 @@ def build_app(
register_generate_api_routers(app)
from vllm.entrypoints.serve.disagg.api_router import (
attach_router as attach_disagg_router,
)
attach_disagg_router(app)
from vllm.entrypoints.serve.elastic_ep.api_router import (
attach_router as elastic_ep_attach_router,
)
@@ -221,11 +214,9 @@ def build_app(
elastic_ep_attach_router(app)
if "generate" in supported_tasks or "render" in supported_tasks:
from vllm.entrypoints.serve.render.api_router import (
attach_router as attach_render_router,
)
from vllm.entrypoints.scale_out.factories import register_scale_out_api_routers
attach_render_router(app)
register_scale_out_api_routers(app, supported_tasks)
if "transcription" in supported_tasks or "realtime" in supported_tasks:
from vllm.entrypoints.speech_to_text.factories import (
@@ -401,12 +392,6 @@ async def init_app_state(
default_chat_template_kwargs=args.default_chat_template_kwargs,
trust_request_chat_template=args.trust_request_chat_template,
)
state.serving_render = ServingRender(
state.openai_serving_models,
state.online_renderer,
state.online_derenderer,
request_logger=request_logger,
)
if "generate" in supported_tasks:
from vllm.entrypoints.generate.api_router import init_generate_state
@@ -415,6 +400,10 @@ async def init_app_state(
engine_client, state, args, request_logger, supported_tasks
)
from vllm.entrypoints.scale_out.factories import init_scale_out_state
init_scale_out_state(state, args, engine_client, request_logger)
if "transcription" in supported_tasks or "realtime" in supported_tasks:
from vllm.entrypoints.speech_to_text.factories import init_speech_to_text_state
@@ -505,12 +494,10 @@ async def init_render_app_state(
default_chat_template_kwargs=args.default_chat_template_kwargs,
trust_request_chat_template=args.trust_request_chat_template,
)
state.serving_render = ServingRender(
model_registry,
state.online_renderer,
state.online_derenderer,
request_logger=request_logger,
)
from vllm.entrypoints.scale_out.factories import init_render_state
init_render_state(state, request_logger)
state.vllm_config = vllm_config
# Disable stats logging — there is no engine to poll.