[Frontend] Split ServingRender into renderer and entrypoint. (#44285)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
wang.yuqi
2026-06-23 11:19:09 +00:00
committed by GitHub
parent d32575a2d2
commit 31ca9504b1
26 changed files with 1155 additions and 939 deletions
@@ -38,7 +38,6 @@ from vllm.entrypoints.openai.models.serving import (
OpenAIServingModels,
)
from vllm.entrypoints.openai.parser.harmony_utils import get_encoding
from vllm.entrypoints.serve.render.serving import OpenAIServingRender
from vllm.exceptions import VLLMValidationError
from vllm.inputs import TokensPrompt
from vllm.multimodal.inputs import PlaceholderRange
@@ -46,6 +45,7 @@ from vllm.outputs import CompletionOutput, RequestOutput
from vllm.parser import HarmonyParser
from vllm.renderers.hf import HfRenderer
from vllm.renderers.mistral import MistralRenderer
from vllm.renderers.online_renderer import OnlineRenderer
from vllm.tokenizers import get_tokenizer
from vllm.tokenizers.mistral import MistralTokenizer
from vllm.tokenizers.registry import cached_tokenizer_from_config
@@ -567,13 +567,12 @@ def _build_renderer(model_config: MockModelConfig):
)
def _build_serving_render(
def _build_online_renderer(
engine, model_registry: OpenAIModelRegistry
) -> OpenAIServingRender:
return OpenAIServingRender(
) -> OnlineRenderer:
return OnlineRenderer(
model_config=engine.model_config,
renderer=engine.renderer,
model_registry=model_registry,
request_logger=None,
chat_template=CHAT_TEMPLATE,
chat_template_content_format="auto",
@@ -591,13 +590,13 @@ def _build_serving_chat(
engine_client=engine,
base_model_paths=BASE_MODEL_PATHS,
)
openai_serving_render = _build_serving_render(engine, models.registry)
online_renderer = _build_online_renderer(engine, models.registry)
serving_chat = OpenAIServingChat(
engine,
models,
response_role="assistant",
openai_serving_render=openai_serving_render,
online_renderer=online_renderer,
chat_template=CHAT_TEMPLATE,
chat_template_content_format="auto",
request_logger=None,
@@ -621,13 +620,13 @@ async def _async_serving_chat_init():
engine = MockEngine()
models = OpenAIServingModels(engine, BASE_MODEL_PATHS)
openai_serving_render = _build_serving_render(engine, models.registry)
online_renderer = _build_online_renderer(engine, models.registry)
serving_completion = OpenAIServingChat(
engine,
models,
response_role="assistant",
openai_serving_render=openai_serving_render,
online_renderer=online_renderer,
chat_template=CHAT_TEMPLATE,
chat_template_content_format="auto",
request_logger=None,
@@ -1396,9 +1395,7 @@ class TestServingChatWithHarmony:
messages=messages,
include_reasoning=include_reasoning,
)
input_messages, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req)
)
input_messages, _ = serving_chat.online_renderer._make_request_with_harmony(req)
verify_harmony_messages(
input_messages,
[
@@ -1429,8 +1426,8 @@ class TestServingChatWithHarmony:
# Test the Harmony messages for the second turn's input
req_2 = ChatCompletionRequest(model=MODEL_NAME, messages=messages)
input_messages_2, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req_2)
input_messages_2, _ = serving_chat.online_renderer._make_request_with_harmony(
req_2
)
verify_harmony_messages(
input_messages_2,
@@ -1452,9 +1449,7 @@ class TestServingChatWithHarmony:
{"role": "user", "content": "Hello"},
]
req = ChatCompletionRequest(model=MODEL_NAME, messages=messages)
input_messages, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req)
)
input_messages, _ = serving_chat.online_renderer._make_request_with_harmony(req)
verify_harmony_messages(
input_messages,
[
@@ -1478,9 +1473,7 @@ class TestServingChatWithHarmony:
req = ChatCompletionRequest(
model=MODEL_NAME, messages=messages, tools=weather_tools
)
input_messages, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req)
)
input_messages, _ = serving_chat.online_renderer._make_request_with_harmony(req)
verify_harmony_messages(
input_messages,
[
@@ -1503,9 +1496,7 @@ class TestServingChatWithHarmony:
# Test the Harmony messages for the first turn's input
req = ChatCompletionRequest(model=MODEL_NAME, messages=messages, tools=tools)
input_messages, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req)
)
input_messages, _ = serving_chat.online_renderer._make_request_with_harmony(req)
verify_harmony_messages(
input_messages,
[
@@ -1549,8 +1540,8 @@ class TestServingChatWithHarmony:
# Test the Harmony messages for the second turn's input
req_2 = ChatCompletionRequest(model=MODEL_NAME, messages=messages, tools=tools)
input_messages_2, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req_2)
input_messages_2, _ = serving_chat.online_renderer._make_request_with_harmony(
req_2
)
verify_harmony_messages(
input_messages_2,
@@ -1588,9 +1579,7 @@ class TestServingChatWithHarmony:
# Test the Harmony messages for the first turn's input
req = ChatCompletionRequest(model=MODEL_NAME, messages=messages, tools=tools)
input_messages, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req)
)
input_messages, _ = serving_chat.online_renderer._make_request_with_harmony(req)
verify_harmony_messages(
input_messages,
[
@@ -1634,8 +1623,8 @@ class TestServingChatWithHarmony:
# Test the Harmony messages for the second turn's input
req_2 = ChatCompletionRequest(model=MODEL_NAME, messages=messages, tools=tools)
input_messages_2, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req_2)
input_messages_2, _ = serving_chat.online_renderer._make_request_with_harmony(
req_2
)
verify_harmony_messages(
input_messages_2,
@@ -1686,8 +1675,8 @@ class TestServingChatWithHarmony:
# Test the Harmony messages for the third turn's input
req_3 = ChatCompletionRequest(model=MODEL_NAME, messages=messages, tools=tools)
input_messages_3, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req_3)
input_messages_3, _ = serving_chat.online_renderer._make_request_with_harmony(
req_3
)
verify_harmony_messages(
input_messages_3,
@@ -1751,8 +1740,8 @@ class TestServingChatWithHarmony:
# Test the Harmony messages for the fourth turn's input
req_4 = ChatCompletionRequest(model=MODEL_NAME, messages=messages, tools=tools)
input_messages_4, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req_4)
input_messages_4, _ = serving_chat.online_renderer._make_request_with_harmony(
req_4
)
verify_harmony_messages(
input_messages_4,
@@ -1802,9 +1791,7 @@ class TestServingChatWithHarmony:
},
]
req = ChatCompletionRequest(model=MODEL_NAME, messages=messages)
input_messages, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req)
)
input_messages, _ = serving_chat.online_renderer._make_request_with_harmony(req)
verify_harmony_messages(
input_messages,
@@ -1835,9 +1822,7 @@ class TestServingChatWithHarmony:
},
]
req = ChatCompletionRequest(model=MODEL_NAME, messages=messages)
input_messages, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req)
)
input_messages, _ = serving_chat.online_renderer._make_request_with_harmony(req)
verify_harmony_messages(
input_messages,
@@ -1866,9 +1851,7 @@ class TestServingChatWithHarmony:
},
]
req = ChatCompletionRequest(model=MODEL_NAME, messages=messages)
input_messages, _ = (
serving_chat.openai_serving_render._make_request_with_harmony(req)
)
input_messages, _ = serving_chat.online_renderer._make_request_with_harmony(req)
verify_harmony_messages(
input_messages,
@@ -1898,14 +1881,14 @@ async def test_tool_choice_validation_without_parser():
engine_client=mock_engine,
base_model_paths=BASE_MODEL_PATHS,
)
openai_serving_render = _build_serving_render(mock_engine, models.registry)
online_renderer = _build_online_renderer(mock_engine, models.registry)
# Create serving_chat without tool_parser (enable_auto_tools=False)
serving_chat = OpenAIServingChat(
mock_engine,
models,
response_role="assistant",
openai_serving_render=openai_serving_render,
online_renderer=online_renderer,
chat_template=CHAT_TEMPLATE,
chat_template_content_format="auto",
request_logger=None,
@@ -1967,13 +1950,13 @@ async def test_streaming_n_gt1_independent_tool_parsers():
engine_client=mock_engine,
base_model_paths=BASE_MODEL_PATHS,
)
openai_serving_render = _build_serving_render(mock_engine, models.registry)
online_renderer = _build_online_renderer(mock_engine, models.registry)
serving_chat = OpenAIServingChat(
mock_engine,
models,
response_role="assistant",
openai_serving_render=openai_serving_render,
online_renderer=online_renderer,
chat_template=CHAT_TEMPLATE,
chat_template_content_format="auto",
request_logger=None,