[Frontend] Refine the entrypoint class's inheritance hierarchy. (#47498)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
wang.yuqi
2026-07-03 10:50:06 +00:00
committed by GitHub
parent 18f658bb31
commit a14f57a3ac
23 changed files with 90 additions and 191 deletions
@@ -78,7 +78,6 @@ def _create_mock_engine():
mock_engine.model_config = MockModelConfig()
mock_engine.input_processor = MagicMock()
# renderer is accessed by OpenAIServing.__init__ and serving.py
mock_renderer = MagicMock()
mock_renderer.tokenizer = get_tokenizer(MODEL_NAME)
mock_engine.renderer = mock_renderer
@@ -7,20 +7,20 @@ from unittest.mock import MagicMock
import pytest
import vllm.envs as envs
from vllm.entrypoints.openai.engine.serving import GenerationError, OpenAIServing
from vllm.entrypoints.generate.base.serving import GenerateBaseServing, GenerationError
from vllm.envs import disable_envs_cache
@pytest.mark.asyncio
async def test_raise_if_error_raises_generation_error():
"""test _raise_if_error raises GenerationError"""
# create a minimal OpenAIServing instance
# create a minimal GenerateBaseServing instance
mock_engine = MagicMock()
mock_engine.model_config = MagicMock()
mock_engine.model_config.max_model_len = 100
mock_models = MagicMock()
serving = OpenAIServing(
serving = GenerateBaseServing(
engine_client=mock_engine,
models=mock_models,
request_logger=None,
@@ -47,7 +47,7 @@ async def test_convert_generation_error_to_streaming_response():
mock_engine.model_config.max_model_len = 100
mock_models = MagicMock()
serving = OpenAIServing(
serving = GenerateBaseServing(
engine_client=mock_engine,
models=mock_models,
request_logger=None,
@@ -77,7 +77,7 @@ def test_is_model_supported_skip_name_validation_env(
mock_models = MagicMock()
mock_models.is_base_model.return_value = False
serving = OpenAIServing(
serving = GenerateBaseServing(
engine_client=mock_engine,
models=mock_models,
request_logger=None,
@@ -127,13 +127,13 @@ def test_responses_api_logprobs_with_return_tokens_as_token_ids():
"""Test that return_tokens_as_token_ids works in Responses API logprobs."""
from unittest.mock import MagicMock
from vllm.entrypoints.openai.engine.serving import OpenAIServing
from vllm.entrypoints.generate.base.serving import GenerateBaseServing
from vllm.entrypoints.openai.responses.serving import OpenAIServingResponses
from vllm.logprobs import Logprob as SampleLogprob
serving = MagicMock(spec=OpenAIServingResponses)
serving.return_tokens_as_token_ids = True
serving._get_decoded_token = OpenAIServing._get_decoded_token
serving._get_decoded_token = GenerateBaseServing._get_decoded_token
tokenizer = MagicMock()
tokenizer.decode = lambda token_id: "decoded"
@@ -14,7 +14,7 @@ from vllm.entrypoints.openai.engine.protocol import (
)
from vllm.entrypoints.openai.models.protocol import BaseModelPath
from vllm.entrypoints.openai.models.serving import OpenAIServingModels
from vllm.entrypoints.pooling.base.serving import PoolingServingBase
from vllm.entrypoints.pooling.base.serving import PoolingBaseServing
from vllm.entrypoints.pooling.typing import PoolingServeContext
from vllm.entrypoints.serve.lora.protocol import (
LoadLoRAAdapterRequest,
@@ -136,7 +136,7 @@ async def test_unload_lora_adapter_not_found():
assert response.error.code == HTTPStatus.NOT_FOUND
class _ConcretePoolingServing(PoolingServingBase):
class _ConcretePoolingServing(PoolingBaseServing):
"""Minimal concrete subclass used only in these unit tests."""
request_id_prefix = "test"
@@ -178,7 +178,7 @@ def test_pooling_maybe_get_adapters_lora_name_sets_lora_request():
serving = _make_pooling_serving(lora_name)
ctx = _make_pooling_ctx(lora_name)
serving._maybe_get_adapters(ctx)
ctx.lora_request = serving._maybe_get_adapters(ctx.request)
assert ctx.lora_request is not None
assert ctx.lora_request.lora_name == lora_name
@@ -190,4 +190,4 @@ def test_pooling_maybe_get_adapters_unknown_model_raises():
ctx = _make_pooling_ctx("unknown-model")
with pytest.raises(VLLMNotFoundError):
serving._maybe_get_adapters(ctx)
serving._maybe_get_adapters(ctx.request)
@@ -7,7 +7,7 @@ from unittest.mock import AsyncMock, Mock
import pytest
from vllm.entrypoints.speech_to_text.base.serving import OpenAISpeechToText
from vllm.entrypoints.speech_to_text.base.serving import SpeechToTextBaseServing
from vllm.entrypoints.speech_to_text.transcription.protocol import TranscriptionResponse
@@ -43,7 +43,7 @@ async def test_non_streaming_cancel_aborts_engine_requests(
is_tracing_enabled=AsyncMock(return_value=False),
)
server = OpenAISpeechToText.__new__(OpenAISpeechToText)
server = SpeechToTextBaseServing.__new__(SpeechToTextBaseServing)
server.engine_client = engine_client
server.task_type = "transcribe"
server.models = SimpleNamespace(model_name=lambda: "audio")
@@ -112,7 +112,7 @@ async def test_non_streaming_cancel_advances_all_chunk_generators():
{"prompt": "chunk-1"},
{"prompt": "chunk-2"},
]
server = OpenAISpeechToText.__new__(OpenAISpeechToText)
server = SpeechToTextBaseServing.__new__(SpeechToTextBaseServing)
server.engine_client = engine_client
server.task_type = "transcribe"
server.models = SimpleNamespace(model_name=lambda: "audio")
@@ -170,7 +170,7 @@ async def test_language_detection_cancel_aborts_engine_request():
abort=AsyncMock(),
)
server = OpenAISpeechToText.__new__(OpenAISpeechToText)
server = SpeechToTextBaseServing.__new__(SpeechToTextBaseServing)
server.engine_client = engine_client
server.asr_config = SimpleNamespace()
server.tokenizer = Mock()
@@ -25,7 +25,7 @@ from vllm.entrypoints.openai.engine.protocol import (
)
from vllm.entrypoints.openai.models.serving import OpenAIServingModels
from vllm.entrypoints.speech_to_text.base.serving import (
OpenAISpeechToText,
SpeechToTextBaseServing,
asr_inter_chunk_separator,
)
from vllm.entrypoints.speech_to_text.transcription.protocol import TranscriptionRequest
@@ -234,7 +234,9 @@ async def test_create_transcription_non_streaming_joins_chunks_by_language():
"vllm.model_executor.model_loader.get_model_cls",
return_value=_StubTranscriptionModel,
),
patch.object(OpenAISpeechToText, "_preprocess_speech_to_text", preprocess_mock),
patch.object(
SpeechToTextBaseServing, "_preprocess_speech_to_text", preprocess_mock
),
):
serving = OpenAIServingTranscription(engine_client, models, request_logger=None)