biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Revert "[Renderer] Separate out RendererConfig from ModelConfig (#30145)" (#30199)

Browse Source
This commit is contained in:
Cyrus Leung
2025-12-07 16:00:22 +08:00
committed by GitHub
parent 27f4c2fd46
commit e83b7e379c
105 changed files with 797 additions and 969 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/compile/distributed/test_sequence_parallelism.py
View File

@@ -17,7 +17,6 @@ from vllm.config import (
DeviceConfig,
ModelConfig,
PassConfig,
RendererConfig,
VllmConfig,
get_current_vllm_config,
set_current_vllm_config,
@@ -277,7 +276,6 @@ def sequence_parallelism_pass_on_test_model(
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
device_config=device_config,
compilation_config=compilation_config,
)

6
tests/compile/test_functionalization.py
View File

@@ -15,7 +15,6 @@ from vllm.config import (
CompilationConfig,
ModelConfig,
PassConfig,
RendererConfig,
VllmConfig,
set_current_vllm_config,
)
@@ -220,11 +219,8 @@ def test_fix_functionalization(
torch.set_default_device("cuda")
torch.set_default_dtype(dtype)
model_config = ModelConfig(dtype=dtype)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
model_config=ModelConfig(dtype=dtype),
compilation_config=CompilationConfig(
custom_ops=["all"],
pass_config=PassConfig(

6
tests/compile/test_fusion.py
View File

@@ -15,7 +15,6 @@ from vllm.config import (
CompilationMode,
ModelConfig,
PassConfig,
RendererConfig,
VllmConfig,
)
from vllm.model_executor.layers.layernorm import RMSNorm
@@ -155,11 +154,8 @@ def test_fusion_rmsnorm_quant(
custom_ops.append("+rms_norm")
if enable_quant_fp8_custom_op:
custom_ops.append("+quant_fp8")
model_config = ModelConfig(dtype=dtype)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
model_config=ModelConfig(dtype=dtype),
compilation_config=CompilationConfig(
mode=CompilationMode.VLLM_COMPILE,
custom_ops=custom_ops,

2
tests/compile/test_fusion_attn.py
View File

@@ -24,7 +24,6 @@ from vllm.config import (
CompilationMode,
ModelConfig,
PassConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
set_current_vllm_config,
@@ -326,7 +325,6 @@ def test_attention_quant_pattern(
)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
scheduler_config=SchedulerConfig(
max_num_seqs=1024,
max_model_len=model_config.max_model_len,

8
tests/compile/test_pass_manager.py
View File

@@ -7,7 +7,7 @@ import torch
from vllm.compilation.inductor_pass import CallableInductorPass, InductorPass
from vllm.compilation.pass_manager import PostGradPassManager
from vllm.config import ModelConfig, RendererConfig, VllmConfig
from vllm.config import ModelConfig, VllmConfig
# dummy custom pass that doesn't inherit
@@ -43,11 +43,7 @@ class ProperPass(InductorPass):
)
def test_pass_manager_uuid(callable):
# Some passes need dtype to be set
model_config = ModelConfig(dtype=torch.bfloat16)
config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
)
config = VllmConfig(model_config=ModelConfig(dtype=torch.bfloat16))
pass_manager = PostGradPassManager()
pass_manager.configure(config)

5
tests/compile/test_qk_norm_rope_fusion.py
View File

@@ -19,7 +19,6 @@ from vllm.config import (
CompilationMode,
ModelConfig,
PassConfig,
RendererConfig,
VllmConfig,
set_current_vllm_config,
)
@@ -134,10 +133,8 @@ def test_qk_norm_rope_fusion(
if enable_rope_custom_op:
custom_ops.append("+rotary_embedding")
model_config = ModelConfig(dtype=dtype)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
model_config=ModelConfig(dtype=dtype),
compilation_config=CompilationConfig(
mode=CompilationMode.VLLM_COMPILE,
custom_ops=custom_ops,

3
tests/distributed/test_kvlayout.py
View File

@@ -5,7 +5,6 @@ from vllm.config import (
DeviceConfig,
KVTransferConfig,
ModelConfig,
RendererConfig,
VllmConfig,
set_current_vllm_config,
)
@@ -48,7 +47,6 @@ def test_get_kv_connector_cache_layout_with_nixl_connector():
vllm_config = VllmConfig(
device_config=DeviceConfig("cpu"),
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
kv_transfer_config=kv_transfer_config,
)
with set_current_vllm_config(vllm_config):
@@ -72,7 +70,6 @@ def test_get_kv_connector_cache_layout_with_multi_connector():
vllm_config = VllmConfig(
device_config=DeviceConfig("cpu"),
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
kv_transfer_config=kv_transfer_config,
)
with set_current_vllm_config(vllm_config):

22
tests/entrypoints/openai/test_chat_template.py
View File

@@ -3,6 +3,7 @@
import pytest
from vllm.config import ModelConfig
from vllm.entrypoints.chat_utils import apply_hf_chat_template, load_chat_template
from vllm.entrypoints.openai.protocol import ChatCompletionRequest
from vllm.tokenizers import get_tokenizer
@@ -106,11 +107,24 @@ def test_get_gen_prompt(
model_info = HF_EXAMPLE_MODELS.find_hf_info(model)
model_info.check_available_online(on_fail="skip")
renderer_config = model_info.build_renderer_config(model)
model_config = ModelConfig(
model,
tokenizer=model_info.tokenizer or model,
tokenizer_mode=model_info.tokenizer_mode,
trust_remote_code=model_info.trust_remote_code,
revision=model_info.revision,
hf_overrides=model_info.hf_overrides,
skip_tokenizer_init=model_info.require_embed_inputs,
enable_prompt_embeds=model_info.require_embed_inputs,
enable_mm_embeds=model_info.require_embed_inputs,
enforce_eager=model_info.enforce_eager,
dtype=model_info.dtype,
)
# Initialize the tokenizer
tokenizer = get_tokenizer(
renderer_config.tokenizer,
trust_remote_code=renderer_config.trust_remote_code,
tokenizer_name=model_config.tokenizer,
trust_remote_code=model_config.trust_remote_code,
)
template_content = load_chat_template(chat_template=template)
@@ -129,7 +143,7 @@ def test_get_gen_prompt(
tokenizer=tokenizer,
conversation=mock_request.messages,
chat_template=mock_request.chat_template or template_content,
renderer_config=renderer_config,
model_config=model_config,
tools=None,
add_generation_prompt=mock_request.add_generation_prompt,
continue_final_message=mock_request.continue_final_message,

21
tests/entrypoints/openai/test_lora_resolvers.py
View File

@@ -33,34 +33,26 @@ class MockModelConfig:
"""Minimal mock ModelConfig for testing."""
model: str = MODEL_NAME
tokenizer: str = MODEL_NAME
trust_remote_code: bool = False
tokenizer_mode: str = "auto"
max_model_len: int = 100
tokenizer_revision: str | None = None
multimodal_config: MultiModalConfig = field(default_factory=MultiModalConfig)
hf_config: MockHFConfig = field(default_factory=MockHFConfig)
logits_processors: list[str] | None = None
logits_processor_pattern: str | None = None
diff_sampling_param: dict | None = None
allowed_local_media_path: str = ""
allowed_media_domains: list[str] | None = None
encoder_config = None
generation_config: str = "auto"
skip_tokenizer_init: bool = False
def get_diff_sampling_param(self):
return self.diff_sampling_param or {}
@dataclass
class MockRendererConfig:
"""Minimal mock RendererConfig for testing."""
model_config: MockModelConfig
tokenizer: str = MODEL_NAME
tokenizer_mode: str = "auto"
tokenizer_revision: str | None = None
skip_tokenizer_init: bool = False
allowed_local_media_path: str = ""
allowed_media_domains: list[str] | None = None
class MockLoRAResolver(LoRAResolver):
async def resolve_lora(
self, base_model_name: str, lora_name: str
@@ -122,7 +114,6 @@ def mock_serving_setup():
mock_engine.add_lora.reset_mock()
mock_engine.model_config = MockModelConfig()
mock_engine.renderer_config = MockRendererConfig(mock_engine.model_config)
mock_engine.input_processor = MagicMock()
mock_engine.io_processor = MagicMock()

28
tests/entrypoints/openai/test_serving_chat.py
View File

@@ -346,33 +346,27 @@ class MockHFConfig:
class MockModelConfig:
task = "generate"
runner_type = "generate"
tokenizer = MODEL_NAME
trust_remote_code = False
tokenizer_mode = "auto"
max_model_len = 100
tokenizer_revision = None
multimodal_config = MultiModalConfig()
hf_config = MockHFConfig()
logits_processors: list[str] | None = None
logits_processor_pattern = None
diff_sampling_param: dict | None = None
allowed_local_media_path: str = ""
allowed_media_domains: list[str] | None = None
encoder_config = None
generation_config: str = "auto"
media_io_kwargs: dict[str, dict[str, Any]] = field(default_factory=dict)
skip_tokenizer_init = False
def get_diff_sampling_param(self):
return self.diff_sampling_param or {}
@dataclass
class MockRendererConfig:
model_config: MockModelConfig = field(default_factory=MockModelConfig)
tokenizer = MODEL_NAME
tokenizer_mode = "auto"
tokenizer_revision = None
skip_tokenizer_init = False
media_io_kwargs: dict[str, dict[str, Any]] = field(default_factory=dict)
allowed_local_media_path: str = ""
allowed_media_domains: list[str] | None = None
def _build_serving_chat(engine: AsyncLLM) -> OpenAIServingChat:
models = OpenAIServingModels(
engine_client=engine,
@@ -405,7 +399,6 @@ def _build_serving_chat(engine: AsyncLLM) -> OpenAIServingChat:
@dataclass
class MockEngine:
model_config: MockModelConfig = field(default_factory=MockModelConfig)
renderer_config: MockRendererConfig = field(default_factory=MockRendererConfig)
input_processor: MagicMock = field(default_factory=MagicMock)
io_processor: MagicMock = field(default_factory=MagicMock)
@@ -436,7 +429,6 @@ async def test_serving_chat_returns_correct_model_name():
mock_engine.get_tokenizer.return_value = get_tokenizer(MODEL_NAME)
mock_engine.errored = False
mock_engine.model_config = MockModelConfig()
mock_engine.renderer_config = MockRendererConfig(mock_engine.model_config)
mock_engine.input_processor = MagicMock()
mock_engine.io_processor = MagicMock()
@@ -467,7 +459,6 @@ async def test_serving_chat_should_set_correct_max_tokens():
mock_engine.get_tokenizer.return_value = get_tokenizer(MODEL_NAME)
mock_engine.errored = False
mock_engine.model_config = MockModelConfig()
mock_engine.renderer_config = MockRendererConfig(mock_engine.model_config)
mock_engine.input_processor = MagicMock()
mock_engine.io_processor = MagicMock()
@@ -501,7 +492,6 @@ async def test_serving_chat_should_set_correct_max_tokens():
mock_engine.get_tokenizer.return_value = get_tokenizer(MODEL_NAME)
mock_engine.errored = False
mock_engine.model_config = mock_model_config
mock_engine.renderer_config = MockRendererConfig(mock_model_config)
mock_engine.input_processor = MagicMock()
mock_engine.io_processor = MagicMock()
@@ -547,7 +537,6 @@ async def test_serving_chat_should_set_correct_max_tokens():
mock_engine.get_tokenizer.return_value = get_tokenizer(MODEL_NAME)
mock_engine.errored = False
mock_engine.model_config = mock_model_config
mock_engine.renderer_config = MockRendererConfig(mock_model_config)
mock_engine.input_processor = MagicMock()
mock_engine.io_processor = MagicMock()
@@ -594,7 +583,6 @@ async def test_serving_chat_could_load_correct_generation_config():
mock_engine.get_tokenizer.return_value = get_tokenizer(MODEL_NAME)
mock_engine.errored = False
mock_engine.model_config = mock_model_config
mock_engine.renderer_config = MockRendererConfig(mock_model_config)
mock_engine.input_processor = MagicMock()
mock_engine.io_processor = MagicMock()
@@ -641,7 +629,6 @@ async def test_serving_chat_did_set_correct_cache_salt(model_type):
mock_engine.get_tokenizer.return_value = get_tokenizer(MODEL_NAME)
mock_engine.errored = False
mock_engine.model_config = mock_model_config
mock_engine.renderer_config = MockRendererConfig(mock_model_config)
mock_engine.input_processor = MagicMock()
mock_engine.io_processor = MagicMock()
@@ -675,7 +662,6 @@ async def test_serving_chat_data_parallel_rank_extraction():
mock_engine.get_tokenizer.return_value = get_tokenizer(MODEL_NAME)
mock_engine.errored = False
mock_engine.model_config = MockModelConfig()
mock_engine.renderer_config = MockRendererConfig(mock_engine.model_config)
mock_engine.input_processor = MagicMock()
mock_engine.io_processor = MagicMock()

8
tests/entrypoints/openai/test_serving_engine.py
View File

@@ -7,7 +7,7 @@ from unittest.mock import Mock
import pytest
from vllm.config import ModelConfig, RendererConfig
from vllm.config import ModelConfig
from vllm.entrypoints.openai.serving_engine import OpenAIServing
from vllm.entrypoints.openai.serving_models import OpenAIServingModels
from vllm.tokenizers import MistralTokenizer
@@ -19,16 +19,10 @@ def serving() -> OpenAIServing:
# Create minimal mocks
engine_client = Mock()
model_config = Mock(spec=ModelConfig)
model_config.max_model_len = 32768
renderer_config = Mock(spec=RendererConfig)
renderer_config.model_config = model_config
models = Mock(spec=OpenAIServingModels)
models.model_config = model_config
models.renderer_config = renderer_config
models.input_processor = Mock()
models.io_processor = Mock()

8
tests/entrypoints/openai/test_serving_models.py
View File

@@ -6,7 +6,7 @@ from unittest.mock import MagicMock
import pytest
from vllm.config import ModelConfig, RendererConfig
from vllm.config import ModelConfig
from vllm.engine.protocol import EngineClient
from vllm.entrypoints.openai.protocol import (
ErrorResponse,
@@ -27,15 +27,9 @@ LORA_UNLOADING_SUCCESS_MESSAGE = (
async def _async_serving_models_init() -> OpenAIServingModels:
mock_engine_client = MagicMock(spec=EngineClient)
# Set the max_model_len attribute to avoid missing attribute
mock_model_config = MagicMock(spec=ModelConfig)
mock_model_config.max_model_len = 2048
mock_renderer_config = MagicMock(spec=RendererConfig)
mock_renderer_config.model_config = mock_model_config
mock_engine_client.model_config = mock_model_config
mock_engine_client.renderer_config = mock_renderer_config
mock_engine_client.input_processor = MagicMock()
mock_engine_client.io_processor = MagicMock()

190
tests/entrypoints/test_chat_utils.py
View File

@@ -12,7 +12,7 @@ from mistral_common.tokens.tokenizers.base import SpecialTokenPolicy
from vllm.assets.audio import AudioAsset
from vllm.assets.image import ImageAsset
from vllm.assets.video import VideoAsset
from vllm.config import ModelConfig, RendererConfig
from vllm.config import ModelConfig
from vllm.entrypoints.chat_utils import (
_try_extract_ast,
apply_mistral_chat_template,
@@ -233,7 +233,7 @@ def test_parse_chat_messages_single_image(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -265,7 +265,7 @@ def test_parse_chat_messages_single_image_with_uuid(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -295,7 +295,7 @@ def test_parse_chat_messages_single_empty_image_with_uuid(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -328,7 +328,7 @@ def test_parse_chat_messages_single_image_with_bad_uuid_format(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -369,7 +369,7 @@ def test_parse_chat_messages_multiple_images_with_uuids(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -409,7 +409,7 @@ def test_parse_chat_messages_multiple_empty_images_with_uuids(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -451,7 +451,7 @@ def test_parse_chat_messages_mixed_empty_images_with_uuids(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -485,7 +485,7 @@ async def test_parse_chat_messages_single_image_with_uuid_async(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -516,7 +516,7 @@ async def test_parse_chat_messages_empty_image_with_uuid_async(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -554,7 +554,7 @@ async def test_parse_chat_messages_multiple_images_with_uuids_async(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -595,7 +595,7 @@ async def test_parse_chat_messages_multiple_empty_images_with_uuids_async(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -634,7 +634,7 @@ async def test_parse_chat_messages_multiple_images_with_partial_uuids_async(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -660,7 +660,7 @@ def test_parse_chat_messages_empty_system(
"content": [{"type": "text", "text": "Who are you?"}],
},
],
RendererConfig(model_config=mistral_model_config),
mistral_model_config,
content_format="string",
)
assert conversation == [
@@ -677,7 +677,7 @@ def test_parse_chat_messages_empty_system(
"content": [{"type": "text", "text": "Who are you?"}],
},
],
RendererConfig(model_config=mistral_model_config),
mistral_model_config,
content_format="openai",
)
assert conversation == [
@@ -701,7 +701,7 @@ async def test_parse_chat_messages_single_image_async(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -730,7 +730,7 @@ def test_parse_chat_messages_multiple_images(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -758,7 +758,7 @@ def test_parse_chat_messages_empty_pil_image_with_uuid(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -786,7 +786,7 @@ def test_parse_chat_messages_empty_image_embeds_with_uuid(
],
}
],
RendererConfig(model_config=phi3v_model_config_image_embeds),
phi3v_model_config_image_embeds,
content_format="string",
)
@@ -818,7 +818,7 @@ def test_parse_chat_messages_empty_audio_embeds_with_uuid(
],
}
],
RendererConfig(model_config=audio_embeds_model_config),
audio_embeds_model_config,
content_format="string",
)
@@ -858,7 +858,7 @@ def test_parse_chat_messages_audio_embeds_with_string(
],
}
],
RendererConfig(model_config=audio_embeds_model_config),
audio_embeds_model_config,
content_format="string",
)
@@ -900,7 +900,7 @@ async def test_parse_chat_messages_audio_embeds_async(
],
}
],
RendererConfig(model_config=audio_embeds_model_config),
audio_embeds_model_config,
content_format="string",
)
@@ -1108,7 +1108,7 @@ async def test_parse_chat_messages_empty_image_embeds_with_uuid_async(
],
}
],
RendererConfig(model_config=phi3v_model_config_image_embeds),
phi3v_model_config_image_embeds,
content_format="string",
)
@@ -1144,7 +1144,7 @@ async def test_parse_chat_messages_multiple_images_async(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -1176,7 +1176,7 @@ def test_parse_chat_messages_placeholder_already_in_prompt(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
assert conversation == [
@@ -1208,7 +1208,7 @@ def test_parse_chat_messages_placeholder_one_already_in_prompt(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -1245,7 +1245,7 @@ def test_parse_chat_messages_multiple_images_across_messages(
],
},
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -1289,7 +1289,7 @@ def test_parse_chat_messages_multiple_images_with_uuids_across_messages(
],
},
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -1314,7 +1314,7 @@ def test_parse_chat_messages_context_text_format(
{"role": "assistant", "content": "Some stuff."},
{"role": "user", "content": "What about this one?"},
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="openai",
)
@@ -1367,7 +1367,7 @@ def test_parse_chat_messages_rejects_too_many_images_in_one_message(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -1410,7 +1410,7 @@ def test_parse_chat_messages_rejects_too_many_images_across_messages(
],
},
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -1430,7 +1430,7 @@ def test_parse_chat_messages_multiple_images_uncommon_input(
],
}
],
RendererConfig(model_config=phi3v_model_config),
phi3v_model_config,
content_format="string",
)
@@ -1464,7 +1464,7 @@ def test_parse_chat_messages_multiple_images_interleave(
],
}
],
RendererConfig(model_config=phi3v_model_config_mm_interleaved),
phi3v_model_config_mm_interleaved,
content_format="string",
)
@@ -1500,7 +1500,7 @@ async def test_parse_chat_messages_multiple_images_interleave_async(
],
}
],
RendererConfig(model_config=phi3v_model_config_mm_interleaved),
phi3v_model_config_mm_interleaved,
content_format="string",
)
@@ -1545,7 +1545,7 @@ async def test_parse_chat_messages_multiple_images_with_uuids_interleave_async(
],
}
],
RendererConfig(model_config=phi3v_model_config_mm_interleaved),
phi3v_model_config_mm_interleaved,
content_format="string",
)
@@ -1583,7 +1583,7 @@ def test_parse_chat_messages_multiple_images_multiple_messages_interleave(
],
},
],
RendererConfig(model_config=phi3v_model_config_mm_interleaved),
phi3v_model_config_mm_interleaved,
content_format="string",
)
@@ -1631,7 +1631,7 @@ def test_parse_chat_messages_multiple_images_with_uuids_multiple_messages_interl
],
},
],
RendererConfig(model_config=phi3v_model_config_mm_interleaved),
phi3v_model_config_mm_interleaved,
content_format="string",
)
@@ -1675,7 +1675,7 @@ def test_parse_chat_messages_multiple_modals_multiple_messages_interleave(
],
},
],
RendererConfig(model_config=qwen25omni_model_config_mm_interleaved),
qwen25omni_model_config_mm_interleaved,
content_format="string",
)
@@ -1743,7 +1743,7 @@ def test_parse_chat_messages_multiple_modals_with_uuids_multiple_messages_interl
],
},
],
RendererConfig(model_config=qwen25omni_model_config_mm_interleaved),
qwen25omni_model_config_mm_interleaved,
content_format="string",
)
@@ -1813,7 +1813,7 @@ def test_parse_chat_messages_multiple_modals_with_uuids_multiple_empty_media_mes
],
},
],
RendererConfig(model_config=qwen25omni_model_config_mm_interleaved),
qwen25omni_model_config_mm_interleaved,
content_format="string",
)
@@ -1879,7 +1879,7 @@ def test_parse_chat_messages_multiple_modals_with_partial_uuids_multiple_message
],
},
],
RendererConfig(model_config=qwen25omni_model_config_mm_interleaved),
qwen25omni_model_config_mm_interleaved,
content_format="string",
)
@@ -1927,7 +1927,7 @@ def test_parse_chat_messages_multiple_images_interleave_with_placeholders(
],
}
],
RendererConfig(model_config=phi3v_model_config_mm_interleaved),
phi3v_model_config_mm_interleaved,
content_format="string",
)
@@ -1945,11 +1945,24 @@ def test_resolve_hf_chat_template(sample_json_schema, model, use_tools):
model_info = HF_EXAMPLE_MODELS.find_hf_info(model)
model_info.check_available_online(on_fail="skip")
renderer_config = model_info.build_renderer_config(model)
model_config = ModelConfig(
model,
tokenizer=model_info.tokenizer or model,
tokenizer_mode=model_info.tokenizer_mode,
revision=model_info.revision,
trust_remote_code=model_info.trust_remote_code,
hf_overrides=model_info.hf_overrides,
skip_tokenizer_init=model_info.require_embed_inputs,
enable_prompt_embeds=model_info.require_embed_inputs,
enable_mm_embeds=model_info.require_embed_inputs,
enforce_eager=model_info.enforce_eager,
dtype=model_info.dtype,
)
# Build the tokenizer
tokenizer = get_tokenizer(
renderer_config.tokenizer,
trust_remote_code=renderer_config.trust_remote_code,
model,
trust_remote_code=model_config.trust_remote_code,
)
tools = (
@@ -1972,7 +1985,7 @@ def test_resolve_hf_chat_template(sample_json_schema, model, use_tools):
tokenizer,
chat_template=None,
tools=tools,
model_config=renderer_config.model_config,
model_config=model_config,
)
assert isinstance(chat_template, str)
@@ -2034,11 +2047,24 @@ def test_resolve_hf_chat_template_kwargs(sample_json_schema, model, expected_kwa
"enable_thinking": True,
}
renderer_config = model_info.build_renderer_config(model)
model_config = ModelConfig(
model,
tokenizer=model_info.tokenizer or model,
tokenizer_mode=model_info.tokenizer_mode,
revision=model_info.revision,
trust_remote_code=model_info.trust_remote_code,
hf_overrides=model_info.hf_overrides,
skip_tokenizer_init=model_info.require_embed_inputs,
enable_prompt_embeds=model_info.require_embed_inputs,
enable_mm_embeds=model_info.require_embed_inputs,
enforce_eager=model_info.enforce_eager,
dtype=model_info.dtype,
)
# Build the tokenizer
tokenizer = get_tokenizer(
renderer_config.tokenizer,
trust_remote_code=renderer_config.trust_remote_code,
model,
trust_remote_code=model_config.trust_remote_code,
)
# Test detecting the tokenizer's chat_template
@@ -2046,7 +2072,7 @@ def test_resolve_hf_chat_template_kwargs(sample_json_schema, model, expected_kwa
tokenizer,
chat_template=None,
tools=tools,
model_config=renderer_config.model_config,
model_config=model_config,
)
with pytest.raises(
ValueError, match="Found unexpected chat template kwargs from request"
@@ -2117,11 +2143,23 @@ def test_resolve_content_format_hf_defined(model, expected_format):
model_info = HF_EXAMPLE_MODELS.find_hf_info(model)
model_info.check_available_online(on_fail="skip")
renderer_config = model_info.build_renderer_config(model)
model_config = ModelConfig(
model,
tokenizer=model_info.tokenizer or model,
tokenizer_mode=model_info.tokenizer_mode,
revision=model_info.revision,
trust_remote_code=model_info.trust_remote_code,
hf_overrides=model_info.hf_overrides,
skip_tokenizer_init=model_info.require_embed_inputs,
enable_prompt_embeds=model_info.require_embed_inputs,
enable_mm_embeds=model_info.require_embed_inputs,
enforce_eager=model_info.enforce_eager,
dtype=model_info.dtype,
)
tokenizer = get_tokenizer(
renderer_config.tokenizer,
trust_remote_code=renderer_config.trust_remote_code,
model,
trust_remote_code=model_config.trust_remote_code,
)
# Test detecting the tokenizer's chat_template
@@ -2129,7 +2167,7 @@ def test_resolve_content_format_hf_defined(model, expected_format):
tokenizer,
chat_template=None,
tools=None,
model_config=renderer_config.model_config,
model_config=model_config,
)
assert isinstance(chat_template, str)
@@ -2143,7 +2181,7 @@ def test_resolve_content_format_hf_defined(model, expected_format):
None,
"auto",
tokenizer,
renderer_config=renderer_config,
model_config=model_config,
)
assert resolved_format == expected_format
@@ -2165,11 +2203,23 @@ def test_resolve_content_format_fallbacks(model, expected_format):
model_info = HF_EXAMPLE_MODELS.find_hf_info(model)
model_info.check_available_online(on_fail="skip")
renderer_config = model_info.build_renderer_config(model)
model_config = ModelConfig(
model,
tokenizer=model_info.tokenizer or model,
tokenizer_mode=model_info.tokenizer_mode,
revision=model_info.revision,
trust_remote_code=model_info.trust_remote_code,
hf_overrides=model_info.hf_overrides,
skip_tokenizer_init=model_info.require_embed_inputs,
enable_prompt_embeds=model_info.require_embed_inputs,
enable_mm_embeds=model_info.require_embed_inputs,
enforce_eager=model_info.enforce_eager,
dtype=model_info.dtype,
)
tokenizer = get_tokenizer(
renderer_config.tokenizer,
trust_remote_code=renderer_config.trust_remote_code,
model_config.tokenizer,
trust_remote_code=model_config.trust_remote_code,
)
# Test detecting the tokenizer's chat_template
@@ -2177,7 +2227,7 @@ def test_resolve_content_format_fallbacks(model, expected_format):
tokenizer,
chat_template=None,
tools=None,
model_config=renderer_config.model_config,
model_config=model_config,
)
assert isinstance(chat_template, str)
@@ -2191,7 +2241,7 @@ def test_resolve_content_format_fallbacks(model, expected_format):
None,
"auto",
tokenizer,
renderer_config=renderer_config,
model_config=model_config,
)
assert resolved_format == expected_format
@@ -2222,13 +2272,15 @@ def test_resolve_content_format_fallbacks(model, expected_format):
],
)
def test_resolve_content_format_examples(template_path, expected_format):
model = PHI3V_MODEL_ID # Dummy
model_config = ModelConfig(model, trust_remote_code=True)
renderer_config = RendererConfig(model_config=model_config, tokenizer=model)
model_config = ModelConfig(
PHI3V_MODEL_ID, # Dummy
tokenizer=PHI3V_MODEL_ID, # Dummy
trust_remote_code=True,
)
dummy_tokenizer = get_tokenizer(
renderer_config.tokenizer,
trust_remote_code=renderer_config.trust_remote_code,
PHI3V_MODEL_ID, # Dummy
trust_remote_code=model_config.trust_remote_code,
)
dummy_tokenizer.chat_template = None
@@ -2245,7 +2297,7 @@ def test_resolve_content_format_examples(template_path, expected_format):
None,
"auto",
dummy_tokenizer,
renderer_config=renderer_config,
model_config=model_config,
)
assert resolved_format == expected_format
@@ -2280,7 +2332,7 @@ def test_parse_chat_messages_include_thinking_chunk(mistral_model_config):
conversation_with_thinking, _, _ = parse_chat_messages(
messages,
RendererConfig(model_config=mistral_model_config),
mistral_model_config,
content_format="openai",
)
@@ -2380,7 +2432,7 @@ def test_parse_chat_messages_single_empty_audio_with_uuid(
],
}
],
RendererConfig(model_config=qwen2_audio_model_config),
qwen2_audio_model_config,
content_format="string",
)
@@ -2414,7 +2466,7 @@ async def test_parse_chat_messages_single_empty_audio_with_uuid_async(
],
}
],
RendererConfig(model_config=qwen2_audio_model_config),
qwen2_audio_model_config,
content_format="string",
)

14
tests/lora/test_lora_manager.py
View File

@@ -8,7 +8,7 @@ import torch
from safetensors.torch import load_file
from torch import nn
from vllm.config import ModelConfig, RendererConfig, VllmConfig
from vllm.config import ModelConfig, VllmConfig
from vllm.config.lora import LoRAConfig
from vllm.lora.layers import (
ColumnParallelLinearWithLoRA,
@@ -422,11 +422,7 @@ def test_lru_cache_worker_adapter_manager(dist_init, dummy_model, device, tmp_pa
)
model_config = ModelConfig(max_model_len=16)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
lora_config=lora_config,
)
vllm_config = VllmConfig(model_config=model_config, lora_config=lora_config)
vllm_config.scheduler_config.max_num_seqs = 4
vllm_config.scheduler_config.max_num_batched_tokens = 2
@@ -529,11 +525,7 @@ def test_worker_adapter_manager(dist_init, dummy_model_gate_up, device, tmp_path
)
model_config = ModelConfig(max_model_len=16)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
lora_config=lora_config,
)
vllm_config = VllmConfig(model_config=model_config, lora_config=lora_config)
vllm_config.scheduler_config.max_num_seqs = 4
vllm_config.scheduler_config.max_num_batched_tokens = 2

2
tests/lora/test_worker.py
View File

@@ -11,7 +11,6 @@ from vllm.config import (
DeviceConfig,
ModelConfig,
ParallelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
)
@@ -44,7 +43,6 @@ def test_worker_apply_lora(qwen3_lora_files):
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
load_config=LoadConfig(
download_dir=None,
load_format="dummy",

22
tests/model_executor/test_model_load_with_params.py
View File

@@ -42,10 +42,8 @@ def test_model_loading_with_params(vllm_runner, monkeypatch):
"Write a short story about a robot that dreams for the first time.\n"
)
llm_engine = vllm_model.llm.llm_engine
model_config = llm_engine.model_config
renderer_config = llm_engine.renderer_config
tokenizer = llm_engine.tokenizer
model_config = vllm_model.llm.llm_engine.model_config
model_tokenizer = vllm_model.llm.llm_engine.tokenizer
# asserts on the bert model config file
assert model_config.encoder_config["max_seq_length"] == 512
@@ -56,8 +54,8 @@ def test_model_loading_with_params(vllm_runner, monkeypatch):
assert model_config.pooler_config.normalize
# asserts on the tokenizer loaded
assert renderer_config.tokenizer == "BAAI/bge-base-en-v1.5"
assert tokenizer.model_max_length == 512
assert model_config.tokenizer == "BAAI/bge-base-en-v1.5"
assert model_tokenizer.model_max_length == 512
def check_model(model):
assert isinstance(model, BertEmbeddingModel)
@@ -88,10 +86,8 @@ def test_roberta_model_loading_with_params(vllm_runner, monkeypatch):
"Write a short story about a robot that dreams for the first time.\n"
)
llm_engine = vllm_model.llm.llm_engine
model_config = llm_engine.model_config
renderer_config = llm_engine.renderer_config
tokenizer = llm_engine.tokenizer
model_config = vllm_model.llm.llm_engine.model_config
model_tokenizer = vllm_model.llm.llm_engine.tokenizer
# asserts on the bert model config file
assert model_config.encoder_config["max_seq_length"] == 512
@@ -102,8 +98,8 @@ def test_roberta_model_loading_with_params(vllm_runner, monkeypatch):
assert model_config.pooler_config.normalize
# asserts on the tokenizer loaded
assert renderer_config.tokenizer == "intfloat/multilingual-e5-base"
assert tokenizer.model_max_length == 512
assert model_config.tokenizer == "intfloat/multilingual-e5-base"
assert model_tokenizer.model_max_length == 512
def check_model(model):
assert isinstance(model, RobertaEmbeddingModel)
@@ -132,7 +128,7 @@ def test_facebook_roberta_model_loading_with_params(vllm_runner, monkeypatch):
"Write a short story about a robot that dreams for the first time.\n"
)
assert vllm_model.llm.llm_engine.renderer_config.tokenizer == model_name
assert vllm_model.llm.llm_engine.model_config.tokenizer == model_name
def check_model(model):
assert isinstance(model, RobertaEmbeddingModel)

5
tests/models/language/pooling/test_gritlm.py
View File

@@ -6,7 +6,7 @@ import pytest
from scipy.spatial.distance import cosine
from vllm import LLM, SamplingParams
from vllm.config import ModelConfig, RendererConfig
from vllm.config import ModelConfig
from ....utils import RemoteOpenAIServer
@@ -31,8 +31,7 @@ def test_find_array():
dtype="bfloat16",
seed=0,
)
renderer_config = RendererConfig(model_config=model_config)
pooling = GritLMMeanPool(renderer_config=renderer_config)
pooling = GritLMMeanPool(model_config=model_config)
arr = _arr([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])

22
tests/models/multimodal/processing/test_common.py
View File

@@ -25,6 +25,7 @@ from vllm.multimodal.processing import BaseMultiModalProcessor, InputProcessingC
from vllm.tokenizers import (
MistralTokenizer,
TokenizerLike,
cached_tokenizer_from_config,
)
from ....multimodal.utils import random_audio, random_image, random_video
@@ -211,20 +212,31 @@ def _test_processing_correctness(
else:
model_info = HF_EXAMPLE_MODELS.find_hf_info(model_id_or_arch)
model_id = model_id_or_arch
model_info.check_available_online(on_fail="skip")
model_info.check_transformers_version(on_fail="skip")
renderer_config = model_info.build_renderer_config(
model=model_id,
model_config = ModelConfig(
model_id,
tokenizer=model_info.tokenizer or model_id,
tokenizer_mode=model_info.tokenizer_mode,
revision=model_info.revision,
trust_remote_code=model_info.trust_remote_code,
hf_overrides=model_info.hf_overrides,
# Ensure that the cache can fit all of the data
mm_processor_cache_gb=2048,
skip_tokenizer_init=model_info.require_embed_inputs,
enable_prompt_embeds=model_info.require_embed_inputs,
enable_mm_embeds=model_info.require_embed_inputs,
enforce_eager=model_info.enforce_eager,
dtype=model_info.dtype,
)
model_config = renderer_config.model_config
model_cls = MULTIMODAL_REGISTRY._get_model_cls(model_config)
factories = model_cls._processor_factory
ctx = InputProcessingContext.from_config(renderer_config)
ctx = InputProcessingContext(
model_config,
tokenizer=cached_tokenizer_from_config(model_config),
)
cache = MultiModalProcessorOnlyCache(model_config)
processing_info = factories.info(ctx)

4
tests/models/multimodal/processing/test_glm4_1v.py
View File

@@ -40,7 +40,7 @@ def test_processor_override(
mm_processor_kwargs=None,
limit_mm_per_prompt={"video": 1},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
tokenizer = processor.info.get_tokenizer()
hf_processor_mm_kwargs = {"fps": fps}
@@ -79,7 +79,7 @@ def test_video_loader_consistency(
mm_processor_kwargs=None,
limit_mm_per_prompt={"video": 1},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
hf_processor_mm_kwargs = {"fps": fps}
# Build the image str / prompt based on the number of images we pass

2
tests/models/multimodal/processing/test_h2ovl.py
View File

@@ -162,7 +162,7 @@ def test_processor_override(
mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
limit_mm_per_prompt={"image": len(size_factors)},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
hf_processor_mm_kwargs = {} if kwargs_on_init else mm_processor_kwargs
min_num = min_dynamic_patch if dynamic_image_size else 1

2
tests/models/multimodal/processing/test_idefics3.py
View File

@@ -38,7 +38,7 @@ def test_processor_override(
mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
hf_processor_mm_kwargs = {} if kwargs_on_init else mm_processor_kwargs
# Build the image str / prompt based on the number of images we pass

2
tests/models/multimodal/processing/test_internvl.py
View File

@@ -116,7 +116,7 @@ def test_processor_override(
mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
limit_mm_per_prompt={"image": len(size_factors)},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
hf_processor_mm_kwargs = {} if kwargs_on_init else mm_processor_kwargs
min_num = min_dynamic_patch if dynamic_image_size else 1

2
tests/models/multimodal/processing/test_llama4.py
View File

@@ -30,7 +30,7 @@ def test_processor_override(
limit_mm_per_prompt={"image": num_imgs},
mm_processor_cache_gb=mm_processor_cache_gb,
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
config = processor.info.get_hf_config()
tokenizer = processor.info.get_tokenizer()
hf_processor = processor.info.get_hf_processor()

6
tests/models/multimodal/processing/test_llava_next.py
View File

@@ -42,7 +42,7 @@ def test_processor_max_tokens(model_id):
mm_processor_kwargs=None,
limit_mm_per_prompt={"image": 1},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
info = processor.info
seen_aspect_ratios = set[float]()
@@ -140,7 +140,7 @@ def test_processor_prompt_replacements_regression(model_id, num_imgs):
mm_processor_kwargs=None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
image_ratios = [
(171, 152),
@@ -173,7 +173,7 @@ def test_processor_prompt_replacements_all(model_id, num_imgs):
mm_processor_kwargs=None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
seen_aspect_ratios = set[float]()
image_sizes = list[ImageSize]()

6
tests/models/multimodal/processing/test_llava_onevision.py
View File

@@ -42,7 +42,7 @@ def test_processor_max_tokens(model_id):
mm_processor_kwargs=None,
limit_mm_per_prompt={"image": 1},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
info = processor.info
seen_aspect_ratios = set[float]()
@@ -138,7 +138,7 @@ def test_processor_prompt_replacements_regression(model_id, num_imgs):
mm_processor_kwargs=None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
image_ratios = [
(171, 152),
@@ -171,7 +171,7 @@ def test_processor_prompt_replacements_all(model_id, num_imgs):
mm_processor_kwargs=None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
seen_aspect_ratios = set[float]()
image_sizes = list[ImageSize]()

4
tests/models/multimodal/processing/test_minimax_vl_01.py
View File

@@ -24,7 +24,7 @@ def test_processor_override(
mm_processor_kwargs=None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
prompt = "<image>" * num_imgs
image = Image.new("RGB", size=(364, 364))
mm_data = {"image": [image] * num_imgs}
@@ -83,7 +83,7 @@ def test_processor_prompt_replacements_regression(model_id, num_imgs):
mm_processor_kwargs=None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
image_ratios = [
(171, 152),

2
tests/models/multimodal/processing/test_mllama4.py
View File

@@ -25,7 +25,7 @@ def test_profiling(model_id: str, max_model_len: int):
limit_mm_per_prompt=mm_counts,
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
profiler = MultiModalProfiler(processor)
decoder_dummy_data = profiler.get_decoder_dummy_data(

2
tests/models/multimodal/processing/test_nemotron_vl.py
View File

@@ -118,7 +118,7 @@ def test_processor_override(
mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
limit_mm_per_prompt={"image": len(size_factors)},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
hf_processor_mm_kwargs = {} if kwargs_on_init else mm_processor_kwargs
min_num = min_dynamic_patch if dynamic_image_size else 1

2
tests/models/multimodal/processing/test_phi3v.py
View File

@@ -39,7 +39,7 @@ def test_processor_override(
mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
hf_processor_mm_kwargs = {} if kwargs_on_init else mm_processor_kwargs
# Build the image str / prompt based on the number of images we pass

2
tests/models/multimodal/processing/test_phi4mm.py
View File

@@ -39,7 +39,7 @@ def test_processor_override(
mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
hf_processor_mm_kwargs = {} if kwargs_on_init else mm_processor_kwargs
# Build the image str / prompt based on the number of images we pass

2
tests/models/multimodal/processing/test_qwen2_vl.py
View File

@@ -34,7 +34,7 @@ def test_processor_override(
mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
tokenizer = processor.info.get_tokenizer()
hf_processor_mm_kwargs = {} if kwargs_on_init else mm_processor_kwargs

2
tests/models/multimodal/processing/test_smolvlm.py
View File

@@ -38,7 +38,7 @@ def test_processor_override(
mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
limit_mm_per_prompt={"image": num_imgs},
)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
hf_processor_mm_kwargs = {} if kwargs_on_init else mm_processor_kwargs
# Build the image str / prompt based on the number of images we pass

24
tests/models/multimodal/processing/test_tensor_schema.py
View File

@@ -11,7 +11,7 @@ import pytest
import torch.nn as nn
from PIL import Image
from vllm.config import ModelConfig, RendererConfig, VllmConfig, set_current_vllm_config
from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config
from vllm.config.multimodal import (
AudioDummyOptions,
BaseDummyOptions,
@@ -31,6 +31,7 @@ from vllm.multimodal import MULTIMODAL_REGISTRY, BatchedTensorInputs
from vllm.multimodal.processing import BaseMultiModalProcessor, InputProcessingContext
from vllm.multimodal.utils import group_mm_kwargs_by_modality
from vllm.platforms import current_platform
from vllm.tokenizers import cached_tokenizer_from_config
from vllm.utils.collection_utils import is_list_of
from vllm.utils.torch_utils import set_default_torch_dtype
@@ -149,10 +150,7 @@ def initialize_dummy_model(
backend="nccl",
)
initialize_model_parallel(tensor_model_parallel_size=1)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
)
vllm_config = VllmConfig(model_config=model_config)
with set_current_vllm_config(vllm_config=vllm_config):
with set_default_torch_dtype(model_config.dtype):
model = model_cls(vllm_config=vllm_config)
@@ -184,12 +182,19 @@ def test_model_tensor_schema(model_id: str):
else:
dtype = model_info.dtype
renderer_config = model_info.build_renderer_config(
model_config = ModelConfig(
model_id,
tokenizer=model_info.tokenizer or model_id,
tokenizer_mode=model_info.tokenizer_mode,
revision=model_info.revision,
trust_remote_code=model_info.trust_remote_code,
hf_overrides=hf_overrides_fn,
skip_tokenizer_init=model_info.require_embed_inputs,
enable_prompt_embeds=model_info.require_embed_inputs,
enable_mm_embeds=model_info.require_embed_inputs,
enforce_eager=model_info.enforce_eager,
dtype=dtype,
)
model_config = renderer_config.model_config
model_cls = MULTIMODAL_REGISTRY._get_model_cls(model_config)
assert supports_multimodal(model_cls)
@@ -207,7 +212,10 @@ def test_model_tensor_schema(model_id: str):
if not any(inputs_parse_methods):
pytest.skip(f"{model_arch} does not support tensor schema validation.")
ctx = InputProcessingContext.from_config(renderer_config)
ctx = InputProcessingContext(
model_config,
tokenizer=cached_tokenizer_from_config(model_config),
)
processing_info = factories.info(ctx)
supported_mm_limits = processing_info.get_supported_mm_limits()
limit_mm_per_prompt = {

5
tests/models/multimodal/processing/test_transformers.py
View File

@@ -3,7 +3,7 @@
import pytest
from vllm.assets.image import ImageAsset
from vllm.config import ModelConfig, RendererConfig
from vllm.config import ModelConfig
from vllm.multimodal import MULTIMODAL_REGISTRY
@@ -13,9 +13,8 @@ def test_multimodal_processor(model_id):
model=model_id,
model_impl="transformers",
)
renderer_config = RendererConfig(model_config=model_config)
mm_processor = MULTIMODAL_REGISTRY.create_processor(renderer_config)
mm_processor = MULTIMODAL_REGISTRY.create_processor(model_config)
image_pil = ImageAsset("cherry_blossom").pil_image
mm_data = {"image": image_pil}

33
tests/models/multimodal/test_mapping.py
View File

@@ -7,6 +7,7 @@ import torch
import transformers
from transformers import AutoConfig, PreTrainedModel
from vllm.config import ModelConfig
from vllm.model_executor.models.utils import WeightsMapper
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.transformers_utils.config import try_get_safetensors_metadata
@@ -49,11 +50,37 @@ def test_hf_model_weights_mapper(model_arch: str):
model_info.check_available_online(on_fail="skip")
model_info.check_transformers_version(on_fail="skip")
model_config = model_info.build_model_config(config_format="hf")
is_mistral_model = model_arch in [
"Mistral3ForConditionalGeneration",
"PixtralForConditionalGeneration",
"VoxtralForConditionalGeneration",
]
if not is_mistral_model or model_info.tokenizer_mode == "mistral":
tokenizer_mode = model_info.tokenizer_mode
else:
tokenizer_mode = "hf"
model_id = model_info.default
model_config = ModelConfig(
model_id,
tokenizer=model_info.tokenizer or model_id,
tokenizer_mode=tokenizer_mode,
config_format="hf",
revision=model_info.revision,
trust_remote_code=model_info.trust_remote_code,
hf_overrides=model_info.hf_overrides,
skip_tokenizer_init=model_info.require_embed_inputs,
enable_prompt_embeds=model_info.require_embed_inputs,
enable_mm_embeds=model_info.require_embed_inputs,
enforce_eager=model_info.enforce_eager,
dtype=model_info.dtype,
)
model_cls = MULTIMODAL_REGISTRY._get_model_cls(model_config)
original_weights = create_repo_dummy_weights(model_config.model)
hf_dummy_model = create_dummy_model(model_config.model, model_arch)
original_weights = create_repo_dummy_weights(model_id)
hf_dummy_model = create_dummy_model(model_id, model_arch)
hf_converted_weights = hf_dummy_model.named_parameters()
hf_converted_buffers = hf_dummy_model.named_buffers()
mapper: WeightsMapper = model_cls.hf_to_vllm_mapper

33
tests/models/registry.py
View File

@@ -9,8 +9,7 @@ import pytest
from packaging.version import Version
from transformers import __version__ as TRANSFORMERS_VERSION
from vllm.config.model import ModelConfig, ModelDType
from vllm.config.renderer import RendererConfig, TokenizerMode
from vllm.config.model import ModelDType, TokenizerMode
@dataclass(frozen=True)
@@ -171,36 +170,6 @@ class _HfExamplesInfo:
else:
pytest.skip(msg)
def build_model_config(self, model: str | None = None, **kwargs) -> ModelConfig:
if model is None:
model = self.default
return ModelConfig(
**{
"model": model,
"revision": self.revision,
"trust_remote_code": self.trust_remote_code,
"hf_overrides": self.hf_overrides,
"enable_prompt_embeds": self.require_embed_inputs,
"enable_mm_embeds": self.require_embed_inputs,
"enforce_eager": self.enforce_eager,
"dtype": self.dtype,
**kwargs,
}
)
def build_renderer_config(
self, model: str | None = None, **kwargs
) -> RendererConfig:
model_config = self.build_model_config(model, **kwargs)
return RendererConfig(
model_config=model_config,
tokenizer=self.tokenizer or model_config.model,
tokenizer_mode=self.tokenizer_mode,
skip_tokenizer_init=self.require_embed_inputs,
)
_TEXT_GENERATION_EXAMPLE_MODELS = {
# [Decoder-only]

17
tests/models/utils.py
View File

@@ -13,6 +13,7 @@ from transformers import PretrainedConfig
from vllm.config.model import ModelConfig, ModelDType, RunnerOption
from vllm.logprobs import Logprob, PromptLogprobs, SampleLogprobs
from vllm.multimodal.processing import InputProcessingContext
from vllm.tokenizers import cached_tokenizer_from_config
from .. import ci_envs
from .registry import HF_EXAMPLE_MODELS
@@ -295,18 +296,30 @@ def build_model_context(
model_config_kwargs = model_config_kwargs or {}
limit_mm_per_prompt = limit_mm_per_prompt or {}
renderer_config = model_info.build_renderer_config(
model_config = ModelConfig(
model_id,
runner=runner,
tokenizer=model_info.tokenizer or model_id,
tokenizer_mode=model_info.tokenizer_mode,
revision=model_info.revision,
trust_remote_code=model_info.trust_remote_code,
dtype=dtype,
seed=0,
mm_processor_kwargs=mm_processor_kwargs,
limit_mm_per_prompt=limit_mm_per_prompt,
mm_processor_cache_gb=mm_processor_cache_gb,
hf_overrides=model_info.hf_overrides,
skip_tokenizer_init=model_info.require_embed_inputs,
enable_prompt_embeds=model_info.require_embed_inputs,
enable_mm_embeds=model_info.require_embed_inputs,
enforce_eager=model_info.enforce_eager,
**model_config_kwargs,
)
return InputProcessingContext.from_config(renderer_config)
return InputProcessingContext(
model_config,
tokenizer=cached_tokenizer_from_config(model_config),
)
def check_embeddings_close(

27
tests/multimodal/test_cache.py
View File

@@ -6,7 +6,7 @@ import numpy as np
import pytest
import torch
from vllm.config import ModelConfig, ParallelConfig, RendererConfig, VllmConfig
from vllm.config import ModelConfig, ParallelConfig, VllmConfig
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.cache import (
BaseMultiModalProcessorCache,
@@ -110,14 +110,11 @@ def _create_vllm_config(
mm_processor_cache_gb: float,
enable_ipc: bool,
):
model_config = ModelConfig(
model="llava-hf/llava-onevision-qwen2-0.5b-ov-hf",
mm_processor_cache_gb=mm_processor_cache_gb,
)
return VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
model_config=ModelConfig(
model="llava-hf/llava-onevision-qwen2-0.5b-ov-hf",
mm_processor_cache_gb=mm_processor_cache_gb,
),
parallel_config=ParallelConfig(data_parallel_size=1 if enable_ipc else 2),
)
@@ -509,15 +506,13 @@ def _run_test_cache_eviction_shm(
def test_cache_eviction_shm_cache():
model_config = ModelConfig(
model="llava-hf/llava-onevision-qwen2-0.5b-ov-hf",
mm_processor_cache_type="shm",
mm_shm_cache_max_object_size_mb=6,
mm_processor_cache_gb=15.2 * MiB_bytes / GiB_bytes,
)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
model_config=ModelConfig(
model="llava-hf/llava-onevision-qwen2-0.5b-ov-hf",
mm_processor_cache_type="shm",
mm_shm_cache_max_object_size_mb=6,
mm_processor_cache_gb=15.2 * MiB_bytes / GiB_bytes,
),
)
sender_cache = ShmObjectStoreSenderCache(vllm_config)
receiver_cache = ShmObjectStoreReceiverCache(vllm_config, mp.Lock())

24
tests/multimodal/test_processing.py
View File

@@ -7,7 +7,7 @@ from contextlib import nullcontext
import numpy as np
import pytest
from vllm.config import ModelConfig, RendererConfig
from vllm.config import ModelConfig
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.processing import (
InputProcessingContext,
@@ -920,9 +920,8 @@ def test_limit_mm_per_prompt_dummy(model_id, limit, num_supported, is_valid):
model=model_id,
limit_mm_per_prompt=limit_mm_per_prompt,
)
renderer_config = RendererConfig(model_config=model_config)
processor = MULTIMODAL_REGISTRY.create_processor(renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(model_config)
processor._supported_mm_limits = {"image": num_supported}
profiler = MultiModalProfiler(processor)
@@ -956,9 +955,8 @@ def test_limit_mm_per_prompt_apply(model_id, num_images, limit, is_valid):
model=model_id,
limit_mm_per_prompt=limit_mm_per_prompt,
)
renderer_config = RendererConfig(model_config=model_config)
processor = MULTIMODAL_REGISTRY.create_processor(renderer_config)
processor = MULTIMODAL_REGISTRY.create_processor(model_config)
rng = np.random.RandomState(0)
image = random_image(rng, min_wh=128, max_wh=256)
@@ -1014,13 +1012,11 @@ def test_hf_processor_init_kwargs(
inference_kwargs,
expected_kwargs,
):
model_config = ModelConfig(model_id, mm_processor_kwargs=config_kwargs)
renderer_config = RendererConfig(
model_config=model_config,
tokenizer=model_id,
ctx = InputProcessingContext(
model_config=ModelConfig(model_id, mm_processor_kwargs=config_kwargs),
tokenizer=None,
)
ctx = InputProcessingContext.from_config(renderer_config)
processor = ctx.get_hf_processor(
DummyProcessor, # type: ignore[arg-type]
**inference_kwargs,
@@ -1049,13 +1045,11 @@ def test_hf_processor_call_kwargs(
inference_kwargs,
expected_kwargs,
):
model_config = ModelConfig(model_id, mm_processor_kwargs=config_kwargs)
renderer_config = RendererConfig(
model_config=model_config,
tokenizer=model_id,
ctx = InputProcessingContext(
model_config=ModelConfig(model_id, mm_processor_kwargs=config_kwargs),
tokenizer=None,
)
ctx = InputProcessingContext.from_config(renderer_config)
processor = ctx.get_hf_processor(DummyProcessor) # type: ignore[arg-type]
result = ctx.call_hf_processor(processor, {}, inference_kwargs)

4
tests/multimodal/test_registry.py
View File

@@ -31,6 +31,4 @@ def test_supports_multimodal_inputs(model_id, limit_mm_per_prompt, expected):
model_id,
limit_mm_per_prompt=limit_mm_per_prompt,
)
assert (
MULTIMODAL_REGISTRY.supports_multimodal_inputs(ctx.renderer_config) is expected
)
assert MULTIMODAL_REGISTRY.supports_multimodal_inputs(ctx.model_config) is expected

131
tests/test_config.py
View File

@@ -13,7 +13,6 @@ from vllm.config import (
CompilationConfig,
ModelConfig,
PoolerConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
update_config,
@@ -477,41 +476,27 @@ def test_load_config_pt_load_map_location(pt_load_map_location):
("deepseek-ai/DeepSeek-R1-Distill-Qwen-7B", 131073, 131072, True),
],
)
def test_recalculate_max_model_len(
def test_get_and_verify_max_len(
model_id, max_model_len, expected_max_len, should_raise
):
"""Test recalculate_max_model_len with different configurations."""
"""Test get_and_verify_max_len with different configurations."""
model_config = ModelConfig(model_id)
if should_raise:
with pytest.raises(ValueError):
model_config.recalculate_max_model_len(
max_model_len,
tokenizer=model_id,
tokenizer_revision=None,
)
model_config.get_and_verify_max_len(max_model_len)
else:
model_config.recalculate_max_model_len(
max_model_len,
tokenizer=model_id,
tokenizer_revision=None,
)
assert model_config.max_model_len == expected_max_len
actual_max_len = model_config.get_and_verify_max_len(max_model_len)
assert actual_max_len == expected_max_len
class MockModelConfig:
"""Simple mock object for testing maybe_pull_model_for_runai"""
class MockConfig:
"""Simple mock object for testing maybe_pull_model_tokenizer_for_runai"""
def __init__(self, model: str):
def __init__(self, model: str, tokenizer: str):
self.model = model
class MockRendererConfig:
"""Simple mock object for testing maybe_pull_tokenizer_for_runai"""
def __init__(self, model_config: MockModelConfig):
self.model_config = model_config
self.tokenizer = model_config.model
self.tokenizer = tokenizer
self.model_weights = None
@pytest.mark.parametrize(
@@ -529,65 +514,59 @@ def test_s3_url_model_tokenizer_paths(mock_pull_files, s3_url):
mock_pull_files.return_value = None
# Create first mock and run the method
model_config1 = MockModelConfig(model=s3_url)
renderer_config1 = MockRendererConfig(model_config=model_config1)
ModelConfig.maybe_pull_model_for_runai(model_config1, s3_url)
RendererConfig.maybe_pull_tokenizer_for_runai(renderer_config1, s3_url)
config1 = MockConfig(model=s3_url, tokenizer=s3_url)
ModelConfig.maybe_pull_model_tokenizer_for_runai(config1, s3_url, s3_url)
# Check that model and tokenizer point to existing directories
assert os.path.exists(model_config1.model), (
f"Model directory does not exist: {model_config1.model}"
assert os.path.exists(config1.model), (
f"Model directory does not exist: {config1.model}"
)
assert os.path.isdir(model_config1.model), (
f"Model path is not a directory: {model_config1.model}"
assert os.path.isdir(config1.model), (
f"Model path is not a directory: {config1.model}"
)
assert os.path.exists(renderer_config1.tokenizer), (
f"Tokenizer directory does not exist: {renderer_config1.tokenizer}"
assert os.path.exists(config1.tokenizer), (
f"Tokenizer directory does not exist: {config1.tokenizer}"
)
assert os.path.isdir(renderer_config1.tokenizer), (
f"Tokenizer path is not a directory: {renderer_config1.tokenizer}"
assert os.path.isdir(config1.tokenizer), (
f"Tokenizer path is not a directory: {config1.tokenizer}"
)
# Verify that the paths are different from the original S3 URL
assert model_config1.model != s3_url, (
"Model path should be converted to local directory"
)
assert renderer_config1.tokenizer != s3_url, (
assert config1.model != s3_url, "Model path should be converted to local directory"
assert config1.tokenizer != s3_url, (
"Tokenizer path should be converted to local directory"
)
# Store the original paths
created_model_dir = model_config1.model
create_tokenizer_dir = renderer_config1.tokenizer
created_model_dir = config1.model
create_tokenizer_dir = config1.tokenizer
# Create a new mock and run the method with the same S3 URL
model_config2 = MockModelConfig(model=s3_url)
renderer_config2 = MockRendererConfig(model_config=model_config2)
ModelConfig.maybe_pull_model_for_runai(model_config2, s3_url)
RendererConfig.maybe_pull_tokenizer_for_runai(renderer_config2, s3_url)
config2 = MockConfig(model=s3_url, tokenizer=s3_url)
ModelConfig.maybe_pull_model_tokenizer_for_runai(config2, s3_url, s3_url)
# Check that the new directories exist
assert os.path.exists(model_config2.model), (
f"Model directory does not exist: {model_config2.model}"
assert os.path.exists(config2.model), (
f"Model directory does not exist: {config2.model}"
)
assert os.path.isdir(model_config2.model), (
f"Model path is not a directory: {model_config2.model}"
assert os.path.isdir(config2.model), (
f"Model path is not a directory: {config2.model}"
)
assert os.path.exists(renderer_config2.tokenizer), (
f"Tokenizer directory does not exist: {renderer_config2.tokenizer}"
assert os.path.exists(config2.tokenizer), (
f"Tokenizer directory does not exist: {config2.tokenizer}"
)
assert os.path.isdir(renderer_config2.tokenizer), (
f"Tokenizer path is not a directory: {renderer_config2.tokenizer}"
assert os.path.isdir(config2.tokenizer), (
f"Tokenizer path is not a directory: {config2.tokenizer}"
)
# Verify that the paths are deterministic (same as before)
assert model_config2.model == created_model_dir, (
assert config2.model == created_model_dir, (
f"Model paths are not deterministic. "
f"Original: {created_model_dir}, New: {model_config2.model}"
f"Original: {created_model_dir}, New: {config2.model}"
)
assert renderer_config2.tokenizer == create_tokenizer_dir, (
assert config2.tokenizer == create_tokenizer_dir, (
f"Tokenizer paths are not deterministic. "
f"Original: {create_tokenizer_dir}, New: {renderer_config2.tokenizer}"
f"Original: {create_tokenizer_dir}, New: {config2.tokenizer}"
)
@@ -601,36 +580,28 @@ def test_s3_url_different_models_create_different_directories(mock_pull_files):
s3_url2 = "s3://example-bucket-2/model/"
# Create mocks with different S3 URLs and run the method
model_config1 = MockModelConfig(model=s3_url1)
renderer_config1 = MockRendererConfig(model_config=model_config1)
ModelConfig.maybe_pull_model_for_runai(model_config1, s3_url1)
RendererConfig.maybe_pull_tokenizer_for_runai(renderer_config1, s3_url1)
config1 = MockConfig(model=s3_url1, tokenizer=s3_url1)
ModelConfig.maybe_pull_model_tokenizer_for_runai(config1, s3_url1, s3_url1)
model_config2 = MockModelConfig(model=s3_url2)
renderer_config2 = MockRendererConfig(model_config=model_config2)
ModelConfig.maybe_pull_model_for_runai(model_config2, s3_url2)
RendererConfig.maybe_pull_tokenizer_for_runai(renderer_config2, s3_url2)
config2 = MockConfig(model=s3_url2, tokenizer=s3_url2)
ModelConfig.maybe_pull_model_tokenizer_for_runai(config2, s3_url2, s3_url2)
# Verify that different URLs produce different directories
assert model_config1.model != model_config2.model, (
assert config1.model != config2.model, (
f"Different S3 URLs should create different model directories. "
f"URL1 model: {model_config1.model}, URL2 model: {model_config2.model}"
f"URL1 model: {config1.model}, URL2 model: {config2.model}"
)
assert renderer_config1.tokenizer != renderer_config2.tokenizer, (
assert config1.tokenizer != config2.tokenizer, (
f"Different S3 URLs should create different tokenizer directories. "
f"URL1 tokenizer: {renderer_config1.tokenizer}, "
f"URL2 tokenizer: {renderer_config2.tokenizer}"
f"URL1 tokenizer: {config1.tokenizer}, "
f"URL2 tokenizer: {config2.tokenizer}"
)
# Verify that both sets of directories exist
assert os.path.exists(model_config1.model) and os.path.isdir(model_config1.model)
assert os.path.exists(renderer_config1.tokenizer) and os.path.isdir(
renderer_config1.tokenizer
)
assert os.path.exists(model_config2.model) and os.path.isdir(model_config2.model)
assert os.path.exists(renderer_config2.tokenizer) and os.path.isdir(
renderer_config2.tokenizer
)
assert os.path.exists(config1.model) and os.path.isdir(config1.model)
assert os.path.exists(config1.tokenizer) and os.path.isdir(config1.tokenizer)
assert os.path.exists(config2.model) and os.path.isdir(config2.model)
assert os.path.exists(config2.tokenizer) and os.path.isdir(config2.tokenizer)
@pytest.mark.parametrize(

7
tests/test_inputs.py
View File

@@ -3,7 +3,7 @@
import pytest
from vllm.config import ModelConfig, RendererConfig
from vllm.config import ModelConfig
from vllm.inputs import zip_enc_dec_prompts
from vllm.inputs.parse import parse_raw_prompts
from vllm.inputs.preprocess import InputPreprocessor
@@ -108,9 +108,8 @@ def test_zip_enc_dec_prompts(mm_processor_kwargs, expected_mm_kwargs):
)
def test_preprocessor_always_mm_code_path(model_id, prompt):
model_config = ModelConfig(model=model_id)
renderer_config = RendererConfig(model_config=model_config)
tokenizer = init_tokenizer_from_config(renderer_config)
input_preprocessor = InputPreprocessor(renderer_config, tokenizer)
tokenizer = init_tokenizer_from_config(model_config)
input_preprocessor = InputPreprocessor(model_config, tokenizer)
# HF processor adds sep token
sep_token_id = tokenizer.vocab[tokenizer.sep_token]

2
tests/v1/attention/utils.py
View File

@@ -16,7 +16,6 @@ from vllm.config import (
LoadConfig,
ModelConfig,
ParallelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
)
@@ -217,7 +216,6 @@ def create_vllm_config(
return VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
parallel_config=parallel_config,
scheduler_config=scheduler_config,

20
tests/v1/core/test_kv_cache_utils.py
View File

@@ -8,7 +8,7 @@ import pytest
import torch
import vllm.v1.core.kv_cache_utils as kv_cache_utils
from vllm.config import ModelConfig, RendererConfig, SchedulerConfig, VllmConfig
from vllm.config import ModelConfig, SchedulerConfig, VllmConfig
from vllm.lora.request import LoRARequest
from vllm.multimodal.inputs import (
MultiModalFeatureSpec,
@@ -667,10 +667,7 @@ def test_metrics_empty_stats():
def test_get_kv_cache_configs_multiple_workers():
model_config = ModelConfig(max_model_len=16)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
)
vllm_config = VllmConfig(model_config=model_config)
ref_kv_cache_spec = new_kv_cache_spec()
same_kv_cache_specs = [
@@ -1139,7 +1136,6 @@ def test_estimate_max_model_len(model_id, max_model_len, want_estimated_max_len)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
scheduler_config=scheduler_config,
)
@@ -1179,7 +1175,6 @@ def test_get_max_concurrency_for_kv_cache_config():
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
scheduler_config=scheduler_config,
)
@@ -1298,10 +1293,7 @@ def test_allocate_with_lookahead():
def test_get_kv_cache_config_one_worker():
# pass max_model_len to pass check_enough_kv_cache_memory
model_config = ModelConfig(max_model_len=16)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
)
vllm_config = VllmConfig(model_config=model_config)
mem_per_block_per_layer = 16 * 2 * 64 * 4 * 2
# all layers are full attention -> single group
@@ -1592,11 +1584,7 @@ def test_get_kv_cache_config_one_worker():
def test_get_kv_cache_configs_attention_free():
kv_cache_specs: dict[str, KVCacheSpec] = {}
model_config = ModelConfig(max_model_len=16)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
)
vllm_config = VllmConfig(model_config=ModelConfig(max_model_len=16))
kv_cache_configs = get_kv_cache_configs(vllm_config, [kv_cache_specs], [0])
assert kv_cache_configs == [
KVCacheConfig(

2
tests/v1/core/test_scheduler.py
View File

@@ -11,7 +11,6 @@ from vllm.config import (
ECTransferConfig,
KVTransferConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
SpeculativeConfig,
VllmConfig,
@@ -1564,7 +1563,6 @@ def create_scheduler_with_priority(
vllm_config = VllmConfig(
scheduler_config=scheduler_config,
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
kv_transfer_config=kv_transfer_config,
speculative_config=speculative_config,

2
tests/v1/core/utils.py
View File

@@ -9,7 +9,6 @@ from vllm.config import (
ECTransferConfig,
KVTransferConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
SpeculativeConfig,
VllmConfig,
@@ -133,7 +132,6 @@ def create_scheduler(
vllm_config = VllmConfig(
scheduler_config=scheduler_config,
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
kv_transfer_config=kv_transfer_config,
speculative_config=speculative_config,

2
tests/v1/engine/test_engine_core.py
View File

@@ -15,7 +15,6 @@ from vllm.config import (
ECTransferConfig,
KVTransferConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
)
@@ -523,7 +522,6 @@ def test_encoder_instance_zero_kv_cache(
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
scheduler_config=scheduler_config,
kv_transfer_config=kv_transfer_config,

24
tests/v1/engine/test_process_multi_modal_uuids.py
View File

@@ -5,14 +5,7 @@ import pytest
from vllm.assets.image import ImageAsset
from vllm.assets.video import VideoAsset
from vllm.config import (
CacheConfig,
DeviceConfig,
ModelConfig,
MultiModalConfig,
RendererConfig,
VllmConfig,
)
from vllm.config import CacheConfig, DeviceConfig, ModelConfig, VllmConfig
from vllm.sampling_params import SamplingParams
from vllm.v1.engine import input_processor as input_processor_mod
from vllm.v1.engine.input_processor import InputProcessor
@@ -51,21 +44,22 @@ def _mock_input_processor(
monkeypatch.setattr(VllmConfig, "__post_init__", lambda self: None, raising=True)
model_config = ModelConfig(
skip_tokenizer_init=True,
max_model_len=128,
mm_processor_cache_gb=mm_cache_gb,
generation_config="vllm",
)
model_config.multimodal_config = MultiModalConfig(mm_processor_cache_gb=mm_cache_gb)
renderer_config = RendererConfig(
model_config=model_config,
tokenizer="dummy",
skip_tokenizer_init=True,
)
# Minimal multimodal_config to satisfy references in
# Processor.process_inputs.
class _MockMMConfig:
def __init__(self, gb: float):
self.mm_processor_cache_gb = gb
model_config.multimodal_config = _MockMMConfig(mm_cache_gb) # type: ignore[attr-defined]
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=renderer_config,
cache_config=CacheConfig(enable_prefix_caching=enable_prefix_caching),
device_config=DeviceConfig(device="cpu"),
)

2
tests/v1/kv_connector/unit/utils.py
View File

@@ -15,7 +15,6 @@ from vllm.config import (
DeviceConfig,
KVTransferConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
)
@@ -128,7 +127,6 @@ def create_vllm_config(
return VllmConfig(
scheduler_config=scheduler_config,
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
kv_transfer_config=kv_transfer_config,
device_config=DeviceConfig("cpu"),

2
tests/v1/spec_decode/test_eagle.py
View File

@@ -19,7 +19,6 @@ from vllm.config import (
DeviceConfig,
ModelConfig,
ParallelConfig,
RendererConfig,
SchedulerConfig,
SpeculativeConfig,
VllmConfig,
@@ -62,7 +61,6 @@ def _create_proposer(
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=CacheConfig(),
speculative_config=speculative_config,
device_config=DeviceConfig(device=current_platform.device_type),

2
tests/v1/spec_decode/test_mtp.py
View File

@@ -18,7 +18,6 @@ from vllm.config import (
DeviceConfig,
ModelConfig,
ParallelConfig,
RendererConfig,
SchedulerConfig,
SpeculativeConfig,
VllmConfig,
@@ -47,7 +46,6 @@ def _create_mtp_proposer(num_speculative_tokens: int) -> EagleProposer:
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=CacheConfig(),
speculative_config=speculative_config,
device_config=DeviceConfig(device=current_platform.device_type),

2
tests/v1/spec_decode/test_ngram.py
View File

@@ -4,7 +4,6 @@ import numpy as np
from vllm.config import (
ModelConfig,
RendererConfig,
SpeculativeConfig,
VllmConfig,
)
@@ -70,7 +69,6 @@ def test_ngram_proposer():
return NgramProposer(
vllm_config=VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
speculative_config=SpeculativeConfig(
prompt_lookup_min=min_n,
prompt_lookup_max=max_n,

12
tests/v1/structured_output/test_backend_guidance.py
View File

@@ -6,7 +6,7 @@ from concurrent.futures import Future
import pytest
from transformers import AutoTokenizer
from vllm.config import RendererConfig, StructuredOutputsConfig, VllmConfig
from vllm.config import StructuredOutputsConfig, VllmConfig
from vllm.config.model import ModelConfig
from vllm.config.parallel import ParallelConfig
from vllm.config.speculative import SpeculativeConfig
@@ -72,11 +72,8 @@ def test_backend_guidance_rollback_terminated():
def test_grammar_bitmask_with_specdec():
tokenizer = AutoTokenizer.from_pretrained(TOKENIZER)
prompt = tokenizer.encode('{"a": "b"}')
model_config = ModelConfig(tokenizer=TOKENIZER)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config, tokenizer=TOKENIZER),
model_config=ModelConfig(tokenizer=TOKENIZER),
structured_outputs_config=StructuredOutputsConfig(backend="guidance"),
speculative_config=SpeculativeConfig(model="[ngram]", num_speculative_tokens=3),
)
@@ -140,11 +137,8 @@ def test_grammar_init_async_and_sync(async_grammar):
# Use "external_launcher" for sync mode, None for async mode
executor_backend = None if async_grammar else "external_launcher"
model_config = ModelConfig(tokenizer=TOKENIZER)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config, tokenizer=TOKENIZER),
model_config=ModelConfig(tokenizer=TOKENIZER),
structured_outputs_config=StructuredOutputsConfig(backend="guidance"),
parallel_config=ParallelConfig(distributed_executor_backend=executor_backend),
)

35
tests/v1/structured_output/test_reasoning_structured_output.py
View File

@@ -7,7 +7,7 @@ from unittest.mock import Mock
import pytest
from vllm.config import ModelConfig, RendererConfig, SchedulerConfig, VllmConfig
from vllm.config import ModelConfig, SchedulerConfig, VllmConfig
from vllm.reasoning import ReasoningParser
from vllm.v1.request import Request
from vllm.v1.structured_output import StructuredOutputManager
@@ -17,26 +17,19 @@ class TestReasoningStructuredOutput:
"""Test reasoning-aware structured output functionality."""
@pytest.fixture
def mock_renderer_config(self):
"""Create a mock RendererConfig."""
renderer_config = Mock(spec=RendererConfig)
renderer_config.skip_tokenizer_init = (
True # Skip tokenizer init to avoid network calls
)
model_config = Mock(spec=ModelConfig)
model_config.get_vocab_size = Mock(return_value=50000)
model_config.trust_remote_code = False
def mock_model_config(self):
"""Create a mock ModelConfig."""
config = Mock(spec=ModelConfig)
config.skip_tokenizer_init = True # Skip tokenizer init to avoid network calls
config.get_vocab_size = Mock(return_value=50000)
# Add missing runner_type attribute that tokenizer initialization expects
model_config.runner_type = "generate"
renderer_config.model_config = model_config
config.runner_type = "generate"
# Add other attributes that tokenizer initialization might need
renderer_config.tokenizer = "test-tokenizer"
renderer_config.tokenizer_mode = "auto"
renderer_config.tokenizer_revision = None
return renderer_config
config.tokenizer = "test-tokenizer"
config.tokenizer_mode = "auto"
config.trust_remote_code = False
config.tokenizer_revision = None
return config
@pytest.fixture
def mock_scheduler_config(self):
@@ -46,10 +39,10 @@ class TestReasoningStructuredOutput:
return config
@pytest.fixture
def mock_vllm_config(self, mock_renderer_config, mock_scheduler_config):
def mock_vllm_config(self, mock_model_config, mock_scheduler_config):
"""Create a mock VllmConfig."""
config = Mock(spec=VllmConfig)
config.renderer_config = mock_renderer_config
config.model_config = mock_model_config
config.scheduler_config = mock_scheduler_config
config.structured_outputs_config = Mock()
config.structured_outputs_config.reasoning_parser = None

2
tests/v1/tpu/worker/test_tpu_model_runner.py
View File

@@ -7,7 +7,6 @@ from vllm.attention.layer import Attention
from vllm.config import (
CacheConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
set_current_vllm_config,
@@ -46,7 +45,6 @@ def get_vllm_config():
)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
scheduler_config=scheduler_config,
)

3
tests/v1/worker/test_gpu_model_runner.py
View File

@@ -13,7 +13,6 @@ from vllm.config import (
CacheConfig,
ModelConfig,
ParallelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
set_current_vllm_config,
@@ -102,7 +101,6 @@ def get_vllm_config():
parallel_config = ParallelConfig()
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
scheduler_config=scheduler_config,
parallel_config=parallel_config,
@@ -813,7 +811,6 @@ def test_hybrid_attention_mamba_tensor_shapes():
attention_config = AttentionConfig(backend=AttentionBackendEnum.FLASHINFER)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
scheduler_config=scheduler_config,
parallel_config=parallel_config,