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

28
vllm/transformers_utils/processor.py
View File

@@ -23,7 +23,7 @@ from vllm.transformers_utils.utils import convert_model_repo_to_path
from vllm.utils.func_utils import get_allowed_kwarg_only_overrides
if TYPE_CHECKING:
from vllm.config import ModelConfig, RendererConfig
from vllm.config import ModelConfig
_P = TypeVar("_P", bound=ProcessorMixin, default=ProcessorMixin)
_V = TypeVar("_V", bound=BaseVideoProcessor, default=BaseVideoProcessor)
@@ -233,18 +233,17 @@ def cached_get_processor_without_dynamic_kwargs(
def cached_processor_from_config(
renderer_config: "RendererConfig",
model_config: "ModelConfig",
processor_cls: type[_P] | tuple[type[_P], ...] = ProcessorMixin,
**kwargs: Any,
) -> _P:
model_config = renderer_config.model_config
if is_gguf(model_config.model):
assert not is_gguf(renderer_config.tokenizer), (
assert not is_gguf(model_config.tokenizer), (
"For multimodal GGUF models, the original tokenizer "
"should be used to correctly load processor."
)
model = renderer_config.tokenizer
revision = renderer_config.tokenizer_revision
model = model_config.tokenizer
revision = model_config.tokenizer_revision
else:
model = model_config.model
revision = model_config.revision
@@ -298,11 +297,9 @@ cached_get_feature_extractor = lru_cache(get_feature_extractor)
def cached_feature_extractor_from_config(
renderer_config: "RendererConfig",
model_config: "ModelConfig",
**kwargs: Any,
):
model_config = renderer_config.model_config
return cached_get_feature_extractor(
model_config.model,
revision=model_config.revision,
@@ -351,17 +348,16 @@ cached_get_image_processor = lru_cache(get_image_processor)
def cached_image_processor_from_config(
renderer_config: "RendererConfig",
model_config: "ModelConfig",
**kwargs: Any,
):
model_config = renderer_config.model_config
if is_gguf(model_config.model):
assert not is_gguf(renderer_config.tokenizer), (
assert not is_gguf(model_config.tokenizer), (
"For multimodal GGUF models, the original tokenizer "
"should be used to correctly load image processor."
)
model = renderer_config.tokenizer
revision = renderer_config.tokenizer_revision
model = model_config.tokenizer
revision = model_config.tokenizer_revision
else:
model = model_config.model
revision = model_config.revision
@@ -415,12 +411,10 @@ cached_get_video_processor = lru_cache(get_video_processor)
def cached_video_processor_from_config(
renderer_config: "RendererConfig",
model_config: "ModelConfig",
processor_cls: type[_V] | None = None,
**kwargs: Any,
):
model_config = renderer_config.model_config
return cached_get_video_processor(
model_config.model,
revision=model_config.revision,