[Bugfix] Standardize custom HF Processor init (#37289)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-03-17 15:38:55 +00:00
committed by GitHub
parent 979ff44cea
commit c781fbbab3
7 changed files with 41 additions and 35 deletions
+3 -1
View File
@@ -196,8 +196,10 @@ class DeepseekOCRProcessingInfo(BaseProcessingInfo):
crop_mode=CROP_MODE,
strategy="v1",
)
return self.ctx.get_hf_processor(
DeepseekOCRProcessor, **{**kwargs, **v1_processor_config}
DeepseekOCRProcessor,
**{**v1_processor_config, **kwargs},
)
def get_supported_mm_limits(self) -> Mapping[str, int | None]:
+3 -1
View File
@@ -76,8 +76,10 @@ class DeepseekOCR2ProcessingInfo(BaseProcessingInfo):
crop_mode=CROP_MODE,
strategy="v2",
)
return self.ctx.get_hf_processor(
DeepseekOCRProcessor, **{**kwargs, **v2_processor_config}
DeepseekOCRProcessor,
**{**v2_processor_config, **kwargs},
)
def get_supported_mm_limits(self) -> Mapping[str, int | None]:
+12 -4
View File
@@ -47,7 +47,10 @@ from vllm.multimodal.processing import (
)
from vllm.sequence import IntermediateTensors
from vllm.transformers_utils.configs.chatglm import ChatGLMConfig
from vllm.transformers_utils.processors.glm4v import GLM4VProcessor
from vllm.transformers_utils.processors.glm4v import (
GLM4VImageProcessorFast,
GLM4VProcessor,
)
from vllm.utils.tensor_schema import TensorSchema, TensorShape
from .chatglm import ChatGLMBaseModel, ChatGLMModel, GLMTransformer
@@ -387,15 +390,20 @@ class GLM4VProcessingInfo(BaseProcessingInfo):
def get_hf_config(self):
return self.ctx.get_hf_config(ChatGLMConfig)
def get_hf_processor(self, **kwargs: object) -> GLM4VProcessor:
def get_image_processor(self, **kwargs):
config = self.get_hf_config()
vision_config = config.vision_config
image_size = vision_config["image_size"]
image_size = vision_config["image_size"]
kwargs.setdefault("size", {"width": image_size, "height": image_size})
return GLM4VImageProcessorFast(**kwargs)
def get_hf_processor(self, **kwargs: object) -> GLM4VProcessor:
return self.ctx.init_processor(
GLM4VProcessor,
tokenizer=self.get_tokenizer(),
**{**kwargs, "image_size": image_size},
image_processor=self.get_image_processor(**kwargs),
)
def get_supported_mm_limits(self) -> Mapping[str, int | None]:
+12 -4
View File
@@ -44,7 +44,10 @@ from vllm.multimodal.processing import (
PromptUpdateDetails,
)
from vllm.sequence import IntermediateTensors
from vllm.transformers_utils.processors.qwen_vl import QwenVLProcessor
from vllm.transformers_utils.processors.qwen_vl import (
QwenVLImageProcessorFast,
QwenVLProcessor,
)
from vllm.utils.tensor_schema import TensorSchema, TensorShape
from .interfaces import (
@@ -432,15 +435,20 @@ class QwenVLModel(QWenModel):
class QwenVLProcessingInfo(BaseProcessingInfo):
def get_hf_processor(self, **kwargs: object) -> QwenVLProcessor:
def get_image_processor(self, **kwargs):
config = self.get_hf_config()
vision_config = config.visual
image_size = vision_config["image_size"]
image_size = vision_config["image_size"]
kwargs.setdefault("size", {"width": image_size, "height": image_size})
return QwenVLImageProcessorFast(**kwargs)
def get_hf_processor(self, **kwargs: object) -> QwenVLProcessor:
return self.ctx.init_processor(
QwenVLProcessor,
tokenizer=self.get_tokenizer(),
**{**kwargs, "image_size": image_size},
image_processor=self.get_image_processor(**kwargs),
)
def get_supported_mm_limits(self) -> Mapping[str, int | None]: