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

[Refactor] Define MM data parser in processing info instead of processor itself (#33260)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-01-29 13:55:17 +08:00
committed by GitHub
parent 07ea184f00
commit 51550179fc
34 changed files with 399 additions and 347 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/model_executor/models/qwen2_vl.py
View File

@@ -806,6 +806,12 @@ class Qwen2VLProcessingInfo(BaseProcessingInfo):
def get_image_processor(self, **kwargs: object) -> Qwen2VLImageProcessor:
return self.get_hf_processor(**kwargs).image_processor
def get_data_parser(self):
return Qwen2VLMultiModalDataParser(
self.get_hf_config().vision_config.spatial_merge_size,
expected_hidden_size=self._get_expected_hidden_size(),
)
def get_supported_mm_limits(self) -> Mapping[str, int | None]:
return {"image": None, "video": None}
@@ -1039,11 +1045,6 @@ class Qwen2VLDummyInputsBuilder(BaseDummyInputsBuilder[Qwen2VLProcessingInfo]):
class Qwen2VLMultiModalProcessor(BaseMultiModalProcessor[Qwen2VLProcessingInfo]):
def _get_data_parser(self) -> MultiModalDataParser:
return Qwen2VLMultiModalDataParser(
self.info.get_hf_config().vision_config.spatial_merge_size
)
def _get_prompt_updates(
self,
mm_items: MultiModalDataItems,