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

[Misc][V1] Avoid using envs.VLLM_USE_V1 in mm processing (#14256)

Browse Source 
Signed-off-by: Roger Wang <ywang@roblox.com>
This commit is contained in:
Roger Wang
2025-03-04 23:37:16 -08:00
committed by GitHub
parent 32985bed7c
commit ec79b67c77
7 changed files with 38 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/llava.py
View File

@@ -767,6 +767,7 @@ class MantisMultiModalProcessor(LlavaMultiModalProcessor):
prompt: Union[str, list[int]],
mm_data: MultiModalDataDict,
hf_processor_mm_kwargs: Mapping[str, object],
return_mm_hashes: bool = False,
) -> MultiModalInputs:
hf_config = self.info.get_hf_config()
image_token_id = hf_config.image_token_index
@@ -777,7 +778,8 @@ class MantisMultiModalProcessor(LlavaMultiModalProcessor):
image_height=-1,
)
result = super().apply(prompt, mm_data, hf_processor_mm_kwargs)
result = super().apply(prompt, mm_data, hf_processor_mm_kwargs,
return_mm_hashes)
mm_items = self._to_mm_items(mm_data)
mm_item_counts = mm_items.get_all_counts()