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

[Bugfix][Perf] Revert applying HF processor on text-only inputs for multimodal models (#28858)

Browse Source 
Signed-off-by: Roger Wang <hey@rogerw.io>
This commit is contained in:
Roger Wang
2025-11-17 06:49:25 -08:00
committed by GitHub
parent 64e39d667c
commit 7f064491f8
2 changed files with 11 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/inputs/preprocess.py
View File

@@ -348,18 +348,15 @@ class InputPreprocessor:
)
inputs: TokenInputs | MultiModalInputs
if self.model_config.is_multimodal_model:
if multi_modal_data := parsed_content.get("multi_modal_data"):
inputs = self._process_multimodal(
prompt_token_ids,
parsed_content.get("multi_modal_data") or {},
multi_modal_data,
parsed_content.get("mm_processor_kwargs") or {},
tokenization_kwargs=tokenization_kwargs,
mm_uuids=mm_uuids,
)
else:
if parsed_content.get("multi_modal_data"):
raise ValueError("This model does not support multimodal inputs")
inputs = token_inputs(prompt_token_ids)
if cache_salt := parsed_content.get("cache_salt"):
@@ -377,18 +374,15 @@ class InputPreprocessor:
prompt_text = parsed_content["prompt"]
inputs: TokenInputs | MultiModalInputs
if self.model_config.is_multimodal_model:
if multi_modal_data := parsed_content.get("multi_modal_data"):
inputs = self._process_multimodal(
prompt_text,
parsed_content.get("multi_modal_data") or {},
multi_modal_data,
parsed_content.get("mm_processor_kwargs") or {},
tokenization_kwargs=tokenization_kwargs,
mm_uuids=mm_uuids,
)
else:
if parsed_content.get("multi_modal_data"):
raise ValueError("This model does not support multimodal inputs")
prompt_token_ids = self._tokenize_prompt(
prompt_text,
tokenization_kwargs=tokenization_kwargs,