[MM] Further cleanup _apply_hf_processor_main (#53610)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-09-16 10:21:00 -07:00
committed by GitHub
parent 6ca2b23e22
commit 0b9e018fb9
74 changed files with 1048 additions and 1323 deletions
+2 -8
View File
@@ -321,12 +321,6 @@ class InputProcessingContext:
merged_kwargs = self.get_merged_mm_kwargs(kwargs)
# vLLM needs the full untruncated sequence to keep multi-modal
# placeholder tokens aligned; note that the text inputs in
# call_hf_processor are just dummy text, not the original prompt.
# The original prompt is already tokenized by the renderer.
merged_kwargs.setdefault("truncation", False)
allowed_kwargs = get_allowed_kwarg_only_overrides(
hf_processor,
merged_kwargs,
@@ -426,7 +420,7 @@ class BaseProcessingInfo:
def get_data_parser(self) -> MultiModalDataParser:
"""Constructs a parser to preprocess multi-modal data items
before passing them to
[`_get_hf_mm_data`][vllm.multimodal.processing.BaseMultiModalProcessor._get_hf_mm_data].
[`_get_hf_mm_inputs`][vllm.multimodal.processing.BaseMultiModalProcessor._get_hf_mm_inputs].
You can support additional modalities by creating a subclass
of [`MultiModalDataParser`][vllm.multimodal.parse.MultiModalDataParser]
@@ -506,7 +500,7 @@ class BaseProcessingInfo:
"""Normalize [`MultiModalDataDict`][vllm.inputs.MultiModalDataDict]
to [`MultiModalDataItems`][vllm.multimodal.parse.MultiModalDataItems]
before passing them to
[`_get_hf_mm_data`][vllm.multimodal.processing.BaseMultiModalProcessor._get_hf_mm_data].
[`_get_hf_mm_inputs`][vllm.multimodal.processing.BaseMultiModalProcessor._get_hf_mm_inputs].
"""
mm_items = self.data_parser.parse_mm_data(mm_data)