[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
@@ -261,44 +261,8 @@ class HCXVisionV2MultiModalProcessor(
):
"""Multimodal processor for HyperCLOVAX V2 (32B Think model)."""
def _apply_hf_processor_main(
self,
mm_items: MultiModalDataItems,
hf_processor_mm_kwargs: Mapping[str, object],
) -> BatchFeature:
valid_mm_items = mm_items.select(
{k for k, c in mm_items.get_all_counts().items() if c > 0}
)
mm_data, passthrough_data = self._get_hf_mm_data(valid_mm_items)
if not mm_data:
return BatchFeature(dict(passthrough_data))
prompt_text = self.dummy_inputs.get_dummy_text(mm_items.get_all_counts())
images = mm_data.get("images")
videos = mm_data.get("videos")
# Get the HF processor
hf_processor = self.info.get_hf_processor(**hf_processor_mm_kwargs)
# Build data dict for HF processor (images/videos only)
# NOTE: We pass the prompt as-is without token normalization.
# Token expansion is handled by vLLM via _get_prompt_updates.
data: dict[str, object] = dict(
text=prompt_text,
images=images,
videos=videos,
)
processed_data = self.info.ctx.call_hf_processor(
hf_processor=hf_processor,
data=data,
kwargs=hf_processor_mm_kwargs,
)
processed_data.update(passthrough_data)
return processed_data
def _get_hf_mm_text(self, mm_counts: Mapping[str, int]) -> str:
return self.dummy_inputs.get_dummy_text(mm_counts)
def _get_prompt_updates(
self,