[MM] Simplify _apply_hf_processor_main (#53275)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-08-22 01:49:37 -07:00
committed by GitHub
parent 7ca49fbe4b
commit 0b19ebcacd
90 changed files with 1848 additions and 1706 deletions
+12 -22
View File
@@ -332,53 +332,43 @@ class Ovis2_5MultiModalProcessor(BaseMultiModalProcessor[Ovis2_5ProcessingInfo])
if x >= INDICATOR_IDS[0]
]
def _call_hf_processor(
def _get_hf_processor_text(self, mm_counts: Mapping[str, int]) -> str:
return self.dummy_inputs.get_dummy_text(mm_counts)
def _postprocess_hf_mm_data(
self,
prompt: str,
mm_data: Mapping[str, object],
mm_kwargs: Mapping[str, object],
hf_processor_mm_kwargs: Mapping[str, object],
processed_data: BatchFeature,
) -> BatchFeature:
if not mm_data:
# Avoid warning from HF logger for text-only input
tokenizer = self.info.get_tokenizer()
prompt_ids = tokenizer.encode(prompt, add_special_tokens=False)
return BatchFeature(dict(input_ids=[prompt_ids]), tensor_type="pt")
return processed_data
processed_outputs = super()._call_hf_processor(
prompt=prompt,
mm_data=mm_data,
mm_kwargs=mm_kwargs,
)
hf_processor = self.info.get_hf_processor()
if "videos" in mm_data:
visual_indicators = [
hf_processor.construct_visual_indicators((1, 1, 1), True)
for grid in processed_outputs["video_grids"]
for grid in processed_data["video_grids"]
]
indicator_tokens = [
self.visual_indicators_to_visual_tokens(indicator)
for indicator in visual_indicators
]
processed_outputs["video_indicator_tokens"] = torch.tensor(indicator_tokens)
processed_data["video_indicator_tokens"] = torch.tensor(indicator_tokens)
if "images" in mm_data:
visual_indicators = [
hf_processor.construct_visual_indicators((1, 1, 1), False)
for grid in processed_outputs["grids"]
for grid in processed_data["grids"]
]
indicator_tokens = [
self.visual_indicators_to_visual_tokens(indicator)
for indicator in visual_indicators
]
processed_outputs["indicator_tokens"] = torch.tensor(indicator_tokens)
return processed_outputs
processed_data["indicator_tokens"] = torch.tensor(indicator_tokens)
def _apply_hf_processor_tokens_only(
self,
prompt_tokens: list[int],
) -> list[int]:
return prompt_tokens
return processed_data
def _get_mm_fields_config(
self,