[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
+21 -8
View File
@@ -612,20 +612,31 @@ class FunAudioChatDummyInputsBuilder(
class FunAudioChatMultiModalProcessor(
BaseMultiModalProcessor[FunAudioChatProcessingInfo]
):
def _call_hf_processor(
def _apply_hf_processor_main(
self,
prompt: str,
mm_data: Mapping[str, object],
mm_kwargs: Mapping[str, object],
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())
tokenizer = self.info.get_tokenizer()
input_ids = torch.tensor([tokenizer.encode(prompt)])
input_ids = torch.tensor([tokenizer.encode(prompt_text)])
audios = mm_data.get("audios", [])
if not audios:
return BatchFeature({"input_ids": input_ids})
processed_data = BatchFeature({"input_ids": input_ids})
processed_data.update(passthrough_data)
return processed_data
feature_extractor = self.info.get_feature_extractor(**mm_kwargs)
feature_extractor = self.info.get_feature_extractor(**hf_processor_mm_kwargs)
sr = int(feature_extractor.sampling_rate)
min_samples = int(getattr(feature_extractor, "n_fft", 400) or 400)
@@ -676,7 +687,9 @@ class FunAudioChatMultiModalProcessor(
"feature_exist_mask": torch.ones((len(wavs),), dtype=torch.bool),
}
return BatchFeature({"input_ids": input_ids, **mm_inputs})
processed_data = BatchFeature({"input_ids": input_ids, **mm_inputs})
processed_data.update(passthrough_data)
return processed_data
def _get_mm_fields_config(
self,