[Misc] Automatically resolve HF processor init kwargs (#22005)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-07-31 22:44:10 -07:00
committed by GitHub
parent ad57f23f6a
commit 82de9b9d46
40 changed files with 334 additions and 727 deletions
+8 -14
View File
@@ -696,19 +696,12 @@ class Phi4MMProcessingInfo(BaseProcessingInfo):
def get_hf_config(self) -> Phi4MultimodalConfig:
return self.ctx.get_hf_config(Phi4MultimodalConfig)
def get_hf_processor(
self,
*,
dynamic_hd: Optional[int] = None,
**kwargs: object,
) -> Phi4MMProcessor:
if dynamic_hd is not None:
kwargs["dynamic_hd"] = dynamic_hd
def get_hf_processor(self, **kwargs: object) -> Phi4MMProcessor:
return self.ctx.get_hf_processor(Phi4MMProcessor, **kwargs)
return self.ctx.get_hf_processor(**kwargs)
def get_feature_extractor(self) -> Phi4MultimodalFeatureExtractor:
return self.get_hf_processor().audio_processor
def get_feature_extractor(
self, **kwargs: object) -> Phi4MultimodalFeatureExtractor:
return self.get_hf_processor(**kwargs).audio_processor
def get_image_processor(
self,
@@ -1007,7 +1000,7 @@ class Phi4MMMultiModalProcessor(BaseMultiModalProcessor[Phi4MMProcessingInfo]):
if audio_data:
audio_features = processed_outputs['audio_input_features']
sr = self.info.get_feature_extractor().sampling_rate
sr = self.info.get_feature_extractor(**mm_kwargs).sampling_rate
feature_sizes = [
self.info.get_audio_num_frames(len(audio), sr)
for audio in audio_data
@@ -1043,7 +1036,8 @@ class Phi4MMMultiModalProcessor(BaseMultiModalProcessor[Phi4MMProcessingInfo]):
audio_token_id = tokenizer.vocab[tokenizer.audio_token]
hf_processor = self.info.get_hf_processor(**hf_processor_mm_kwargs)
audio_processor = self.info.get_feature_extractor()
audio_processor = self.info.get_feature_extractor(
**hf_processor_mm_kwargs)
def get_image_replacement_phi4mm(item_idx: int):
images = mm_items.get_items(