bridge.models.nemotron_omni.processing_super_vl#

Bridge-local loading adapter for the native VL processor in Transformers 5.18.0 through 5.20.0.

Imported lazily only by the version-guarded Super-VL pretrained wrapper. Image, video and text processing use the upstream implementation without monkey patches or changes to the source checkpoint. Loading these exports on unpatched HF 5.18.0 through 5.20.0 still requires this Bridge adapter because HF insists on an audio extractor.

Module Contents#

Classes#

Nemotron35VLProcessor

Reuse the native processor without requiring an unused audio component.

API#

class bridge.models.nemotron_omni.processing_super_vl.Nemotron35VLProcessor#

Bases: transformers.models.nemotron_h_omni.processing_nemotron_h_omni.NemotronH_Omni_Reasoning_V3Processor

Reuse the native processor without requiring an unused audio component.

classmethod get_attributes() → list[str]#
classmethod _get_arguments_from_pretrained(
pretrained_model_name_or_path,
processor_dict=None,
**kwargs,
)#
_process_audio(audio, audio_kwargs)#
to_dict() → dict#

Save native metadata, without introducing a remote-code dependency.