bridge.models.nemotron_omni.modeling_nemotron_omni#
Module Contents#
Classes#
Extends NemotronVLModel with sound modality freeze support. |
API#
- class bridge.models.nemotron_omni.modeling_nemotron_omni.NemotronOmniModel#
Bases:
megatron.bridge.models.nemotron_vl.modeling_nemotron_vl.NemotronVLModelExtends NemotronVLModel with sound modality freeze support.
forward() is inherited from NemotronVLModel (which delegates to LLaVAModel), so sound kwargs (sound_clips, sound_length) pass through automatically when the vlm2 LLaVAModel is used.
- freeze(
- *,
- freeze_language_model: bool = False,
- freeze_vision_model: bool = False,
- freeze_vision_projection: bool = False,
- freeze_sound_model: bool = False,
- freeze_sound_projection: bool = False,