`bridge.models.gemma_vl.gemma3_vl_provider`#

Module Contents#

Classes#

Gemma3VLModelProvider

Base model provider for Gemma VL Models.

API#

class bridge.models.gemma_vl.gemma3_vl_provider.Gemma3VLModelProvider#

Bases: megatron.bridge.models.gemma.gemma3_provider.Gemma3ModelProvider

Base model provider for Gemma VL Models.

scatter_embedding_sequence_parallel: bool#: False

vision_config: transformers.SiglipVisionConfig#: ‘field(…)’

vision_projector_config: megatron.bridge.models.gemma_vl.modeling_gemma3_vl.Gemma3VLMultimodalProjectorConfig#: ‘field(…)’

mm_tokens_per_image: int#: 256

bos_token_id: int#: 2

eos_token_id: int#: 1

vision_start_token_id: int#: 255999

vision_end_token_id: int#: 256000

image_token_id: int#: 262144

freeze_language_model: bool#: False

freeze_vision_model: bool#: False

freeze_vision_projection: bool#: False

provide( pre_process=None, post_process=None, vp_stage=None, ) → megatron.bridge.models.gemma_vl.modeling_gemma3_vl.Gemma3VLModel#

provide_language_model( pre_process=None, post_process=None, vp_stage=None, ) → megatron.core.models.gpt.GPTModel#