`bridge.recipes.gemma3_vl.gemma3_vl`#

Module Contents#

Classes#

Gemma3VLCommonKwargs

Typed options accepted by Gemma3-VL recipe helper functions.

Functions#

`gemma3_vl_4b_finetune_config`	Return a fine-tuning config for Gemma3-VL 4B Instruct.
`gemma3_vl_12b_finetune_config`	Return a fine-tuning config for Gemma3-VL 12B Instruct.
`gemma3_vl_27b_finetune_config`	Return a fine-tuning config for Gemma3-VL 27B Instruct.
`_gemma3_vl_common`	Create a fine-tuning configuration for Gemma3-VL models using a given HuggingFace path.

API#

class bridge.recipes.gemma3_vl.gemma3_vl.Gemma3VLCommonKwargs#

Bases: typing_extensions.TypedDict

Typed options accepted by Gemma3-VL recipe helper functions.

Initialization

Initialize self. See help(type(self)) for accurate signature.

hf_path: str#: None

dir: Optional[str]#: None

name: str#: None

train_data_path: Optional[List[str]]#: None

valid_data_path: Optional[List[str]]#: None

test_data_path: Optional[List[str]]#: None

dataset_type: Optional[str]#: None

image_folder: Optional[str]#: None

tokenizer_model: Optional[str]#: None

tensor_model_parallel_size: int#: None

pipeline_model_parallel_size: int#: None

pipeline_dtype: Optional[torch.dtype]#: None

virtual_pipeline_model_parallel_size: Optional[int]#: None

context_parallel_size: int#: None

sequence_parallel: bool#: None

use_megatron_fsdp: bool#: None

train_iters: int#: None

global_batch_size: int#: None

micro_batch_size: int#: None

seq_length: int#: None

lr: float#: None

min_lr: float#: None

lr_warmup_iters: int#: None

lr_decay_iters: Optional[int]#: None

eval_interval: int#: None

save_interval: int#: None

precision_config: Optional[Union[megatron.bridge.training.mixed_precision.MixedPrecisionConfig, str]]#: None

comm_overlap_config: Optional[megatron.bridge.training.comm_overlap.CommOverlapConfig]#: None

freeze_language_model: bool#: None

freeze_vision_model: bool#: None

freeze_vision_projection: bool#: None

pretrained_checkpoint: Optional[str]#: None

peft: Optional[Union[str, megatron.bridge.peft.base.PEFT]]#: None

finetune_lr: float#: None

wandb_project: Optional[str]#: None

wandb_entity: Optional[str]#: None

wandb_exp_name: Optional[str]#: None

bridge.recipes.gemma3_vl.gemma3_vl.gemma3_vl_4b_finetune_config(

**user_kwargs: typing_extensions.Unpack[bridge.recipes.gemma3_vl.gemma3_vl.Gemma3VLCommonKwargs],

) → megatron.bridge.training.config.ConfigContainer#

Return a fine-tuning config for Gemma3-VL 4B Instruct.

Default configuration: 1 node, 8 GPUs

LoRA/DoRA: TP=1, PP=1, LR=1e-4
Full SFT: TP=1, PP=1, LR=5e-6

See _gemma3_vl_common for the full list of parameters.

bridge.recipes.gemma3_vl.gemma3_vl.gemma3_vl_12b_finetune_config(

**user_kwargs: typing_extensions.Unpack[bridge.recipes.gemma3_vl.gemma3_vl.Gemma3VLCommonKwargs],

) → megatron.bridge.training.config.ConfigContainer#

Return a fine-tuning config for Gemma3-VL 12B Instruct.

Default configuration: 1 node, 8 GPUs

LoRA/DoRA: TP=1, PP=1, LR=1e-4
Full SFT: TP=4, PP=1, LR=5e-6

See _gemma3_vl_common for the full list of parameters.

bridge.recipes.gemma3_vl.gemma3_vl.gemma3_vl_27b_finetune_config(

**user_kwargs: typing_extensions.Unpack[bridge.recipes.gemma3_vl.gemma3_vl.Gemma3VLCommonKwargs],

) → megatron.bridge.training.config.ConfigContainer#

Return a fine-tuning config for Gemma3-VL 27B Instruct.

Default configuration: 2 nodes, 16 GPUs total

LoRA/DoRA: TP=4, PP=1, LR=1e-4
Full SFT: TP=8, PP=2, LR=5e-6

See _gemma3_vl_common for the full list of parameters.

bridge.recipes.gemma3_vl.gemma3_vl._gemma3_vl_common( hf_path: str, dir: Optional[str] = None, name: str = 'gemma3_vl_finetune', pretrained_checkpoint: Optional[str] = None, train_data_path: Optional[List[str]] = None, valid_data_path: Optional[List[str]] = None, test_data_path: Optional[List[str]] = None, dataset_type: Optional[str] = None, image_folder: Optional[str] = None, tokenizer_model: Optional[str] = None, tensor_model_parallel_size: int = 2, pipeline_model_parallel_size: int = 1, pipeline_dtype: Optional[torch.dtype] = None, virtual_pipeline_model_parallel_size: Optional[int] = None, context_parallel_size: int = 1, sequence_parallel: bool = False, use_megatron_fsdp: bool = False, train_iters: int = 300000, global_batch_size: int = 32, micro_batch_size: int = 2, seq_length: int = 4096, lr: float = 0.0003, min_lr: float = 3e-05, lr_warmup_iters: int = 500, lr_decay_iters: Optional[int] = None, eval_interval: int = 500, save_interval: int = 500, precision_config: Optional[Union[megatron.bridge.training.mixed_precision.MixedPrecisionConfig, str]] = 'bf16_mixed', comm_overlap_config: Optional[megatron.bridge.training.comm_overlap.CommOverlapConfig] = None, freeze_language_model: bool = False, freeze_vision_model: bool = False, freeze_vision_projection: bool = False, peft: Optional[Union[str, megatron.bridge.peft.base.PEFT]] = None, finetune_lr: Optional[float] = None, wandb_project: Optional[str] = None, wandb_entity: Optional[str] = None, wandb_exp_name: Optional[str] = None, ) → megatron.bridge.training.config.ConfigContainer#

Create a fine-tuning configuration for Gemma3-VL models using a given HuggingFace path.

The dataset pipeline is based on the Gemma3-VL architecture. To train multimodal tokens, ensure your preprocessed data includes appropriate image placeholders.

bridge.recipes.gemma3_vl.gemma3_vl#

Module Contents#

Classes#

Functions#

API#

`bridge.recipes.gemma3_vl.gemma3_vl`#