`bridge.recipes.ministral3.ministral3`#

Module Contents#

Classes#

Ministral3FinetuneKwargs

Typed options accepted by Ministral3 finetuning recipe helper functions.

Functions#

`ministral3_3b_finetune_config`	Return a fine-tuning config for Ministral3 3B.
`ministral3_8b_finetune_config`	Return a fine-tuning config for Ministral3 8B.
`ministral3_14b_finetune_config`	Return a fine-tuning config for Ministral3 14B.
`_ministral3_finetune_common`	Create a fine-tuning configuration for Ministral3 family models using a given HuggingFace path.

API#

class bridge.recipes.ministral3.ministral3.Ministral3FinetuneKwargs#

Bases: typing_extensions.TypedDict

Typed options accepted by Ministral3 finetuning 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

seq_length: Optional[int]#: 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: Optional[int]#: None

micro_batch_size: int#: None

eval_interval: int#: None

save_interval: int#: None

finetune_lr: Optional[float]#: None

min_lr: float#: None

lr_warmup_iters: int#: None

lr_decay_iters: Optional[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

wandb_project: Optional[str]#: None

wandb_entity: Optional[str]#: None

wandb_exp_name: Optional[str]#: None

bridge.recipes.ministral3.ministral3.ministral3_3b_finetune_config(

**user_kwargs: typing_extensions.Unpack[bridge.recipes.ministral3.ministral3.Ministral3FinetuneKwargs],

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

Return a fine-tuning config for Ministral3 3B.

Default configuration: 1 node, 8 GPUs

LoRA/DoRA (default): TP=1, PP=1, LR=1e-4
Full SFT (peft=None): TP=1, PP=1, LR=5e-6

See _ministral3_finetune_common for the full list of parameters.

bridge.recipes.ministral3.ministral3.ministral3_8b_finetune_config(

**user_kwargs: typing_extensions.Unpack[bridge.recipes.ministral3.ministral3.Ministral3FinetuneKwargs],

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

Return a fine-tuning config for Ministral3 8B.

Default configuration: 1 node, 8 GPUs

LoRA/DoRA (default): TP=1, PP=1, LR=1e-4
Full SFT (peft=None): TP=2, PP=1, LR=5e-6

See _ministral3_finetune_common for the full list of parameters.

bridge.recipes.ministral3.ministral3.ministral3_14b_finetune_config(

**user_kwargs: typing_extensions.Unpack[bridge.recipes.ministral3.ministral3.Ministral3FinetuneKwargs],

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

Return a fine-tuning config for Ministral3 14B.

Default configuration: 1 node, 8 GPUs

LoRA/DoRA (default): TP=2, PP=1, LR=1e-4
Full SFT (peft=None): TP=4, PP=1, LR=5e-6

See _ministral3_finetune_common for the full list of parameters.

bridge.recipes.ministral3.ministral3._ministral3_finetune_common( hf_path: str, dir: Optional[str] = None, name: str = 'ministral3_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 = 1, 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 = 1000, global_batch_size: int = 32, micro_batch_size: int = 1, seq_length: int = 4096, eval_interval: int = 30, save_interval: int = 50, finetune_lr: Optional[float] = None, min_lr: float = 0.0, lr_warmup_iters: int = 50, lr_decay_iters: Optional[int] = None, 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, 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 Ministral3 family models using a given HuggingFace path.

The dataset pipeline is conversation-based. To train multimodal tokens, ensure your preprocessed data includes placeholders (e.g., ) as needed.

bridge.recipes.ministral3.ministral3#

Module Contents#

Classes#

Functions#

API#

`bridge.recipes.ministral3.ministral3`#