protomotions.utils.fabric_config module#

Configuration classes for Lightning Fabric distributed training.

class protomotions.utils.fabric_config.FabricConfig( accelerator='gpu', devices=1, num_nodes=1, strategy=<factory>, precision='32-true', loggers=None, callbacks=None, )[source]#

Bases: object

Configuration for Lightning Fabric distributed training.

Attributes:: accelerator: Hardware accelerator: ‘gpu’, ‘cpu’, ‘tpu’, ‘auto’. devices: Number of devices or ‘auto’ for all available. num_nodes: Number of nodes for distributed training. strategy: Distributed training strategy (DDP, FSDP, etc). precision: Training precision: ‘32-true’, ‘16-mixed’, ‘bf16-mixed’. loggers: List of logging backends (WandB, TensorBoard, etc). callbacks: List of training callbacks.

accelerator: str = 'gpu'#

devices: int | str = 1#

num_nodes: int | str = 1#

strategy: Dict | <Mock object at 0x7fbf6b05bf50>[]#

precision: str | int = '32-true'#

loggers: Dict | <Mock object at 0x7fbf6b059010>[]] | None = None#

callbacks: List[Dict | Any] | None = None#

__init__( accelerator='gpu', devices=1, num_nodes=1, strategy=<factory>, precision='32-true', loggers=None, callbacks=None, )#