TensorOpera Federate APIs

fedml.init()

Initialize the fedml library.

Usage

fedml.init(args=None, check_env=True, should_init_logs=True) -> None

Arguments

• args (str=None): Arguments.
• check_env (bool=True): Do we need to check the environment? such as gpu info, network connection, etc.
• should_init_logs (bool=True): Do we need to init logs related environment.

Returns
None

Examples

import fedml
from fedml.cross_silo import Client

if __name__ == "__main__":
    args = fedml.init()

fedml.device.get_device()

Apply the compute device (GPUs) for federated learning.

Usage

fedml.device.get_device(args) -> torch.device

Arguments

• args (str=None): Arguments.

Returns
torch.device.

Examples

import fedml

if __name__ == "__main__":
    # init FedML framework
    args = fedml.init()

    # init device
    device = fedml.device.get_device(args)

fedml.data.load()

Load datasets for federated learning.

Usage

fedml.data.load(args) -> (dataset, output_dim)

Arguments

• args (str=None): Arguments.

Returns
(dataset, output_dim)

Examples

import fedml

if __name__ == "__main__":
    # init FedML framework
    args = fedml.init()

    # init device
    device = fedml.device.get_device(args)

    # load data
    dataset, output_dim = fedml.data.load(args)

fedml.model.create()

Load the model for federated learning.

Usage

fedml.model.create(args, output_dim) -> model

Arguments

• args (str=None): Arguments.
• output_dim (int): Output dimension for model.

Returns
(dataset, output_dim)

Examples

import fedml

if __name__ == "__main__":
    # init FedML framework
    args = fedml.init()

    # init device
    device = fedml.device.get_device(args)

    # load data
    dataset, output_dim = fedml.data.load(args)

    # load model
    model = fedml.model.create(args, output_dim)

fedml.FedMLRunner.run()

Run the federated learning instance (client or server).

Usage

fedml.FedMLRunner.run() -> None

Arguments None

Returns
None

Examples

Example for the federated learning client:

import fedml
from fedml.cross_silo import Client

if __name__ == "__main__":
    args = fedml.init()

    # init device
    device = fedml.device.get_device(args)

    # load data
    dataset, output_dim = fedml.data.load(args)

    # load model
    model = fedml.model.create(args, output_dim)

    # start training
    client = Client(args, device, dataset, model)
    client.run()

Example for the federated learning server:

import fedml
from fedml.cross_silo import Server

if __name__ == "__main__":
    args = fedml.init()

    # init device
    device = fedml.device.get_device(args)

    # load data
    dataset, output_dim = fedml.data.load(args)

    # load model
    model = fedml.model.create(args, output_dim)

    # start training
    server = Server(args, device, dataset, model)
    server.run()