HorovodConfig

class ray.train.horovod.HorovodConfig(nics: Set[str] | None = None, verbose: int = 1, key: str | None = None, ssh_port: int | None = None, ssh_identity_file: str | None = None, ssh_str: str | None = None, timeout_s: int = 300, placement_group_timeout_s: int = 100)

Bases: BackendConfig

Configurations for Horovod setup.

See horovod/horovod # noqa: E501

Parameters:

• nics (Optional[Set[str]) – Network interfaces that can be used for communication.

• verbose (int) – Horovod logging verbosity.

• key (Optional[str]) – Secret used for communication between workers.

• ssh_port (Optional[int]) – Port for SSH server running on worker nodes.

• ssh_identity_file (Optional[str]) – Path to the identity file to ssh into different hosts on the cluster.

• ssh_str (Optional[str]) – CAUTION WHEN USING THIS. Private key file contents. Writes the private key to ssh_identity_file.

• timeout_s (int) – Timeout parameter for Gloo rendezvous.

• placement_group_timeout_s (int) – Timeout parameter for Ray Placement Group creation. Currently unused.

PublicAPI (beta): This API is in beta and may change before becoming stable.

Methods

to_dict

Returns serializable dictionary representation of the backend config.

Attributes

backend_cls	The Backend class that sets up this framework on the workers.
framework	The training framework this config sets up.
key
nics
placement_group_timeout_s
ssh_identity_file
ssh_port
ssh_str
start_timeout	A Horovod Timeout of timeout_s seconds.
timeout_s
train_func_context	The context manager class wrapping each worker's training function.
verbose