Struct MultiNodeConfig 

pub struct MultiNodeConfig {
    pub num_nodes: usize,
    pub devices_per_node: usize,
    pub node_rank: usize,
    pub local_rank: usize,
    pub global_rank: usize,
    pub zero_config: ZeROConfig,
    pub gradient_compression: bool,
    pub comm_backend: CommunicationBackend,
    pub overlap_comm_compute: bool,
    pub gradient_bucket_size_mb: usize,
}

Multi-node training configuration

Fields

num_nodes: usize

Number of nodes in the cluster

devices_per_node: usize

Number of devices per node

node_rank: usize

Node rank (0-based)

local_rank: usize

Local device rank within node

global_rank: usize

Global rank across all nodes

zero_config: ZeROConfig

ZeRO configuration for memory optimization

gradient_compression: bool

Enable gradient compression

comm_backend: CommunicationBackend

Communication backend

overlap_comm_compute: bool

Enable overlap of computation and communication

gradient_bucket_size_mb: usize

Bucket size for gradient synchronization (MB)

Implementations

impl MultiNodeConfig

pub fn new( num_nodes: usize, devices_per_node: usize, node_rank: usize, local_rank: usize, ) -> Self

Create configuration for multi-node training

pub fn world_size(&self) -> usize

Get total world size across all nodes

pub fn is_master(&self) -> bool

Check if this is the master rank

pub fn node_local_ranks(&self) -> Vec<usize>

Get node-local ranks for this node

Trait Implementations

impl Clone for MultiNodeConfig

fn clone(&self) -> MultiNodeConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for MultiNodeConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for MultiNodeConfig

fn default() -> Self

Returns the “default value” for a type.