Struct CollectiveConfig 

pub struct CollectiveConfig {
    pub num_devices: usize,
    pub local_all_reduce: AllReduceStrategy,
    pub local_reduce: ReduceStrategy,
    pub local_broadcast: BroadcastStrategy,
    pub num_nodes: Option<usize>,
    pub global_all_reduce: Option<AllReduceStrategy>,
    pub global_reduce: Option<ReduceStrategy>,
    pub global_broadcast: Option<BroadcastStrategy>,
    pub tree_threshold_bytes: usize,
    pub tree_arity: usize,
    pub timeout: Duration,
}

Configuration for collective operations.

Fields

num_devices: usize

Number of local devices (GPUs).

local_all_reduce: AllReduceStrategy

Local all-reduce strategy.

local_reduce: ReduceStrategy

Local reduce strategy.

local_broadcast: BroadcastStrategy

Local broadcast strategy.

num_nodes: Option<usize>

Number of nodes (None = single node).

global_all_reduce: Option<AllReduceStrategy>

Global all-reduce strategy.

global_reduce: Option<ReduceStrategy>

Global reduce strategy.

global_broadcast: Option<BroadcastStrategy>

Global broadcast strategy.

tree_threshold_bytes: usize

Threshold (bytes) below which tree is preferred over ring.

tree_arity: usize

Tree arity (branching factor).

timeout: Duration

Timeout for collective operations.

Implementations

impl CollectiveConfig

pub fn single_node(num_devices: usize) -> Self

Create a config for a single node with multiple devices.

pub fn multi_node(num_devices: usize, num_nodes: usize) -> Self

Create a config for multi-node training.

pub fn validate(&self) -> DistributedResult<()>

Validate the configuration.

pub fn select_all_reduce( &self, buffer_size: usize, world_size: usize, ) -> AllReduceStrategy

Select the best all-reduce strategy for a given buffer size.

Trait Implementations

impl Clone for CollectiveConfig

fn clone(&self) -> CollectiveConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CollectiveConfig

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

Formats the value using the given formatter.

impl Default for CollectiveConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for CollectiveConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for CollectiveConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.