Struct SimpleDistributedDataParallel 

pub struct SimpleDistributedDataParallel<T, M>
where
    T: DistributedScalar,
    M: Module<T> + Send + Sync + 'static,
{ /* private fields */ }

Simplified DistributedDataParallel for RusTorch RusTorch用簡略化DistributedDataParallel

Implementations

impl<T, M> SimpleDistributedDataParallel<T, M>

where T: DistributedScalar, M: Module<T> + Send + Sync + 'static,

pub fn new(module: M, device_ids: Option<Vec<usize>>) -> RusTorchResult<Self>

Create a new simplified DDP wrapper 新しい簡略化DDPラッパーを作成

pub fn forward(&self, input: &Variable<T>) -> RusTorchResult<Variable<T>>

Forward pass with distributed synchronization 分散同期付きフォワードパス

pub fn sync_gradients(&self) -> RusTorchResult<()>

Synchronize gradients across all processes 全プロセス間での勾配同期

pub fn module(&self) -> Arc<Mutex<M>>

Get the wrapped module ラップされたモジュールを取得

pub fn device_ids(&self) -> &[usize]

Get device IDs デバイスIDを取得

pub fn set_gradient_sync(&mut self, enabled: bool)

Enable or disable automatic gradient synchronization 自動勾配同期の有効/無効を設定

Trait Implementations

impl<T, M> DistributedDataParallelTrait<T> for SimpleDistributedDataParallel<T, M>

where T: DistributedScalar, M: Module<T> + Send + Sync + 'static,

fn device_ids(&self) -> &[usize]

Get device IDs for this DDP instance このDDPインスタンスのデバイスIDを取得

fn distributed_forward( &self, input: &Variable<T>, ) -> RusTorchResult<Variable<T>>

Perform distributed forward pass 分散フォワードパスを実行

fn sync_gradients(&self) -> RusTorchResult<()>

Synchronize gradients across processes プロセス間での勾配同期