Struct MultiGpuTrainer

pub struct MultiGpuTrainer { /* private fields */ }

Multi-GPU training coordinator

Implementations

impl MultiGpuTrainer

pub fn new(device_ids: Vec<u32>) -> Result<Self>

Create new multi-GPU trainer

pub fn set_reduction_strategy(&mut self, strategy: ReductionStrategy)

Set reduction strategy

pub fn all_reduce<T>(&self, tensors: &mut [CudaTensor<T>]) -> Result<()>

where T: Copy + Default + Send + Sync + Add<Output = T> + Div<Output = T> + From<usize>,

Perform all-reduce operation across GPUs

pub fn device_count(&self) -> usize

Get number of devices

pub fn memory_stats_all(&self) -> Result<Vec<MemoryStats>>

Get memory statistics for all devices

Trait Implementations

impl Debug for MultiGpuTrainer

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

Formats the value using the given formatter.