Struct TrainedState 

pub struct TrainedState {
    pub params: Vec<Tensor>,
    pub buffers: Vec<Tensor>,
}

Trained parameters and buffers returned by DdpHandle::join().

Contains the averaged final state from all workers. Parameters are on CPU. Buffers include running statistics (e.g. BatchNorm mean/var) needed for inference.

Example

let state = ddp.join()?;
// state.params[i] corresponds to model.parameters()[i]
// state.buffers[i] corresponds to model.buffers()[i]

Fields

params: Vec<Tensor>

Averaged parameter tensors (CPU). Same order as Module::parameters().

buffers: Vec<Tensor>

Averaged buffer tensors (CPU). Same order as Module::buffers().

Trait Implementations

impl Clone for TrainedState

fn clone(&self) -> TrainedState

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TrainedState

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

Formats the value using the given formatter.