Struct TrainingCheckpoint 

pub struct TrainingCheckpoint<B: Backend> {
    pub model_params: Vec<(String, Tensor<B>)>,
    pub optimizer_state: Option<OptimizerState>,
    pub epoch: u64,
    pub global_step: u64,
    pub best_loss: f64,
    pub loss_history: Vec<f64>,
}

Complete training checkpoint: model weights + optimizer state + training metadata.

Enables full training resume — not just the model parameters, but all the internal state that would be lost if training were interrupted.

Fields

model_params: Vec<(String, Tensor<B>)>

Named model parameters (same as standard checkpoint).

optimizer_state: Option<OptimizerState>

Optimizer internal state (momentum buffers, step counters, etc.)

epoch: u64

Current training epoch (0-indexed).

global_step: u64

Global optimization step counter.

best_loss: f64

Best loss value seen during training.

loss_history: Vec<f64>

Per-epoch loss history.

Implementations

impl<B: Backend> TrainingCheckpoint<B>

pub fn new() -> Self

Create an empty checkpoint (no model params).

pub fn from_executor(executor: &Executor<B>) -> Self

Create a checkpoint from an executor (model params only).

pub fn with_optimizer_state(self, state: OptimizerState) -> Self

Set the optimizer state in this checkpoint.

pub fn with_epoch(self, epoch: u64) -> Self

Set epoch counter.

pub fn with_global_step(self, step: u64) -> Self

Set global step counter.

pub fn with_best_loss(self, loss: f64) -> Self

Set best loss.

pub fn with_loss_history(self, history: Vec<f64>) -> Self

Set loss history.

Trait Implementations

impl<B: Clone + Backend> Clone for TrainingCheckpoint<B>

fn clone(&self) -> TrainingCheckpoint<B>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<B: Debug + Backend> Debug for TrainingCheckpoint<B>

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

Formats the value using the given formatter.

impl<B: Backend> Default for TrainingCheckpoint<B>

fn default() -> Self

Returns the “default value” for a type.