Struct TrainingRecord 

pub struct TrainingRecord<B, M, O, S, U>
where
    B: AutodiffBackend,
    M: AutodiffModule<B>,
    O: Optimizer<M, B>,
    S: LrScheduler,
    U: Record<B>,
{ /* private fields */ }

Available on crate feature std only.

One record containing the trainable state and a caller-defined continuation record.

Capture at a training boundary with no concurrent updates. The caller’s state carries its counters and any available input/RNG records; this type does not infer them or snapshot a DataLoader. Recorder precision settings apply to all component records and must preserve their values for exact continuation.

Implementations

impl<B, M, O, S, U> TrainingRecord<B, M, O, S, U>

where B: AutodiffBackend, M: AutodiffModule<B>, O: Optimizer<M, B>, S: LrScheduler, U: Record<B>,

pub fn capture( model: &M, optimizer: &O, scheduler: &S, accumulator: &GradientsAccumulator<M>, state: U, ) -> Result<Self, RecorderError>

Capture the components without consuming the model or clearing gradients.

pub async fn capture_async( model: &M, optimizer: &O, scheduler: &S, accumulator: &GradientsAccumulator<M>, state: U, ) -> Result<Self, RecorderError>

Capture with asynchronous gradient readback; no recorder I/O is performed.

pub fn save<R: Recorder<B>>( self, recorder: &R, args: R::RecordArgs, ) -> Result<R::RecordOutput, RecorderError>

Save all components in one recorder payload.

pub fn load<R: Recorder<B>>( recorder: &R, args: R::LoadArgs, device: &B::Device, ) -> Result<Self, RecorderError>

Read a combined record using the selected recorder and device.

pub fn restore( self, model: M, optimizer: O, scheduler: S, device: &B::Device, ) -> Result<RestoredTraining<M, O, S, U>, RecorderError>

Restore onto compatible model, optimizer and scheduler configurations.

No scheduler step, optimizer step or gradient reset is performed. Apply the returned caller state before consuming the next training batch.

Trait Implementations

impl<B, M, O, S, U> Record<B> for TrainingRecord<B, M, O, S, U>

where B: AutodiffBackend, M: AutodiffModule<B>, O: Optimizer<M, B>, S: LrScheduler, U: Record<B>,

type Item<P: PrecisionSettings> = (<<M as Module<B>>::Record as Record<B>>::Item<P>, <<O as Optimizer<M, B>>::Record as Record<B>>::Item<P>, <<S as LrScheduler>::Record<B> as Record<B>>::Item<P>, <GradientsParamsRecord as Record<B>>::Item<P>, <U as Record<B>>::Item<P>)

Type of the item that can be serialized and deserialized.

fn into_item<P: PrecisionSettings>(self) -> Self::Item<P>

Convert the current record into the corresponding item that follows the given settings.

fn from_item<P: PrecisionSettings>( item: Self::Item<P>, device: &B::Device, ) -> Self

Convert the given item into a record.