skainet-lang-core/sk.ainet.java/TrainingLoop

TrainingLoop

class TrainingLoop(source)

Java-friendly training loop wrapping SKaiNET's trainStep function.

Example usage from Java:

TrainingLoop loop = TrainingLoop.builder()
    .model(model)
    .loss(Losses.crossEntropy())
    .optimizer(Optimizers.adam(0.001))
    .context(ctx)
    .build();

// Single step
float stepLoss = loop.step(inputBatch, targetBatch);

// Full training
TrainingResult result = loop.train(dataIterator, 10);

Types

Builder for TrainingLoop.

object Companion

Functions

fun model(): Module<DType, Any?>

Returns the model being trained.

fun step(x: Tensor<*, *>, y: Tensor<*, *>): Float

Perform a single training step.

fun train(epochDataProvider: Supplier<Iterator<Pair<Tensor<*, *>, Tensor<*, *>>>>, epochs: Int): TrainingResult

Train the model for the specified number of epochs using an iterable of (input, target) pairs per epoch.

fun trainAsync(epochDataProvider: Supplier<Iterator<Pair<Tensor<*, *>, Tensor<*, *>>>>, epochs: Int): CompletableFuture<TrainingResult>

Train asynchronously using virtual threads.