Scalar Trainer Operations

Packed loss, gradient, and update operations, plus simple training and evaluation loops. Differentiable inputs use scalar type α; non-differentiable data, such as token identifiers, labels, or masks, use a separate type δ.

def Runtime.Autograd.Torch.ScalarTrainer.runLoss {α δ : Type} [TorchLean.Storage α] [TorchLean.Storage δ] {paramShapes inputShapes dataInputShapes : List Spec.Shape} (trainer : ScalarTrainer α δ paramShapes inputShapes dataInputShapes) (inputs : TorchLean.TensorPack α inputShapes) (dataInputs : TorchLean.TensorPack δ dataInputShapes) : IO (TorchLean.Tensor α Spec.Shape.scalar)

Evaluate the scalar loss on packed differentiable and non-differentiable inputs.

def Runtime.Autograd.Torch.ScalarTrainer.runDiff {α δ : Type} [TorchLean.Storage α] [TorchLean.Storage δ] {paramShapes inputShapes dataInputShapes : List Spec.Shape} (trainer : ScalarTrainer α δ paramShapes inputShapes dataInputShapes) (inputs : TorchLean.TensorPack α inputShapes) (dataInputs : TorchLean.TensorPack δ dataInputShapes) : IO (TorchLean.Tensor α Spec.Shape.scalar × TorchLean.TensorPack α paramShapes)

Evaluate one loss and its parameter gradients from the same tape.

def Runtime.Autograd.Torch.ScalarTrainer.runGrad {α δ : Type} [TorchLean.Storage α] [TorchLean.Storage δ] {paramShapes inputShapes dataInputShapes : List Spec.Shape} (trainer : ScalarTrainer α δ paramShapes inputShapes dataInputShapes) (inputs : TorchLean.TensorPack α inputShapes) (dataInputs : TorchLean.TensorPack δ dataInputShapes) : IO (TorchLean.TensorPack α paramShapes)

Evaluate parameter gradients on packed inputs.

def Runtime.Autograd.Torch.ScalarTrainer.runStep {α δ : Type} [TorchLean.Storage α] [TorchLean.Storage δ] {paramShapes inputShapes dataInputShapes : List Spec.Shape} (trainer : ScalarTrainer α δ paramShapes inputShapes dataInputShapes) (learningRate : α) (inputs : TorchLean.TensorPack α inputShapes) (dataInputs : TorchLean.TensorPack δ dataInputShapes) : IO Unit

Apply the trainer's SGD update to packed inputs.

def Runtime.Autograd.Torch.ScalarTrainer.runStepWithLoss {α δ : Type} [TorchLean.Storage α] [TorchLean.Storage δ] {paramShapes inputShapes dataInputShapes : List Spec.Shape} (trainer : ScalarTrainer α δ paramShapes inputShapes dataInputShapes) (learningRate : α) (inputs : TorchLean.TensorPack α inputShapes) (dataInputs : TorchLean.TensorPack δ dataInputShapes) : IO (TorchLean.Tensor α Spec.Shape.scalar)

Apply the trainer's SGD update and return the loss used for the update.

def Runtime.Autograd.Torch.trainCycleSGD {α : Type} [TorchLean.Storage α] [ToString α] {paramShapes inputShapes : List Spec.Shape} (trainer : ScalarTrainer α Unit paramShapes inputShapes) (learningRate : α) (steps : ℕ) (samples : List (TorchLean.TensorPack α inputShapes)) (logEvery : ℕ := 1) : IO Unit

Apply steps SGD updates while cycling through samples without auxiliary data tensors.

Logging reuses each update's loss. Set logEvery := 0 to disable logging. Rejects an empty dataset.

def Runtime.Autograd.Torch.meanLoss {α : Type} [TorchLean.Storage α] [Add α] [Div α] [Zero α] [NatCast α] {paramShapes inputShapes : List Spec.Shape} (trainer : ScalarTrainer α Unit paramShapes inputShapes) (samples : List (TorchLean.TensorPack α inputShapes)) : IO α

Evaluate the arithmetic mean loss over samples without auxiliary data tensors.

Rejects an empty dataset.