TorchLean API

Docs HomeInstallationGuideExamplesGraphsTools

NN.Runtime.Autograd.Torch.Core.CudaBridge

CUDA Tensor Storage Bridge #

Adapt the public TensorTransfer capability to dtype-carrying row-major storage owned by the eager CUDA tape.

source
def Runtime.Autograd.Torch.Internal.CudaBridge.toAnyBuffer {α : Type} [TorchLean.Storage α] [TensorTransfer α] {s : Spec.Shape} (tensor : TorchLean.Tensor α s) :
IO LibTorch.AnyBuffer

Upload complete configured words or the scalar's native dtype, preserving the tensor's shape.

Instances For
    source
    def Runtime.Autograd.Torch.Internal.CudaBridge.ofBuffer {α : Type} [TorchLean.Storage α] [TensorTransfer α] {s : Spec.Shape} (buffer : LibTorch.Buffer) :
    IO (TorchLean.Tensor α s)

    Download a buffer with the requested shape, checking its element count.

    Instances For
      source
      def Runtime.Autograd.Torch.Internal.CudaBridge.ofAnyBuffer {α : Type} [TorchLean.Storage α] [TensorTransfer α] (stored : LibTorch.AnyBuffer) :
      IO (Spec.SomeTensor α)

      Download a shape-erased buffer through the scalar type's runtime transfer representation.

      Instances For