Eager Tensor Operations

PyTorch-style tensor operations backed by the eager CPU/CUDA tapes. These wrappers record runtime nodes, dispatch CUDA kernels when requested, and preserve the typed TensorRef surface.

Tensor ops (eager tape wrappers)

The following definitions are the eager front-end for Runtime.Autograd.Tape.* primitives. Each one:

• reads the current tape from s.tape,
• appends a new node/leaf via a Tape.* constructor,
• writes the updated tape back, and
• returns a fresh TensorRef pointing to the new node id.

PyTorch comparison: this is the standard eager autograd mechanism (a dynamic tape of ops).

def Runtime.Autograd.Torch.Internal.EagerSession.dispatchCudaCapsuleOpt {α β : Type} (s : EagerSession α) (op : NN.Backend.BackendOp) (cudaProviders : List NN.Backend.Provider) (cpu : IO β) (cuda : NN.Backend.KernelCapsule → IO (Option β)) : IO β

Dispatch an eager operation through its selected CPU or CUDA capsule.

cudaProviders names the providers implemented by the supplied CUDA handler. The selected capsule is bound to the matching handler before any implementation runs. Returning none still means that the operation has no implementation in this CUDA runtime; there is no per-operation CPU fallback.

def Runtime.Autograd.Torch.Internal.EagerSession.dispatchCudaOpt {α β : Type} (s : EagerSession α) (op : NN.Backend.BackendOp) (cpu : IO β) (cuda : IO (Option β)) : IO β

Dispatch an eager operation implemented by the reference CPU and TorchLean native CUDA runtimes.