TorchLean-Executable GraphSpec Models #
This module is the architecture-facing home for reusable model constructors that still execute via the TorchLean autograd runtime.
The split is intentional:
NN.GraphSpec.Models.*contains graph-authored architectures and architecture-facing wrappers.NN.GraphSpec.Models.TorchLean.*contains executableTorchLean.NN.Seq/TorchLean.Programconstructors for common models.NN.Runtime.Autograd.TorchLean.*contains runtime machinery: tensors, ops, backends, sessions, losses, optimizers, and training loops.
So users looking for models start from GraphSpec, while runtime internals stay focused on execution.