Backend Registry #
Registry of backend capsules known to TorchLean's planner.
Capsules are contributed by operation or provider modules and flattened into one planner catalog. Model architectures never appear here: they lower to backend operations, and the planner chooses a capsule for each operation. Optional external modules such as LibTorch are included only when the caller enables them and chooses an assurance policy that admits them.
A named, independently maintained contribution to the backend catalog.
- name : String
- capsules : Array KernelCapsule
Instances For
Instances For
Flatten capsule modules while preserving module and local preference order.
Instances For
First repeated module name, if the registry contains two contributions with the same name.
Instances For
First repeated capsule identity after flattening, if one exists.
Instances For
Validate the identities that make registry ordering meaningful.
Different providers may implement the same operation. What is rejected is registering the same named module twice or repeating the same capsule name, operation, provider, and device tuple.
Instances For
Maintained operation/provider modules. A new architecture does not modify this list; only a new primitive implementation or provider does.
Instances For
LibTorch's independently maintained provider module. Profiles opt into it by adding this module to their catalog; provider preference is handled by the kernel policy, so module order does not encode backend selection.