Skip to main content

Module tensor_factory

Module tensor_factory 

Source
Expand description

Candle-tied TensorFactory impl used by the registry’s stub-executor factory.

This module lived at engine/src/backends/ until the ComputeBackend trait was deleted; the directory was renamed to tensor_factory/ so the layout no longer suggests engine-level backend dispatch happens here. Real GPU dispatch goes through ferrum_kernels::backend::Backend<B> and its supertraits / capability bundles. The only remaining job in this module is to mint dummy tensors for the stub model executor (used by tests + the no-model happy path).

Re-exports§

pub use self::candle::CandleTensor;
pub use self::candle::CandleTensorFactory;
pub use self::candle::CandleTensorOps;

Modules§

candle
Candle backend - MVP implementation with core functionality