Lightweight wrapper that binds a Tensor to a specific TensorOps implementation. Useful to evaluate operator overloads with a desired backend (CPU/GPU/etc.).
The data component responsible for storage, indexing, and memory management.
The data type descriptor defining the numerical precision and value representation.
Gradient state tied to this tensor instance.
The operations component responsible for computational algorithms and transformations.
Accumulate a gradient tensor onto this tensor.
Reset accumulated gradient.