Package-level declarations

Data class containing detailed contiguity analysis results.

A materialization strategy that immediately copies all data from a tensor view into a new, standalone tensor with contiguous memory layout.

Default implementation of TensorViewStrategy that prefers views when possible.

Mutable holder for gradient state associated with a tensor instance.

Interface for coordinate transformation between tensor view space and parent tensor space.

Thrown when an index value is outside the allowed range.

Lightweight wrapper marking a Tensor as index-bearing. It does not change storage, only annotates intent and guarantees validation performed by asIndices.

A materialization strategy that defers copying of tensor view data until individual elements are actually accessed.

Strategy interface for materializing tensor views into concrete tensors.

Memory-pressure-aware tensor view strategy.

Specialized IndexMapper optimized for NCHW (Batch, Channel, Height, Width) tensor layout.

Detects NCHW-optimized slicing patterns for efficient memory access.

Thrown when an index tensor contains a non-integral value while strict validation is enabled.

Builder for configuring a single dimension slice within a segment { } block.

Represents a slice descriptor for tensor slicing operations.

A concrete implementation of TensorView that provides zero-copy sliced access to parent tensors.

General-purpose IndexMapper implementation for arbitrary slice patterns.

The core tensor abstraction in the SKaiNET framework, representing a fundamental architectural decision to compose tensors from data and operations.

DSL builder for creating tensor slices with fluent syntax.

A tensor view interface that represents a zero-copy slice or subset of a parent tensor.

Strategy interface for deciding between view creation and tensor copying.

Simple tensor implementation with void Ops as default for the DSL

Accumulate a gradient tensor onto this state.

Provides a more detailed contiguity analysis with diagnostic information.

Validates this tensor as an index tensor and wraps it into IndexTensor.

Materializes a list of tensor views into a list of standalone tensors efficiently.

Materializes a list of tensor views using a specific materialization strategy.

Checks if this tensor view can be materialized using the default strategy.

Checks if this tensor view can be materialized using a specific strategy.

Calculates the cosine distance between two tensors along a given dimension. Cosine distance is defined as 1 - cosine similarity. Formula: 1 - (A dot B) / (||A|| * ||B||)

Detects the NCHW slicing pattern for optimization purposes.

Estimates the total memory overhead for batch materializing a list of views.

Estimates the total memory overhead using a specific materialization strategy.

Estimates the memory overhead of materializing this tensor view.

Estimates the memory overhead of materializing this tensor view with a specific strategy.

Determines if this tensor view represents a contiguous memory layout.

Determines if a list of slices results in a contiguous memory access pattern.

Materializes this tensor view into a standalone tensor using the default strategy.

Materializes this tensor view using a specific materialization strategy.

Creates a tensor view using a direct list of slice descriptors.

Convenience method for extracting specific indices across dimensions.

Creates a new tensor by copying data from the sliced region of this tensor.

Convenience method for creating simple range slices across all dimensions.

Creates a view of an existing tensor view, enabling view chaining.

Creates a zero-copy view of this tensor using the provided slicing DSL.

Mark a tensor as requiring gradients and return it for chaining.

Clear the stored gradient.