Class RuntimeShape<S>

constructor

• new RuntimeShape<S extends Shape = Shape>(dims: S): RuntimeShape<S>

  Type Parameters

  • S extends Shape = Shape

  Parameters

  • dims: S

  Returns RuntimeShape<S>

Readonlydims

isMatrix

• get isMatrix(): boolean

  Check if this shape represents a matrix (2 dimensions)

  Returns boolean

isScalar

• get isScalar(): boolean

  Check if this shape represents a scalar (0 dimensions)

  Returns boolean

isVector

• get isVector(): boolean

  Check if this shape represents a vector (1 dimension)

  Returns boolean

rank

• get rank(): number

  Number of dimensions (rank) of the tensor

  Returns number

size

• get size(): number

  Total number of elements in the tensor

  Returns number

strides

• get strides(): readonly number[]

  Strides for each dimension (row-major order) Used for efficient memory access calculations

  Returns readonly number[]

broadcastWith

• broadcastWith(other: RuntimeShape): RuntimeShape

  Compute the broadcasted shape with another shape

  Parameters

  • other: RuntimeShape

  Returns RuntimeShape

canBroadcastWith

• canBroadcastWith(other: RuntimeShape): boolean

  Check if this shape is compatible with another shape for broadcasting

  Parameters

  • other: RuntimeShape

  Returns boolean

canMatMulWith

• canMatMulWith(other: RuntimeShape): boolean

  Check if this shape is compatible with another shape for matrix multiplication

  Parameters

  • other: RuntimeShape

  Returns boolean

canReshapeTo

• canReshapeTo(newShape: number[]): boolean

  Check if this shape can be reshaped to another shape

  Parameters

  • newShape: number[]

  Returns boolean

defaultStrides

• defaultStrides(): number[]

  Get default strides for this shape (row-major)

  Returns number[]

dim

• dim(index: number): number

  Get a dimension size by index, with support for negative indexing

  Parameters

  • index: number

  Returns number

equals

• equals(other: RuntimeShape): boolean

  Check if this shape is exactly equal to another

  Parameters

  • other: RuntimeShape

  Returns boolean

matMul

• matMul(other: RuntimeShape): RuntimeShape

  Compute the result shape of matrix multiplication with another shape

  Parameters

  • other: RuntimeShape

  Returns RuntimeShape

ravel

• ravel(indices: number[]): number

  Convert multi-dimensional indices to linear index

  Parameters

  • indices: number[]

  Returns number

squeeze

• squeeze(axes?: number[]): RuntimeShape

  Create a new shape with dimensions squeezed (size-1 dimensions removed)

  Parameters

  • Optionalaxes: number[]

  Returns RuntimeShape

toString

• toString(): string

  String representation for debugging

  Returns string

transpose

• transpose(axes?: number[]): RuntimeShape

  Create a new shape by transposing dimensions

  Parameters

  • Optionalaxes: number[]

  Returns RuntimeShape

unravel

• unravel(index: number): number[]

  Convert linear index to multi-dimensional indices

  Parameters

  • index: number

  Returns number[]

unsqueeze

• unsqueeze(axis: number): RuntimeShape

  Create a new shape with a size-1 dimension added at the specified axis

  Parameters

  • axis: number

  Returns RuntimeShape

StaticbroadcastShapes

• broadcastShapes(shape1: Shape, shape2: Shape): Shape

  Compute the result shape when broadcasting two shapes

  Parameters

  • shape1: Shape
  • shape2: Shape

  Returns Shape

StaticcanBroadcast

• canBroadcast(shape1: Shape, shape2: Shape): boolean

  Check if two shapes can be broadcast together

  Parameters

  • shape1: Shape
  • shape2: Shape

  Returns boolean

StaticcanMatmul

• canMatmul(shape1: Shape, shape2: Shape): boolean

  Check if two shapes can be matrix multiplied

  Matrix multiplication is valid when:

  • Both tensors have at least 1 dimension
  • The last dimension of the first tensor equals the second-to-last dimension of the second
  • All batch dimensions (if any) match exactly

  Parameters

  • shape1: Shape

    First tensor shape
  • shape2: Shape

    Second tensor shape

  Returns boolean

  true if shapes can be matrix multiplied

Staticequals

• equals(shape1: Shape, shape2: Shape): boolean

  Check if two shapes are exactly equal

  Parameters

  • shape1: Shape
  • shape2: Shape

  Returns boolean

Staticfrom

• from<S extends Shape>(dims: S): RuntimeShape<S>

  Create a RuntimeShape from array-like input

  Type Parameters

  • S extends Shape

  Parameters

  • dims: S

  Returns RuntimeShape<S>

StaticinferFromNestedArray

• inferFromNestedArray(data: unknown): Shape

  Infer shape from nested array structure

  Parameters

  • data: unknown

  Returns Shape

StaticmatmulShape

• matmulShape(shape1: Shape, shape2: Shape): null | Shape

  Compute the output shape of matrix multiplication

  Parameters

  • shape1: Shape

    First tensor shape
  • shape2: Shape

    Second tensor shape

  Returns null | Shape

  Output shape or null if shapes are incompatible

StaticpadLeft

• padLeft(shape: Shape, targetLength: number): number[]
  Internal

  Pad a shape with 1s on the left to reach target length

  Parameters

  • shape: Shape
  • targetLength: number

  Returns number[]

Staticproduct

• product(shape: Shape): number

  Compute the product of all dimensions in a shape

  Parameters

  • shape: Shape

  Returns number

Staticvalidate

• validate(dims: number[]): boolean

  Validate that a shape is well-formed

  Parameters

  • dims: number[]

  Returns boolean