torch.TransposeShape

export type TransposeShape<S extends Shape> = number extends S['length']
  ? DynamicShape // Dynamic shape remains dynamic
  : S extends readonly []
    ? readonly [] // 0D scalar - return unchanged
    : S extends readonly [infer N extends number]
      ? readonly [N] // 1D vector - return unchanged
      : S extends readonly [infer M extends number, infer N extends number]
        ? readonly [N, M] // 2D matrix - transpose
        : transpose_error_requires_2d_tensor<S>;

Sextends Shape

Compute output shape of 2D transpose (.t()). Matches PyTorch behavior:

0D/1D: returns tensor unchanged
2D: swaps dimensions
3D+: error (transpose_error_requires_2d_tensor)