torch.SVDShape

export type SVDShape<S extends Shape> = S['length'] extends 2
  ? S // Returns multiple tensors, simplify to input shape check
  : S extends readonly [...infer _Batch extends number[], number, number]
    ? S // Valid batched case
    : number extends S['length']
      ? DynamicShape
      : linalg_error_requires_at_least_2d<S['length']>;

Sextends Shape

Output shape of SVD. For [m, n] matrix: U is [m, m], S is [min(m,n)], V is [n, n] (full_matrices=true) For [m, n] matrix: U is [m, k], S is [k], V is [k, n] (full_matrices=false, k=min(m,n))