Compute $A^{-1}$ for a square float tensor.

Validates that a is at least 2-D, square in the trailing two axes, and float-typed; dispatches to the backend; and wires InvBackward into the autograd graph (saving only the output) when grad mode is active.

Shape

• a: (..., N, N).
• Output: same shape as a.

See Also

solve_op : Often a better alternative — solves $AX = B$ without forming $A^{-1}$ explicitly, with better conditioning. det_op : Determinant, which uses inv_op in its backward.