numpy/numpy · error · ValueError

ValueError: order must be one of 'C', 'F', 'A', or 'K' (got

Error message

ValueError: order must be one of 'C', 'F', 'A', or 'K' (got '{order}')

What it means

Raised by _parse_output_order when the order keyword passed to an optimized einsum contraction is not one of 'C', 'F', 'A', 'K' (compared case-insensitively after .upper()). Note the message redundantly prefixes itself with 'ValueError:'; the value of order is echoed. This parser decides the memory layout of the matmul result.

Source

Thrown at numpy/_core/einsumfunc.py:1138

        perm_ab,
        False,  # pure_multiplication=False
    )


@functools.lru_cache(maxsize=64)
def _parse_output_order(order, a_is_fcontig, b_is_fcontig):
    order = order.upper()
    if order == "K":
        return None
    elif order in "CF":
        return order
    elif order == "A":
        if a_is_fcontig and b_is_fcontig:
            return "F"
        else:
            return "C"
    else:
        raise ValueError(
            "ValueError: order must be one of "
            f"'C', 'F', 'A', or 'K' (got '{order}')"
        )


def bmm_einsum(eq, a, b, out=None, **kwargs):
    """Perform arbitrary pairwise einsums using only ``matmul``, or
    ``multiply`` if no contracted indices are involved (plus maybe single term
    ``einsum`` to prepare the terms individually). The logic for each is cached
    based on the equation and array shape, and each step is only performed if
    necessary.

    Parameters
    ----------
    eq : str
        The einsum equation.
    a : array_like
        The first array to contract.

View on GitHub (pinned to e117b3ca4e)

Solutions

1. Use one of 'C', 'F', 'A', 'K' (any case).
2. Use 'K' (default) to keep input memory order, or 'C'/'F' to force a layout.
3. Avoid trailing spaces; the value is uppercased but not stripped.

Example fix

// before
np.einsum('ij,jk->ik', a, b, optimize=True, order='Z')
// after
np.einsum('ij,jk->ik', a, b, optimize=True, order='K')  # C, F, A, or K

VALID_ORDER = {'C', 'F', 'A', 'K'}
def safe_order(order):
    if order is None:
        return 'K'
    o = str(order).upper()
    if o not in VALID_ORDER:
        raise ValueError(f"order must be one of C/F/A/K, got {order!r}")
    return o

null

null

When it happens

Trigger: np.einsum('ij,jk->ik', a, b, optimize=True, order='Z') or order='Row-major' or any string other than C/F/A/K (case-insensitive). Also triggered by a non-string order that has no matching upper().

Common situations: Passing a layout constant from another library; typo like 'c ' with trailing space (actually fine after upper/strip? no—no strip, so ' c' fails); confusing with numpy's set_printoptions or ndarray order semantics.

Related errors

• No input operands
• Character {s} is not a valid symbol.
• For this input type lists must contain either int or Ellipsi
• Subscripts can only contain one '->'.
• Invalid Ellipses.

AI-assisted analysis of numpy/numpy@e117b3ca4e (2026-08-07). Data as JSON: /api/errors/5928ed36b6b408ea. Report an issue: GitHub.