keras-team/keras · error · ValueError

Shapes are incompatible for associative_scan interleaving…

Error message

Shapes are incompatible for associative_scan interleaving. a.shape[{axis}]={a.shape[axis]}, b.shape[{axis}]={b.shape[axis]}

What it means

Internal shape check inside the NumPy `associative_scan`'s `_interleave` helper. While combining partial scan results, it interleaves two tensors a and b along the scan axis; for odd lengths the two operands may differ by exactly one element along that axis, otherwise the interleave is geometrically impossible. If a.shape[axis] and b.shape[axis] differ by more than one (or disagree when they must match), this error reports both sizes and the axis.

Solutions

1. Ensure scan inputs have lengths that are compatible with the halving scheme (differences of at most one element per level) — typically satisfied by any valid associative_scan input, so treat this as a sign of malformed custom elems
2. Check that the combine function `f` returns tensors with the same shapes as its inputs, preserving the scan-axis length
3. Report the issue upstream if it occurs with well-formed inputs, as it indicates an internal inconsistency in the divide-and-conquer path

When it happens

Trigger: Thrown at keras/src/backend/numpy/ops/core.py:240 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/4078d94c82a73312. Report an issue: GitHub.

Appendix: source

Thrown at keras/src/backend/numpy/ops/core.py:240

        c = f(a, b)
        c_flat = tree.flatten(c)
        return c_flat

    num_elems = int(elems_flat[0].shape[axis])
    if not all(int(elem.shape[axis]) == num_elems for elem in elems_flat[1:]):
        raise ValueError(
            "Array inputs to associative_scan must have the same "
            "first dimension. (saw: {})".format(
                [elem.shape for elem in elems_flat]
            )
        )

    def _interleave(a, b, axis):
        """Given two Tensors of static shape, interleave them along axis."""
        if not (
            a.shape[axis] == b.shape[axis] or a.shape[axis] == b.shape[axis] + 1
        ):
            raise ValueError(
                "Shapes are incompatible for associative_scan interleaving. "
                f"a.shape[{axis}]={a.shape[axis]}, "
                f"b.shape[{axis}]={b.shape[axis]}"
            )

        # we want to get a: [a1, a2], b: [b1, b2]
        # to a: [a1, 0, a2, 0], b: [0, b1, 0, b2]
        a_shape = list(a.shape)
        a_shape[axis] = a.shape[axis] * 2 - 1

        b_shape = list(b.shape)
        b_shape[axis] = b.shape[axis] * 2 - 1

        a_dil = np.zeros(a_shape)
        np.copyto(slice_along_axis(a_dil, 0, None, 2, axis), a)
        b_dil = np.zeros(b_shape)
        np.copyto(slice_along_axis(b_dil, 0, None, 2, axis), b)

View on GitHub (pinned to 7a34a03db6)