keras-team/keras · error · ValueError

All `axis` values must be in the range

Error message

All `axis` values must be in the range [-ndim, ndim). Received inputs with ndim={ndim}, while axis={axis}

What it means

Raised by the axis-validation guard in keras/src/backend/tensorflow/linalg.py norm: every axis value must lie in the half-open range [-ndim, ndim) of the input tensor. At least one axis value is out of that range (e.g. axis=3 for a 2D tensor), which would make the reduction dimension undefined, so the guard fails before computing the norm.

Solutions

1. Use axis values within [-x.ndim, x.ndim), e.g. axis=(-1, -2) for the last two dims of a matrix
2. Remember negative indices count from the last axis; prefer -1 instead of ndim-1
3. Validate axis against x.shape.rank before calling norm

When it happens

Trigger: Thrown at keras/src/backend/tensorflow/linalg.py:62 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/8d92642918bd5587. Report an issue: GitHub.

Appendix: source

Thrown at keras/src/backend/tensorflow/linalg.py:62

def lu_factor(a):
    lu, p = tf.linalg.lu(a)
    return lu, tf.math.invert_permutation(p)


def norm(x, ord=None, axis=None, keepdims=False):
    from keras.src.backend.tensorflow.numpy import moveaxis

    x = convert_to_tensor(x)
    x_shape = x.shape
    ndim = x_shape.rank

    if axis is None:
        axis = tuple(range(ndim))
    elif isinstance(axis, int):
        axis = (axis,)
    if any(a < -ndim or a >= ndim for a in axis):
        raise ValueError(
            "All `axis` values must be in the range [-ndim, ndim). "
            f"Received inputs with ndim={ndim}, while axis={axis}"
        )
    axis = axis[0] if len(axis) == 1 else axis
    num_axes = 1 if isinstance(axis, int) else len(axis)

    if standardize_dtype(x.dtype) == "int64":
        dtype = config.floatx()
    else:
        dtype = dtypes.result_type(x.dtype, float)
    x = cast(x, dtype)

    # Ref: jax.numpy.linalg.norm
    if num_axes == 1:
        if ord is None or ord == 2:
            return tf.sqrt(
                tf.reduce_sum(x * tf.math.conj(x), axis=axis, keepdims=keepdims)
            )

View on GitHub (pinned to 7a34a03db6)