keras-team/keras · error · ValueError

Input should have rank >= 1. Received: input.shape = {x.shap

Error message

Input should have rank >= 1. Received: input.shape = {x.shape}

What it means

keras.ops.extract_sequences slices the last axis into windows of sequence_length with stride sequence_stride, so it needs input of rank >= 1. The compute_output_spec raises this when x.shape is empty, i.e. a rank-0 scalar.

Source

Thrown at keras/src/ops/math.py:442

    >>> y = keras.ops.convert_to_tensor([[1.0, 0.0], [0.0, 1.0]])
    >>> keras.ops.cdist(x, y)
    array([[1.       , 1.       ],
           [1.       , 1.4142135]], dtype=float32)
    """
    if any_symbolic_tensors((x, y)):
        return CDist().symbolic_call(x, y)
    return backend.math.cdist(x, y)


class ExtractSequences(Operation):
    def __init__(self, sequence_length, sequence_stride, *, name=None):
        super().__init__(name=name)
        self.sequence_length = sequence_length
        self.sequence_stride = sequence_stride

    def compute_output_spec(self, x):
        if len(x.shape) < 1:
            raise ValueError(
                f"Input should have rank >= 1. "
                f"Received: input.shape = {x.shape}"
            )
        if x.shape[-1] is not None:
            num_sequences = (
                1 + (x.shape[-1] - self.sequence_length) // self.sequence_stride
            )
        else:
            num_sequences = None
        new_shape = x.shape[:-1] + (num_sequences, self.sequence_length)
        return KerasTensor(shape=new_shape, dtype=x.dtype)

    def call(self, x):
        return backend.math.extract_sequences(
            x,
            sequence_length=self.sequence_length,
            sequence_stride=self.sequence_stride,
        )

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Keep at least one axis: wrap scalars with ops.reshape(x, (1,)) or ops.expand_dims; if upstream code squeezed all axes, squeeze only the axes you actually collapsed.
2. Validate rank before the call: if len(x.shape) == 0: reshape to (1,).
3. In custom layers, keep a feature axis (..., 1) rather than fully reducing to scalars before sequence ops.

Example fix

// before
from keras import ops
s = ops.array(7.0)                     # rank 0
seqs = ops.extract_sequences(s, 3, 1)  # ValueError

// after
s = ops.reshape(ops.array(7.0), (1,))   # rank 1
seqs = ops.extract_sequences(s, 3, 1)

from keras import ops

def ensure_rank1(x):
    if len(ops.shape(x)) < 1:
        x = ops.reshape(x, (1,))
    return x

seqs = ops.extract_sequences(ensure_rank1(x), seq_len, stride)

import keras

def rank1_plus(x) -> bool:
    return keras.ops.ndim(x) >= 1

When it happens

Trigger: Calling keras.ops.extract_sequences(x, sequence_length, sequence_stride) on a 0-D scalar tensor; using the op after an ops.squeeze or an all-axes reduction removed every dimension; passing a Python scalar converted via keras.ops.array without a shape.

Common situations: Transformer-encoder preprocessing where squeeze(axis=-1) on scalar outputs precedes sequence extraction; feeding per-sample scalars that should be (1,) shaped; dynamic dimension removal in a functional graph leaving scalar symbolic tensors.

Related errors

• Expected input to have rank >= 2. Received input with shape
• Inputs to `cdist` must have rank >= 2. Received shapes: x.sh
• Architecture configuration does not match {weights_name} var
• The `weights` argument should be either `None` (random initi
• Expected mode to be one of `caffe`, `tf` or `torch`. Receive

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/3e21a9445ca6b2f2. Report an issue: GitHub.