keras-team/keras · error · ValueError

Cannot infer argument `num` from shape

Error message

Cannot infer argument `num` from shape {x.shape}. Either provide a tensor with a concrete shape in the `axis` dimension or explicitly pass the `num` argument.

What it means

For element-count based behavior (e.g. boolean all/any), num defaults to the size of the reduced axis. In symbolic tracing that dimension may be None, so the count cannot be inferred and you must supply num.

Solutions

Pass num explicitly
Run outside symbolic tracing or with concrete shapes
Reshape the tensor so the axis dim is concrete before the reduction

Example fix

# before
keras.ops.all(x)  # x.shape[axis] is None under tracing

# after
keras.ops.all(x, num=128)  # or pass a concrete-shaped tensor

Defensive patterns

Strategy: validation

Validate before calling

if num is None:
    num = x.shape[axis]
assert num is not None, 'pass num explicitly when axis dim is dynamic'

Prevention

Pass num explicitly for count-based reductions
Test ops inside a concrete (non-symbolic) shape context first

When it happens

Trigger: Calling an op with num=None on a KerasTensor whose shape[axis] is None inside a functional model or symbolic call

Common situations: Using count-based reductions on boolean tensors inside a symbolic (KerasTensor) trace or functional model

Related errors

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/3f3d73247c906221. Report an issue: GitHub.

Appendix: source

Thrown at keras/src/ops/core.py:764


class Unstack(Operation):
    def __init__(self, num=None, axis=0, *, name=None):
        super().__init__(name=name)
        self.num = num
        self.axis = axis

    def call(self, x):
        return backend.core.unstack(x, self.num, self.axis)

    def compute_output_spec(self, x):
        axis = canonicalize_axis(self.axis, len(x.shape))
        output_shapes = x.shape[:axis] + x.shape[axis + 1 :]
        num = self.num
        if num is None:
            num = x.shape[axis]
        if num is None:
            raise ValueError(
                "Cannot infer argument `num` from shape "
                f"{x.shape}. Either provide a tensor with a "
                "concrete shape in the `axis` dimension or "
                "explicitly pass the `num` argument."
            )
        output = [
            KerasTensor(shape=output_shapes, dtype=x.dtype) for _ in range(num)
        ]
        return output


@keras_export("keras.ops.unstack")
def unstack(x, num=None, axis=0):
    """Unpacks the given dimension of a rank-R tensor into rank-(R-1) tensors.

    Args:
        x: The input tensor.
        num: The length of the dimension axis. Automatically inferred

View on GitHub (pinned to 7a34a03db6)