keras-team/keras · error · ValueError

Input should have its last dimension fully-defined. Received

Error message

Input should have its last dimension fully-defined. Received: input.shape = {real.shape}

What it means

The 1D FFT needs the FFT length known at symbolic-inference time, so the last dimension of the input must be fully defined (not None). Keras raises when real.shape[-1] is None during functional-model construction or with dynamic shapes.

Source

Thrown at keras/src/ops/math.py:526

        if real.shape != imag.shape:
            raise ValueError(
                "Input `x` should be a tuple of two tensors - real and "
                "imaginary. Both the real and imaginary parts should have the "
                f"same shape. Received: x[0].shape = {real.shape}, "
                f"x[1].shape = {imag.shape}"
            )

        # We are calculating 1D FFT. Hence, rank >= 1.
        if len(real.shape) < 1:
            raise ValueError(
                f"Input should have rank >= 1. "
                f"Received: input.shape = {real.shape}"
            )

        # The axis along which we are calculating FFT should be fully-defined.
        m = real.shape[-1]
        if m is None:
            raise ValueError(
                f"Input should have its last dimension fully-defined. "
                f"Received: input.shape = {real.shape}"
            )

        return (
            KerasTensor(shape=real.shape, dtype=real.dtype),
            KerasTensor(shape=imag.shape, dtype=imag.dtype),
        )

    def call(self, x):
        return backend.math.fft(x)


@keras_export("keras.ops.fft")
def fft(x):
    """Computes the Fast Fourier Transform along last axis of input.

    Args:

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Give the input a fixed last dimension: keras.Input(shape=(fft_len,))
2. Pad/ragged-to-dense to a fixed length before the FFT
3. If lengths vary, bucket inputs to a few fixed sizes or compute the FFT eagerly outside the symbolic graph
4. Rebuild with concrete tensors so the last dim is known

Example fix

# before
inputs = keras.Input(shape=(None,))  # dynamic length
out = keras.ops.fft((inputs, inputs))
# after
inputs = keras.Input(shape=(256,))  # fixed fft length
out = keras.ops.fft((inputs, inputs))

assert real.shape[-1] is not None, 'last dim must be static for fft'

When it happens

Trigger: Using a keras.Input(shape=(None,)) (dynamic sequence length) and calling keras.ops.fft inside a functional model; JAX/TF traces with an undefined last dim.

Common situations: Variable-length audio/text pipelines with dynamic timesteps feeding an FFT layer; migrating from eager execution to a functional/Layer workflow where shapes become symbolic.

Related errors

• Input should have its {axes} axes fully-defined. Received: i
• The `weights` argument should be either `None` (random initi
• RandomCrop requires the input to have a fully defined height
• All `axis` values to be kept must have a known shape. Receiv
• R2Score expects 2D inputs with shape (batch_size, output_dim

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/62d85d3769c00b2b. Report an issue: GitHub.