keras-team/keras · error · ValueError

Received an invalid value for `units`, expected a positive i

Error message

Received an invalid value for `units`, expected a positive integer. Received: units={units}

What it means

Dense requires units to be a positive Python int (strict isinstance check: bools, floats, numpy ints, and strings all fail). Keras 3 hardened this check because the unit count determines the kernel shape; invalid values previously surfaced later as cryptic shape errors.

Source

Thrown at keras/src/layers/core/dense.py:102

    def __init__(
        self,
        units,
        activation=None,
        use_bias=True,
        kernel_initializer="glorot_uniform",
        bias_initializer="zeros",
        kernel_regularizer=None,
        bias_regularizer=None,
        activity_regularizer=None,
        kernel_constraint=None,
        bias_constraint=None,
        lora_rank=None,
        lora_alpha=None,
        quantization_config=None,
        **kwargs,
    ):
        if not isinstance(units, int) or units <= 0:
            raise ValueError(
                "Received an invalid value for `units`, expected a positive "
                f"integer. Received: units={units}"
            )

        super().__init__(activity_regularizer=activity_regularizer, **kwargs)
        self.units = units
        self.activation = activations.get(activation)
        self.use_bias = use_bias
        self.kernel_initializer = initializers.get(kernel_initializer)
        self.bias_initializer = initializers.get(bias_initializer)
        self.kernel_regularizer = regularizers.get(kernel_regularizer)
        self.bias_regularizer = regularizers.get(bias_regularizer)
        self.kernel_constraint = constraints.get(kernel_constraint)
        self.bias_constraint = constraints.get(bias_constraint)
        self.lora_rank = lora_rank
        self.lora_alpha = lora_alpha if lora_alpha is not None else lora_rank
        self.lora_enabled = False
        self.quantization_config = quantization_config

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Coerce to a positive int before construction: units = int(units), ensure units > 0.
2. Fix the config source to store a plain positive integer.
3. Wrap numpy scalars with int() when passing in.

Example fix

# before
units = float(cfg['units'])  # 128.0 -> raises
layer = keras.layers.Dense(units)

# after
units = int(cfg['units'])
assert units > 0
layer = keras.layers.Dense(units)

units = int(cfg['units'])
assert isinstance(units, int) and units > 0, f'bad units: {units!r}'

def valid_units(units) -> bool:
    return isinstance(units, int) and not isinstance(units, bool) and units > 0

When it happens

Trigger: Calling keras.layers.Dense(units) with units as a float (128.0), np.int64, a string from config, or units <= 0.

Common situations: Units read from JSON/YAML config as string/float; hyperparameter sweeps yielding 0; numpy scalars from computations; Keras 2->3 migration where loose types were accepted.

Related errors

• You must build the layer before accessing `kernel`.
• adapt() expects an iterable that yields arrays or tensors wi
• Unsupported data type: {type(data)}. `adapt` supports `np.nd
• Expected an integer value for `n`, got {type(n)}.
• Expected an integer value for `size`, got {type(size)}.

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/bb585994352eb803. Report an issue: GitHub.