keras-team/keras · error · ValueError

The number of input channels must be evenly divisible by…

Error message

The number of input channels must be evenly divisible by kernel's in_channels. Received input channels {channels} and kernel in_channels {kernel_in_channels}. 

What it means

Pre-check in the JAX backend's convolution op: it reads the input's channel dimension (last axis for channels_last, axis 1 for channels_first) and the kernel's in_channels axis (kernel.shape[-2]) and requires the former to be a multiple of the latter, because the channels are split into feature_group_count equal groups for jax.lax.conv_general_dilated.

Solutions

1. For grouped convolution, make input channels an exact multiple of the kernel's in_channels (e.g. in_channels=64, kernel in_channels=32 gives 2 groups)
2. For a normal convolution, set the kernel's in_channels equal to the input's channel count
3. Add a 1x1 conv or channel-padding layer upstream to align the channel counts
4. Check data_format — a mismatched channels_first/channels_last setting reads the wrong axis as channels

When it happens

Trigger: Thrown at keras/src/backend/jax/ops/nn.py:757 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/41e5d1b99cbd4c1d. Report an issue: GitHub.

Appendix: source

Thrown at keras/src/backend/jax/ops/nn.py:757

    strides = _convert_to_spatial_operand(
        strides,
        num_spatial_dims,
        data_format,
        include_batch_and_channels=False,
    )
    dilation_rate = _convert_to_spatial_operand(
        dilation_rate,
        num_spatial_dims,
        data_format,
        include_batch_and_channels=False,
    )
    if data_format == "channels_last":
        channels = inputs.shape[-1]
    else:
        channels = inputs.shape[1]
    kernel_in_channels = kernel.shape[-2]
    if channels % kernel_in_channels > 0:
        raise ValueError(
            "The number of input channels must be evenly divisible by "
            f"kernel's in_channels. Received input channels {channels} and "
            f"kernel in_channels {kernel_in_channels}. "
        )
    feature_group_count = channels // kernel_in_channels
    kernel = convert_to_tensor(kernel)
    inputs = convert_to_tensor(inputs, dtype=kernel.dtype)
    result = jax.lax.conv_general_dilated(
        inputs,
        kernel,
        strides,
        padding,
        rhs_dilation=dilation_rate,
        dimension_numbers=dimension_numbers,
        feature_group_count=feature_group_count,
    )
    if result.size == 0 and inputs.size != 0:
        raise ValueError(

View on GitHub (pinned to 7a34a03db6)