Keras Embedding Layer Issue: Incorrect input length, Rank error when calling prediction

I receive the error

"Input headline is an array of rank 2, but this model only supports single vector inputs (rank 1) or a sequence of batches of vectors (rank 3)"

when I call predication.

If I change shape to [20] I get the error [core] Input feature headline was presented as a vector of length 20, but the model expects an input of length 1.

        guard let mlMultiArray = try? MLMultiArray(shape:[1,20],
                                                  dataType:MLMultiArrayDataType.double) else {
            fatalError("Unexpected runtime error. MLMultiArray")
        }

        for (index, element) in data.enumerated() {
            mlMultiArray[index] = NSNumber(floatLiteral: Double(element))
        }
        print(mlMultiArray)

        if let result = try? model.prediction(headline: mlMultiArray) {
            print("result: \(result.output[0])")
        } else {
            print("Failed!")
        }

The input type is MultiArray<Double,1> for the core ml model. The input to the modal using keras and tensor flow is in the format

array([[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,

0, 0, 0, 0, 0, 0, 387, 2630, 1]], dtype=int32).

So Keras is taking an array of ints of size 20. But core ml expects a MLArray of size one, of type double.

How do I specify type Int for the model, and why is the array count off?

Is this possible, or must I change the model?

Any tips on how to remedy this are appreciated.

EDIT: Here is the model.

def ConvolutionalNet(vocabulary_size, embedding_dimension, input_length, embedding_weights=None):

    model = Sequential()
    if embedding_weights is None:
        model.add(Embedding(vocabulary_size, embedding_dimension, input_length=input_length, trainable=False))
    else:
        model.add(Embedding(vocabulary_size, embedding_dimension, input_length=input_length, weights=[embedding_weights], trainable=False))


    model.add(Convolution1D(32, 2, W_regularizer=l2(0.005)))
    model.add(BatchNormalization())
    model.add(Activation("relu"))


    model.add(Convolution1D(32, 2, W_regularizer=l2(0.001)))
    model.add(BatchNormalization())
    model.add(Activation("relu"))


    model.add(Convolution1D(32, 2, W_regularizer=l2(0.001)))
    model.add(BatchNormalization())
    model.add(Activation("relu"))


    model.add(MaxPooling1D(17))
    model.add(Flatten())


    model.add(Dense(1, bias=True, W_regularizer=l2(0.001)))
    model.add(BatchNormalization())
    model.add(Activation("sigmoid"))


    return model

EDIT 2:

I added model.add(Reshape((20,), input_shape=(20,))) to the top of the model so the core ml model now expects MultiArray<Double,1> but I get the following error:

/mlkitc: compiler error: Embedding layer: embedding_1 must have C,H,W = 1,1,1 input blob