Issues with StringLookUp layer and keras.ops.matmul

[Humbulani1234]

Here is my input to the keras preprocessing StringLookup layer: <tf.Tensor 'args_4:0' shape=(None,) dtype=string> and it returns: <tf.Tensor 'string_lookup_1/Squeeze:0' shape=() dtype=int64>. The code snippet is as follows:

            index = layers.StringLookup(
                vocabulary=vocabulary, mask_token=None, num_oov_indices=1, output_mode="int"
            )
            # Convert the string input values into integer indices.
            value_index = index(feature_name)

Question 1: Isn't the output shape supposed to maintain the same shape = (None,) instead of shape = (). None is a batch dimension for tf.data.Dataset elements.

Question 2: How do I perform matmul in Keras so it runs on tensorflow, jax and torch? The following only runs on tensorflow, but fails on others. Where x=<tf.Tensor 'stack:0' shape=(None, 40, 16) dtype=float32> and v=<tf.Tensor 'ExpandDims_7:0' shape=(None, 40, 1) dtype=float32>. The output shape should be shape = (None, 16) and this are Symbolic tensors.

result = keras.ops.dot(v, keras.ops.transpose(x))
result = keras.ops.reshape(result, (-1, 16))

jax error:

TypeError: dot_general requires contracting dimensions to have the same shape, got (1,) and (40,).

torch error:

RuntimeError: The size of tensor a (83) must match the size of tensor b (16) at non-singleton dimension 0

[Humbulani1234]

Resolved