input shape problem with RaggedTensor input in RNN

Question

I have created ragged tensors from python lists as below.

>  list_a=tf.ragged.constant(list_a)

>  list_b=tf.ragged.constant(list_b)

Checking the shapes gives me:

 list_a.shape

TensorShape([10555, None, None])

list_b.shape

TensorShape([10555])

dataset = tf.data.Dataset.from_tensor_slices((list_a, list_b))
dataset

TensorSliceDataset shapes: ((None, None), ()), types: (tf.float64, tf.float32)>

train_data=dataset.take(9000).batch(5)
train_data

BatchDataset shapes: ((None, None, None), (None,)), types: (tf.float64, tf.float32)>

My problem is I cannot figure out how to specify the input_shape in a SimpleRNN layer with this type of dataset. Thank you,