Windows in tensorflow dataset with sequence AND constant input

Question

I have a dataset of the following form:

import tensorflow as tf

num_inputs = 3
num_outputs = 5
num_params = 5
num_samples = 32
sequence_length = 50

input_sequences = tf.random.normal((num_samples, sequence_length, num_inputs))
output_sequences = tf.random.normal((num_samples, sequence_length, num_outputs))
input_params = tf.random.normal((num_samples, num_params))

dataset = tf.data.Dataset.from_tensor_slices((
    {'sequence': input_sequences, 'parameters': input_params},
    output_sequences
    ))

And I want to train a model on this that takes two inputs: a window of say 10 consecutive steps in a sequence, and the corresponding parameters (which do not have a sequence dimension, but vary from sample to sample). It should output the 10th step of the output sequence.

So I want to create from this a new dataset of samples of the form

windows = {'sequence': (10 steps), 'parameters': (single constant vector of parameters)}, (single output step 10)

What is the clean way to do this, ideally without just copying the sequence into many overlapping windows, and for each window copying the corresponding parameters.

I know there is the window method, but I don't know how to apply it to this case. All the examples I've seen just apply it to a toy dataset that has only one type of data, not an input and output, never mind two inputs as here.