Tensorflow 2.3: tf.feature_column for variable length inputs or multi-hot encoding

Question

I have been trying to multi-hot encode some of my features with tf.feature_column. From my understanding, given a feature, all inputs must be padded to the same length and l