I have about 500k rows of 512 dimensional int16 values, which will be retrieved and use for model inputs during training, with retrieval based on index.
I see that I can either store the data using tf.constant, and look up values using tf.gather, or I can use tf.lookup.StaticHashTable to the store/retrieve the values.
Is there an advantage of doing one over the other?
I will be training my model over a TPU, if that makes a difference.