Gets the first max_user_contribution words from the input dataset.
batch_size: int = 1
The input dataset must yield batched 1-d tensors. This function reads each
coordinate of the tensor as an individual element and caps the total number of
elements to return. Note either none of the elements in one batch is added to
the returned result, or all the elements are added. This means the length of
the returned list of elements could be less than max_user_contribution when
dataset is capped.