A small variant of the Funnel Transformer model that compresses sequence length for efficient encoding.