There's an important distinction between the decoding speed of documents and overall throughput. Wonder what can be done to realistically generate something ap

There's an important distinction between the decoding speed of documents and overall throughput. Wonder what can be done to realistically generate something approximating a real world benchmark to cover the throughput scenario.