HW/SW co-design of sentence-level energy optimizations for latency-aware multi-task NLP inference - View it on GitHub