Prefill and Decode for Concurrent Requests - Optimizing LLM Performance