Increasing GPU Utilization during Generative Inference for Higher Throughput