Does anyone know the impact of the prompt size in terms of throughput? If I'm only generating 10 tokens, does it matter if my initial prompt is 10 tokens or 8000 tokens? How much does it matter?